Merge pull request #5908 from SivanCola/feature/bot-channel-settings-ui

Simplify bot channel settings / 简化机器人渠道设置

.github/workflows/release-desktop.yml

@@ -8,7 +8,8 @@ name: Release desktop
 # Wails cannot cross-compile a CGO/WebKit binary, so build/ fans out to one native
 # runner per platform. Artifacts are minisign-signed (MINISIGN_* secrets), a
 # latest.json manifest is generated, and everything is published to a GitHub
-# release and mirrored to R2 (the updater reads R2 first, GitHub as fallback).
+# release and mirrored to R2 (the updater reads R2 first, then the crash worker
+# release gateway; GitHub's repository-wide "latest" is reserved for the CLI line).
 #
 # The same build/sign/manifest pipeline serves two channels. A `desktop-v*` tag
 # (or manual stable dispatch) publishes a GitHub release and moves R2's latest/
@@ -260,9 +261,9 @@ jobs:
         working-directory: desktop
         run: go run ./cmd/sign manifest ../dist "${{ steps.ver.outputs.version }}" "${{ steps.ver.outputs.tag }}"
 
-      # Canary is R2-only and never appears on the GitHub releases page; the mirror
-      # job picks up the signed dist via the canary-dist artifact below. Stable
-      # publishes a GitHub release as usual.
+      # Canary never appears on the GitHub releases page; the mirror job picks up
+      # the signed dist via the canary-dist artifact below. Stable publishes a
+      # GitHub release as usual.
       - name: Publish GitHub release
         if: steps.ver.outputs.channel != 'canary'
         env:
@@ -285,7 +286,7 @@ jobs:
     needs: publish
     runs-on: ubuntu-latest
     permissions:
-      contents: read # gh release download
+      contents: write # gh release download + compatibility manifest upload
       actions: write # dispatch pages.yml to re-bake the site version
     # Skip cleanly when R2 isn't configured; GitHub release still works as fallback.
     if: ${{ github.repository_owner == 'esengine' }}
@@ -366,6 +367,53 @@ jobs:
             aws s3 cp assets/ "s3://${R2_BUCKET}/latest/" --recursive --endpoint-url "$ENDPOINT"
           fi
 
+      - name: Smoke desktop release pointers
+        if: env.HAS_R2 == 'true'
+        env:
+          TAG: ${{ steps.ver.outputs.tag }}
+          VERSION: ${{ steps.ver.outputs.version }}
+          CHANNEL: ${{ steps.ver.outputs.channel }}
+          PRERELEASE: ${{ steps.ver.outputs.prerelease }}
+        run: |
+          set -euo pipefail
+          f=assets/latest.json
+          jq -e --arg version "$VERSION" --arg prefix "https://dl.reasonix.io/${TAG}/" '
+            .version == $version and
+            .download_page == "https://reasonix.io/#start" and
+            ([.platforms[] | (.url, .sig)] |
+              all(type == "string" and startswith($prefix) and (contains("/releases/latest/") | not)))
+          ' "$f" >/dev/null
+
+          curl --retry 5 --retry-delay 2 -fsSL "https://dl.reasonix.io/${TAG}/latest.json" >/tmp/reasonix-desktop-tag-latest.json
+          if [ "$CHANNEL" = "canary" ]; then
+            pointer="canary"
+          elif [ "$PRERELEASE" = "true" ]; then
+            pointer=""
+          else
+            pointer="latest"
+          fi
+          if [ -n "$pointer" ]; then
+            curl --retry 5 --retry-delay 2 -fsSL "https://dl.reasonix.io/${pointer}/latest.json" >/tmp/reasonix-desktop-pointer-latest.json
+            jq -e --arg version "$VERSION" '.version == $version' /tmp/reasonix-desktop-pointer-latest.json >/dev/null
+          fi
+
+          jq -r '.platforms[] | .url, .sig' "$f" | while IFS= read -r asset; do
+            curl --retry 5 --retry-delay 2 -fsSI "$asset" >/dev/null
+          done
+
+      - name: Attach desktop manifest to matching CLI release
+        if: env.HAS_R2 == 'true' && steps.ver.outputs.channel != 'canary' && steps.ver.outputs.prerelease != 'true'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          VERSION: ${{ steps.ver.outputs.version }}
+        run: |
+          set -euo pipefail
+          if gh release view "$VERSION" >/dev/null 2>&1; then
+            gh release upload "$VERSION" assets/latest.json --clobber
+          else
+            echo "CLI release $VERSION does not exist yet; release.yml will attach the compatibility latest.json when it publishes."
+          fi
+
       # Stable release moved R2 latest/ — rebuild the site so its build-time baked
       # version + JSON-LD follow (site.js's runtime .rxv refresh can't touch first paint / SEO).
       - name: Refresh site to the new version

.github/workflows/release.yml

@@ -48,3 +48,23 @@ jobs:
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           HOMEBREW_TAP_TOKEN: ${{ secrets.HOMEBREW_TAP_TOKEN }}
+
+      - name: Attach desktop manifest compatibility asset
+        env:
+          GH_TOKEN: ${{ github.token }}
+          TAG: ${{ github.ref_name }}
+        run: |
+          set -euo pipefail
+          case "$TAG" in
+            *-*)
+              echo "prerelease $TAG — GitHub latest does not move here; skipping desktop manifest compatibility asset"
+              exit 0
+              ;;
+          esac
+          curl --retry 5 --retry-delay 2 -fsSL https://dl.reasonix.io/latest/latest.json -o latest.raw.json
+          jq '.download_page = "https://reasonix.io/#start"' latest.raw.json > latest.json
+          jq -e '
+            ([.platforms[] | (.url, .sig)] |
+              all(type == "string" and startswith("https://dl.reasonix.io/") and (contains("/releases/latest/") | not)))
+          ' latest.json >/dev/null
+          gh release upload "$TAG" latest.json --clobber

cmd/e2ebench/diff.go

@@ -10,6 +10,8 @@ import (
 	"strings"
 	"time"
 	"unicode/utf8"
+
+	"reasonix/internal/shellparse"
 )
 
 type diffOpts struct {
@@ -546,57 +548,25 @@ func failingTestNames(out string) []string {
 }
 
 func runTests(repo, testCmd string, pkgs []string) (bool, string) {
-	fields := splitShellFields(testCmd)
+	test, err := shellparse.ParseStaticCommand(testCmd, shellparse.StaticCommandPolicy{AllowEnvAssignments: true, AllowStderrToStdout: true})
+	if err != nil {
+		return false, "invalid test command: " + err.Error()
+	}
+	fields := test.Argv
 	if len(fields) == 0 {
 		fields = []string{"go", "test"}
 	}
 	args := append(fields[1:], pkgs...)
 	cmd := exec.Command(fields[0], args...)
 	cmd.Dir = repo
+	if len(test.Env) > 0 {
+		cmd.Env = append(os.Environ(), test.Env...)
+	}
 	cmd.WaitDelay = 5 * time.Minute // bound the wait if `go test` hangs
 	out, err := cmd.CombinedOutput()
 	return err == nil, string(out)
 }
 
-// splitShellFields splits on whitespace, honoring single/double-quoted spans and stripping the quotes.
-func splitShellFields(s string) []string {
-	var out []string
-	var cur strings.Builder
-	inSingle, inDouble := false, false
-	for i := 0; i < len(s); i++ {
-		c := s[i]
-		switch {
-		case inSingle:
-			if c == '\'' {
-				inSingle = false
-			} else {
-				cur.WriteByte(c)
-			}
-		case inDouble:
-			if c == '"' {
-				inDouble = false
-			} else {
-				cur.WriteByte(c)
-			}
-		case c == '\'':
-			inSingle = true
-		case c == '"':
-			inDouble = true
-		case c == ' ' || c == '\t' || c == '\n':
-			if cur.Len() > 0 {
-				out = append(out, cur.String())
-				cur.Reset()
-			}
-		default:
-			cur.WriteByte(c)
-		}
-	}
-	if cur.Len() > 0 {
-		out = append(out, cur.String())
-	}
-	return out
-}
-
 // changedGoFiles lists .go files changed by base...HEAD, excluding *_test.go
 // when includeTests is false (we want the source under test).
 func changedGoFiles(repo, base string, includeTests bool) []string {

cmd/e2ebench/diff_test.go

@@ -0,0 +1,48 @@
+package main
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func TestRunTestsAllowsStaticEnvAssignment(t *testing.T) {
+	repo := t.TempDir()
+	writeE2EFile(t, filepath.Join(repo, "go.mod"), "module example.com/e2ebenchtest\n\ngo 1.23\n")
+	writeE2EFile(t, filepath.Join(repo, "env_test.go"), `package e2ebenchtest
+
+import (
+	"os"
+	"testing"
+)
+
+func TestEnv(t *testing.T) {
+	if got := os.Getenv("REASONIX_E2E_ENV"); got != "ok" {
+		t.Fatalf("REASONIX_E2E_ENV = %q", got)
+	}
+}
+`)
+
+	ok, out := runTests(repo, "GOWORK=off REASONIX_E2E_ENV=ok go test", []string{"./..."})
+	if !ok {
+		t.Fatalf("runTests failed:\n%s", out)
+	}
+}
+
+func TestRunTestsRejectsDynamicEnvAssignment(t *testing.T) {
+	ok, out := runTests(t.TempDir(), "REASONIX_E2E_ENV=$(echo ok) go test", []string{"./..."})
+	if ok {
+		t.Fatal("runTests accepted dynamic env assignment")
+	}
+	if !strings.Contains(out, "invalid test command: shell expansion") {
+		t.Fatalf("output = %q, want shell expansion rejection", out)
+	}
+}
+
+func writeE2EFile(t *testing.T, path, content string) {
+	t.Helper()
+	if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+		t.Fatalf("write %s: %v", path, err)
+	}
+}

desktop/README.md

@@ -101,7 +101,9 @@ Desktop releases ride their own tag namespace, `desktop-v<semver>` (plain `v*`
 tags are the CLI release). Pushing one triggers `.github/workflows/release-desktop.yml`,
 which builds on a native runner per platform (Wails can't cross-compile a
 CGO/WebKit binary), packages each artifact, signs it with minisign, generates a
-`latest.json` manifest, publishes a GitHub release, and mirrors everything to R2.
+`latest.json` manifest, publishes a GitHub release, mirrors everything to R2,
+and attaches the current desktop manifest to the matching CLI release for old
+clients that still ask GitHub's repository-wide `latest` release for it.
 The Linux artifact links against WebKitGTK 4.1 (`-tags webkit2_41`), so it needs
 `libwebkit2gtk-4.1-0` at runtime — present by default on Ubuntu 22.04+, Fedora 40+.
 
@@ -109,9 +111,12 @@ The Linux artifact links against WebKitGTK 4.1 (`-tags webkit2_41`), so it needs
 git tag desktop-v1.1.0 && git push origin desktop-v1.1.0
 ```
 
-The app checks `latest.json` on startup (R2 first, GitHub as fallback) and shows
-an update banner when a newer version is published; **Settings → Software update**
-has a manual check. Self-update behavior by platform:
+The app checks `latest.json` on startup (R2 first, then the
+`crash.reasonix.io` desktop release gateway) and shows an update banner when a
+newer version is published; **Settings → Software update** has a manual check.
+The gateway resolves only the desktop `desktop-v*` release line and never uses
+GitHub's repository-wide `/releases/latest` shortcut, because plain `v*` tags are
+the CLI release line. Self-update behavior by platform:
 
 - **Linux / Windows** — download, verify the minisign signature, then update in
   place: Linux replaces the binary and relaunches; Windows runs the per-user NSIS

desktop/app.go

@@ -43,7 +43,9 @@ import (
 	"reasonix/internal/i18n"
 	"reasonix/internal/mcpdiag"
 	"reasonix/internal/memory"
+	"reasonix/internal/notify"
 	"reasonix/internal/plugin"
+	"reasonix/internal/pluginpkg"
 	"reasonix/internal/provider"
 	"reasonix/internal/skill"
 )
@@ -141,6 +143,9 @@ type App struct {
 
 	metrics atomic.Pointer[metricsAggregator] // non-nil only when desktop.metrics is opted in; swapped live by SetDesktopMetrics
 
+	notificationSenderOnce sync.Once
+	notificationSender     notify.Sender
+
 	runtimeEvents asyncRuntimeEmitter
 
 	// promptHistoryTape is a lazy, cursor-addressed view of prompt history. It
@@ -663,6 +668,7 @@ func (a *App) shutdown(context.Context) {
 				slog.Warn("desktop: shutdown snapshot failed", "tab", t.ID, "err", err)
 			}
 			t.Ctrl.Close()
+			t.releaseSessionLease()
 		}
 	}
 }
@@ -1430,6 +1436,9 @@ func (a *App) ClearSession() error {
 	if err := ctrl.ClearSession(); err != nil {
 		return err
 	}
+	if err := tab.ensureSessionLease(ctrl.SessionPath()); err != nil {
+		return err
+	}
 	tab.resetTelemetry()
 	a.persistTabSessionPath(tab, ctrl.SessionPath())
 	a.invalidatePromptHistoryCache()
@@ -1474,6 +1483,8 @@ func (a *App) clearActiveSessionRuntime(tab *WorkspaceTab, oldCtrl control.Sessi
 		TokenMode:                currentTabTokenMode(tab),
 		SharedHost:               sharedHost,
 		CleanupPendingReconciler: reconcileDesktopCleanupPending,
+		SessionRecoveryMeta:      a.tabSessionRecoveryMeta(tab),
+		OnSessionRecovered:       a.handleTabSessionRecovered(tab),
 	})
 	if err != nil {
 		if teardownTimedOut {
@@ -1505,6 +1516,10 @@ func (a *App) clearActiveSessionRuntime(tab *WorkspaceTab, oldCtrl control.Sessi
 	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
 	newCtrl.SetGoal(tab.goal)
 	path := agent.NewSessionPath(newCtrl.SessionDir(), newCtrl.Label())
+	if err := tab.ensureSessionLease(path); err != nil {
+		newCtrl.Close()
+		return err
+	}
 	newCtrl.SetSessionPath(path)
 
 	a.mu.Lock()
@@ -1901,7 +1916,11 @@ func (a *App) ListSessions() []SessionMeta {
 	out := make([]SessionMeta, 0, len(infos))
 	for _, s := range infos {
 		_, isOpen := open[s.Path]
-		meta := sessionMetaFromInfo(s, titles[filepath.Base(s.Path)], s.Path == active, isOpen, 0)
+		title := strings.TrimSpace(s.CustomTitle)
+		if title == "" {
+			title = titles[filepath.Base(s.Path)]
+		}
+		meta := sessionMetaFromInfo(s, title, s.Path == active, isOpen, 0)
 		if route, ok := channelRoutes[sessionRuntimeKey(s.Path)]; ok {
 			applyChannelSessionRoute(&meta, route)
 		}
@@ -1926,7 +1945,11 @@ func (a *App) ListTrashedSessions() []SessionMeta {
 				continue
 			}
 			deletedAt := trashedSessionDeletedAt(path)
-			out = append(out, sessionMetaFromInfo(infos[0], titles[filepath.Base(path)], false, false, deletedAt))
+			title := strings.TrimSpace(infos[0].CustomTitle)
+			if title == "" {
+				title = titles[filepath.Base(path)]
+			}
+			out = append(out, sessionMetaFromInfo(infos[0], title, false, false, deletedAt))
 		}
 	}
 	sort.Slice(out, func(i, j int) bool {
@@ -2293,7 +2316,11 @@ func (a *App) prepareRemovedSessionRuntimes(removed []removedSessionRuntime) err
 			continue
 		}
 		if err := item.ctrl.Snapshot(); err != nil {
-			return err
+			if !errors.Is(err, agent.ErrSessionSnapshotConflict) {
+				return err
+			}
+			slog.Warn("desktop: skipping stale runtime snapshot before removing session",
+				"session", item.sessionPath, "err", err)
 		}
 		item.ctrl.SetSessionPath("")
 		a.quiesceTabAutosave(item.tab)
@@ -2402,6 +2429,7 @@ func (a *App) closeRemovedSessionRuntime(item removedSessionRuntime, closed map[
 				releasedTabs[item.tab] = true
 			}
 			a.releaseTabSharedHost(item.tab)
+			item.tab.releaseSessionLease()
 		}
 	}
 	if item.ctrl == nil {
@@ -2607,9 +2635,19 @@ func (a *App) PurgeTrashedSession(path string) error {
 }
 
 // RenameSession sets a custom display name for a session (empty clears it back to
-// the preview). It only affects the history panel; the file on disk is unchanged.
+// the preview). The transcript file is unchanged; the canonical name lives in
+// the branch meta sidecar, with the legacy .titles.json map kept as a
+// compatibility write-through for older desktop data paths.
 func (a *App) RenameSession(path, title string) error {
-	if err := setSessionTitle(a.activeSessionDir(), path, title); err != nil {
+	dir := a.activeSessionDir()
+	sessionPath, _, err := validateSessionPath(dir, path)
+	if err != nil {
+		return err
+	}
+	if err := agent.RenameSession(sessionPath, title); err != nil {
+		return err
+	}
+	if err := setSessionTitle(dir, sessionPath, title); err != nil {
 		return err
 	}
 	a.invalidatePromptHistoryCache()
@@ -2806,6 +2844,7 @@ func (a *App) rebindTabToLoadedSessionPath(tab *WorkspaceTab, sessionPath string
 		}
 	} else {
 		ctrl.Close()
+		tab.releaseSessionLease()
 	}
 
 	a.mu.Lock()
@@ -4996,6 +5035,7 @@ func (a *App) Commands() []CommandInfo {
 		{Name: "remember", Description: i18n.M.CmdRemember, Kind: "builtin"},
 		{Name: "mcp", Description: i18n.M.CmdMcp, Kind: "builtin"},
 		{Name: "hooks", Description: i18n.M.CmdHooks, Kind: "builtin"},
+		{Name: "plugins", Description: i18n.M.CmdPlugins, Kind: "builtin"},
 		{Name: "theme", Description: i18n.M.CmdTheme, Kind: "builtin"},
 		{Name: "skill", Description: i18n.M.CmdSkill, Kind: "builtin"},
 		{Name: "reload-cmd", Description: i18n.M.CmdReloadCmd, Kind: "builtin"},
@@ -5062,6 +5102,9 @@ func (a *App) SlashArgs(input string) SlashArgsResult {
 		DisconnectedMCP: ctrl.DisconnectedMCPNames(),
 		CurrentModel:    model,
 	}
+	if names, err := pluginpkg.InstalledNames(config.ReasonixHomeDir()); err == nil {
+		data.PluginNames = names
+	}
 	seen := map[string]bool{}
 	for _, m := range a.Models() {
 		data.ModelRefs = append(data.ModelRefs, m.Ref)
@@ -5092,6 +5135,7 @@ type CapabilitiesView struct {
 	Servers    []ServerView    `json:"servers"`
 	Skills     []SkillView     `json:"skills"`
 	SkillRoots []SkillRootView `json:"skillRoots"`
+	Plugins    []PluginView    `json:"plugins"`
 }
 
 // SkillsSettingsView is the skills management page's data, split from MCP
@@ -5174,6 +5218,7 @@ func (a *App) Capabilities() CapabilitiesView {
 		Servers:    a.MCPServers(),
 		Skills:     skills.Skills,
 		SkillRoots: skills.SkillRoots,
+		Plugins:    a.Plugins(),
 	}
 }
 
@@ -5605,6 +5650,27 @@ func (a *App) PickSkillFolder() (string, error) {
 	return normalizeSkillPath(dir), nil
 }
 
+// PickPluginFolder opens a directory picker for choosing a local plugin package
+// source. It returns the selected directory path; plugin install/plan performs
+// manifest validation and decides whether to copy or link the package.
+func (a *App) PickPluginFolder() (string, error) {
+	if a.ctx == nil {
+		return "", nil
+	}
+	cur := a.activeWorkspaceRoot()
+	if strings.TrimSpace(cur) == "" {
+		cur, _ = os.Getwd()
+	}
+	dir, err := runtime.OpenDirectoryDialog(a.ctx, runtime.OpenDialogOptions{
+		Title:            "Choose plugin folder",
+		DefaultDirectory: dialogDefaultDirectory(cur),
+	})
+	if err != nil || dir == "" {
+		return "", err
+	}
+	return filepath.Clean(dir), nil
+}
+
 // AddSkillPath adds a custom skill root to the user config and rebuilds the
 // controller so the skills index and slash menu reflect it immediately.
 func (a *App) AddSkillPath(path string) error {
@@ -6563,11 +6629,27 @@ func (a *App) SetModelForTab(tabID, name string) error {
 		TokenMode:                currentTabTokenMode(tab),
 		SharedHost:               sharedHost,
 		CleanupPendingReconciler: reconcileDesktopCleanupPending,
+		SessionRecoveryMeta:      a.tabSessionRecoveryMeta(tab),
+		OnSessionRecovered:       a.handleTabSessionRecovered(tab),
 	})
 	if err != nil {
+		tab.releaseSessionLease()
 		return err
 	}
 	a.bindControllerDisplayRecorder(newCtrl)
+	newCtrl.EnableInteractiveApproval()
+	applyTabModeToController(newCtrl, tab.mode)
+	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
+	newCtrl.SetGoal(tab.goal)
+
+	path := agent.ContinueSessionPath(prevPath, newCtrl.SessionDir(), newCtrl.Label())
+	if err := tab.ensureSessionLease(path); err != nil {
+		newCtrl.Close()
+		tab.releaseSessionLease()
+		return err
+	}
+	resumeWithFreshSystemPrompt(newCtrl, carried, path)
+	a.persistTabSessionPath(tab, path)
 	a.mu.Lock()
 	tab.Ctrl = newCtrl
 	tab.model = name
@@ -6575,14 +6657,6 @@ func (a *App) SetModelForTab(tabID, name string) error {
 	tab.Label = newCtrl.Label()
 	a.saveTabsLocked()
 	a.mu.Unlock()
-	newCtrl.EnableInteractiveApproval()
-	applyTabModeToController(newCtrl, tab.mode)
-	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
-	newCtrl.SetGoal(tab.goal)
-
-	path := agent.ContinueSessionPath(prevPath, newCtrl.SessionDir(), newCtrl.Label())
-	resumeWithFreshSystemPrompt(newCtrl, carried, path)
-	a.persistTabSessionPath(tab, path)
 	return nil
 }
 
@@ -6665,11 +6739,26 @@ func (a *App) SetEffortForTab(tabID, level string) error {
 		TokenMode:                currentTabTokenMode(tab),
 		SharedHost:               sharedHost,
 		CleanupPendingReconciler: reconcileDesktopCleanupPending,
+		SessionRecoveryMeta:      a.tabSessionRecoveryMeta(tab),
+		OnSessionRecovered:       a.handleTabSessionRecovered(tab),
 	})
 	if err != nil {
+		tab.releaseSessionLease()
 		return err
 	}
 	a.bindControllerDisplayRecorder(newCtrl)
+	newCtrl.EnableInteractiveApproval()
+	applyTabModeToController(newCtrl, tab.mode)
+	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
+	newCtrl.SetGoal(tab.goal)
+	path := agent.ContinueSessionPath(prevPath, newCtrl.SessionDir(), newCtrl.Label())
+	if err := tab.ensureSessionLease(path); err != nil {
+		newCtrl.Close()
+		tab.releaseSessionLease()
+		return err
+	}
+	resumeWithFreshSystemPrompt(newCtrl, carried, path)
+	a.persistTabSessionPath(tab, path)
 	a.mu.Lock()
 	tab.Ctrl = newCtrl
 	tab.model = modelRef
@@ -6679,13 +6768,6 @@ func (a *App) SetEffortForTab(tabID, level string) error {
 	tab.Ready = true
 	a.saveTabsLocked()
 	a.mu.Unlock()
-	newCtrl.EnableInteractiveApproval()
-	applyTabModeToController(newCtrl, tab.mode)
-	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
-	newCtrl.SetGoal(tab.goal)
-	path := agent.ContinueSessionPath(prevPath, newCtrl.SessionDir(), newCtrl.Label())
-	resumeWithFreshSystemPrompt(newCtrl, carried, path)
-	a.persistTabSessionPath(tab, path)
 	return nil
 }
 
@@ -6743,14 +6825,27 @@ func (a *App) SetTokenModeForTab(tabID, mode string) error {
 		TokenMode:                mode,
 		SharedHost:               sharedHost,
 		CleanupPendingReconciler: reconcileDesktopCleanupPending,
+		SessionRecoveryMeta:      a.tabSessionRecoveryMeta(tab),
+		OnSessionRecovered:       a.handleTabSessionRecovered(tab),
 	})
 	if err != nil {
 		return err
 	}
 	a.bindControllerDisplayRecorder(newCtrl)
+	newCtrl.EnableInteractiveApproval()
+	applyTabModeToController(newCtrl, tab.mode)
+	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
+	newCtrl.SetGoal(tab.goal)
+	path := agent.ContinueSessionPath(prevPath, newCtrl.SessionDir(), newCtrl.Label())
+	if err := tab.ensureSessionLease(path); err != nil {
+		newCtrl.Close()
+		return err
+	}
+	resumeWithFreshSystemPrompt(newCtrl, carried, path)
 	if oldCtrl != nil {
 		oldCtrl.Close()
 	}
+	a.persistTabSessionPath(tab, path)
 	a.mu.Lock()
 	tab.Ctrl = newCtrl
 	tab.model = modelRef
@@ -6760,13 +6855,6 @@ func (a *App) SetTokenModeForTab(tabID, mode string) error {
 	tab.Ready = true
 	a.saveTabsLocked()
 	a.mu.Unlock()
-	newCtrl.EnableInteractiveApproval()
-	applyTabModeToController(newCtrl, tab.mode)
-	applyTabToolApprovalModeToController(newCtrl, tab.toolApprovalMode)
-	newCtrl.SetGoal(tab.goal)
-	path := agent.ContinueSessionPath(prevPath, newCtrl.SessionDir(), newCtrl.Label())
-	resumeWithFreshSystemPrompt(newCtrl, carried, path)
-	a.persistTabSessionPath(tab, path)
 	return nil
 }
 

desktop/app_autosave_test.go

@@ -2,6 +2,7 @@ package main
 
 import (
 	"context"
+	"errors"
 	"os"
 	"path/filepath"
 	"strings"
@@ -173,6 +174,229 @@ func TestAutosaveFailureRetriesAndRecoversOnNextTurnDone(t *testing.T) {
 	}
 }
 
+func TestDesktopSnapshotConflictRecoveryUpdatesTabAndProjectTree(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	root := globalTabWorkspaceRoot()
+	dir := desktopSessionDir(root)
+	if err := os.MkdirAll(dir, 0o755); err != nil {
+		t.Fatalf("mkdir sessions: %v", err)
+	}
+	originalPath := filepath.Join(dir, "session.jsonl")
+	originalTopic := "topic_original"
+	if err := setTopicTitle("", originalTopic, "Original"); err != nil {
+		t.Fatalf("set original topic title: %v", err)
+	}
+	current := agent.NewSession("sys")
+	current.Add(provider.Message{Role: provider.RoleUser, Content: "first"})
+	current.Add(provider.Message{Role: provider.RoleAssistant, Content: "one"})
+	current.Add(provider.Message{Role: provider.RoleUser, Content: "disk second"})
+	if err := current.Save(originalPath); err != nil {
+		t.Fatalf("Save current: %v", err)
+	}
+	if err := agent.SaveBranchMeta(originalPath, agent.BranchMeta{
+		Scope:         "global",
+		TopicID:       originalTopic,
+		TopicTitle:    "Original",
+		Preview:       "first",
+		Turns:         2,
+		SchemaVersion: agent.BranchMetaCountsVersion,
+	}); err != nil {
+		t.Fatalf("SaveBranchMeta original: %v", err)
+	}
+
+	staleSess := agent.NewSession("sys")
+	staleSess.Add(provider.Message{Role: provider.RoleUser, Content: "first"})
+	staleSess.Add(provider.Message{Role: provider.RoleAssistant, Content: "one"})
+	staleSess.Add(provider.Message{Role: provider.RoleUser, Content: "local second"})
+	staleExec := agent.New(stubProvider{}, tool.NewRegistry(), staleSess, agent.Options{}, event.Discard)
+	app := &App{
+		tabs:             map[string]*WorkspaceTab{},
+		detachedSessions: map[string]*WorkspaceTab{},
+		activeTabID:      "recovery_tab",
+	}
+	tab := &WorkspaceTab{
+		ID:            "recovery_tab",
+		Scope:         "global",
+		WorkspaceRoot: root,
+		TopicID:       originalTopic,
+		TopicTitle:    "Original",
+		SessionPath:   originalPath,
+		Ready:         true,
+		model:         "test-model",
+		disabledMCP:   map[string]ServerView{},
+	}
+	tab.sink = &tabEventSink{tabID: tab.ID, app: app}
+	tab.Ctrl = control.New(control.Options{
+		Executor:            staleExec,
+		SessionDir:          dir,
+		SessionPath:         originalPath,
+		Label:               "test",
+		Sink:                tab.sink,
+		SessionRecoveryMeta: app.tabSessionRecoveryMeta(tab),
+		OnSessionRecovered:  app.handleTabSessionRecovered(tab),
+	})
+	app.tabs[tab.ID] = tab
+
+	if err := tab.Ctrl.Snapshot(); err != nil {
+		t.Fatalf("Snapshot: %v", err)
+	}
+	recoveryPath := tab.Ctrl.SessionPath()
+	if recoveryPath == "" || recoveryPath == originalPath {
+		t.Fatalf("recovery path = %q, want distinct path", recoveryPath)
+	}
+	if tab.SessionPath != recoveryPath {
+		t.Fatalf("tab session path = %q, want recovery path %q", tab.SessionPath, recoveryPath)
+	}
+	if tab.TopicID == "" || tab.TopicID == originalTopic {
+		t.Fatalf("tab topic ID = %q, want fresh recovery topic", tab.TopicID)
+	}
+	meta, ok, err := agent.LoadBranchMeta(recoveryPath)
+	if err != nil || !ok {
+		t.Fatalf("LoadBranchMeta recovery ok=%v err=%v", ok, err)
+	}
+	if !meta.Recovered || meta.TopicID != tab.TopicID || meta.TopicTitle != tab.TopicTitle {
+		t.Fatalf("recovery meta = %+v, tab topic=%q/%q", meta, tab.TopicID, tab.TopicTitle)
+	}
+	tabMeta := app.tabMeta(tab, true)
+	if !tabMeta.Recovered || tabMeta.RecoveryDigest != meta.RecoveryDigest || tabMeta.RecoveryParentID != string(meta.ParentID) {
+		t.Fatalf("tab recovery meta = %+v, want digest %q parent %q", tabMeta, meta.RecoveryDigest, meta.ParentID)
+	}
+	nodes := app.ListProjectTree()
+	found := false
+	var walk func([]ProjectNode)
+	walk = func(list []ProjectNode) {
+		for _, node := range list {
+			if node.TopicID == tab.TopicID {
+				found = true
+				if !node.Recovered || node.RecoveryDigest != meta.RecoveryDigest || node.RecoveryParentID != string(meta.ParentID) {
+					t.Fatalf("project tree recovery node = %+v, want digest %q parent %q", node, meta.RecoveryDigest, meta.ParentID)
+				}
+			}
+			walk(node.Children)
+		}
+	}
+	walk(nodes)
+	if !found {
+		t.Fatalf("project tree did not include recovery topic %q: %#v", tab.TopicID, nodes)
+	}
+}
+
+func TestDesktopSnapshotConflictRecoveryRequiresRecoveryLease(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	root := globalTabWorkspaceRoot()
+	dir := desktopSessionDir(root)
+	if err := os.MkdirAll(dir, 0o755); err != nil {
+		t.Fatalf("mkdir sessions: %v", err)
+	}
+	originalPath := filepath.Join(dir, "session.jsonl")
+	current := agent.NewSession("sys")
+	current.Add(provider.Message{Role: provider.RoleUser, Content: "first"})
+	current.Add(provider.Message{Role: provider.RoleAssistant, Content: "one"})
+	current.Add(provider.Message{Role: provider.RoleUser, Content: "disk second"})
+	if err := current.Save(originalPath); err != nil {
+		t.Fatalf("Save current: %v", err)
+	}
+
+	staleSess := agent.NewSession("sys")
+	staleSess.Add(provider.Message{Role: provider.RoleUser, Content: "first"})
+	staleSess.Add(provider.Message{Role: provider.RoleAssistant, Content: "one"})
+	staleSess.Add(provider.Message{Role: provider.RoleUser, Content: "local second"})
+	recovery, err := staleSess.SaveRecoveryBranch(agent.RecoveryBranchOptions{
+		OriginalPath: originalPath,
+		BranchMeta: agent.BranchMeta{
+			Name:       agent.RecoveryBranchDefaultName,
+			Scope:      "global",
+			TopicID:    "topic_recovery",
+			TopicTitle: "Recovery",
+		},
+	})
+	if err != nil {
+		t.Fatalf("SaveRecoveryBranch: %v", err)
+	}
+	lease, err := agent.TryAcquireSessionLease(recovery.Path)
+	if err != nil {
+		t.Fatalf("TryAcquireSessionLease recovery: %v", err)
+	}
+	defer lease.Release()
+
+	staleExec := agent.New(stubProvider{}, tool.NewRegistry(), staleSess, agent.Options{}, event.Discard)
+	runtimeEvents := make(chan runtimeEventEnvelope, 4)
+	app := &App{
+		ctx:              context.Background(),
+		tabs:             map[string]*WorkspaceTab{},
+		detachedSessions: map[string]*WorkspaceTab{},
+		activeTabID:      "recovery_tab",
+	}
+	app.runtimeEvents.emit = func(ctx context.Context, name string, payload ...interface{}) {
+		runtimeEvents <- runtimeEventEnvelope{
+			ctx:     ctx,
+			name:    name,
+			payload: append([]interface{}(nil), payload...),
+		}
+	}
+	tab := &WorkspaceTab{
+		ID:            "recovery_tab",
+		Scope:         "global",
+		WorkspaceRoot: root,
+		TopicID:       "topic_original",
+		TopicTitle:    "Original",
+		SessionPath:   originalPath,
+		Ready:         true,
+		model:         "test-model",
+		disabledMCP:   map[string]ServerView{},
+	}
+	tab.sink = &tabEventSink{tabID: tab.ID, app: app}
+	tab.Ctrl = control.New(control.Options{
+		Executor:            staleExec,
+		SessionDir:          dir,
+		SessionPath:         originalPath,
+		Label:               "test",
+		Sink:                tab.sink,
+		SessionRecoveryMeta: app.tabSessionRecoveryMeta(tab),
+		OnSessionRecovered:  app.handleTabSessionRecovered(tab),
+	})
+	app.tabs[tab.ID] = tab
+
+	err = tab.Ctrl.Snapshot()
+	if !errors.Is(err, agent.ErrSessionLeaseHeld) {
+		t.Fatalf("Snapshot err = %v, want ErrSessionLeaseHeld", err)
+	}
+	if got := tab.Ctrl.SessionPath(); got != originalPath {
+		t.Fatalf("controller session path = %q, want original %q", got, originalPath)
+	}
+	if tab.SessionPath != originalPath {
+		t.Fatalf("tab session path = %q, want original %q", tab.SessionPath, originalPath)
+	}
+	if tab.TopicID != "topic_original" {
+		t.Fatalf("tab topic ID = %q, want original topic", tab.TopicID)
+	}
+
+	deadline := time.After(time.Second)
+	for {
+		select {
+		case emitted := <-runtimeEvents:
+			if emitted.name != "session:recovery-failed" {
+				continue
+			}
+			if len(emitted.payload) != 1 {
+				t.Fatalf("session:recovery-failed payload count = %d, want 1", len(emitted.payload))
+			}
+			failed, ok := emitted.payload[0].(sessionRecoveryFailedEvent)
+			if !ok {
+				t.Fatalf("session:recovery-failed payload type = %T, want sessionRecoveryFailedEvent", emitted.payload[0])
+			}
+			if failed.Reason != "lease_held" {
+				t.Fatalf("session:recovery-failed reason = %q, want lease_held", failed.Reason)
+			}
+			return
+		case <-deadline:
+			t.Fatal("session:recovery-failed event was not emitted")
+		}
+	}
+}
+
 func TestSetActiveTabBlocksWhenCurrentSessionCannotPersist(t *testing.T) {
 	path := filepath.Join(t.TempDir(), "blocked.jsonl")
 	if err := os.Mkdir(path, 0o755); err != nil {

desktop/app_session_dedup_test.go

@@ -3,6 +3,8 @@ package main
 import (
 	"os"
 	"path/filepath"
+	"strconv"
+	"strings"
 	"testing"
 	"time"
 
@@ -189,6 +191,221 @@ func TestEnsureBlankTabCreatesOneBlankPerProject(t *testing.T) {
 	}
 }
 
+func TestEnsureBlankTabStartsProjectRuntimeWithCurrentWorkspacePrompt(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectA := robustTempDir(t)
+	projectB := robustTempDir(t)
+	if err := addProject(projectA, "Project A"); err != nil {
+		t.Fatalf("add project A: %v", err)
+	}
+	if err := addProject(projectB, "Project B"); err != nil {
+		t.Fatalf("add project B: %v", err)
+	}
+
+	app := NewApp()
+	first, err := app.EnsureBlankTab("project", projectA)
+	if err != nil {
+		t.Fatalf("EnsureBlankTab(project A): %v", err)
+	}
+	tabA := waitForTabReady(t, app, first.ID)
+	if got := normalizeProjectRoot(tabA.Ctrl.WorkspaceRoot()); got != normalizeProjectRoot(projectA) {
+		t.Fatalf("project A controller workspace root = %q, want %q", got, normalizeProjectRoot(projectA))
+	}
+
+	second, err := app.EnsureBlankTab("project", projectB)
+	if err != nil {
+		t.Fatalf("EnsureBlankTab(project B): %v", err)
+	}
+	if second.ID == first.ID {
+		t.Fatalf("EnsureBlankTab reused project A tab %q for project B", second.ID)
+	}
+	tabB := waitForTabReady(t, app, second.ID)
+
+	if got := normalizeProjectRoot(tabB.WorkspaceRoot); got != normalizeProjectRoot(projectB) {
+		t.Fatalf("project B tab workspace root = %q, want %q", got, normalizeProjectRoot(projectB))
+	}
+	if got := normalizeProjectRoot(tabB.Ctrl.WorkspaceRoot()); got != normalizeProjectRoot(projectB) {
+		t.Fatalf("project B controller workspace root = %q, want %q", got, normalizeProjectRoot(projectB))
+	}
+	if !sameDesktopPath(tabB.Ctrl.SessionDir(), desktopSessionDir(projectB)) {
+		t.Fatalf("project B controller session dir = %q, want %q", tabB.Ctrl.SessionDir(), desktopSessionDir(projectB))
+	}
+	if !sameDesktopPath(filepath.Dir(tabB.Ctrl.SessionPath()), desktopSessionDir(projectB)) {
+		t.Fatalf("project B controller session path = %q, want under %q", tabB.Ctrl.SessionPath(), desktopSessionDir(projectB))
+	}
+	sys := systemPromptFrom(tabB.Ctrl.History())
+	if !strings.Contains(sys, "Current workspace: "+strconv.Quote(projectB)) {
+		t.Fatalf("project B system prompt missing current workspace %q:\n%s", projectB, sys)
+	}
+	if strings.Contains(sys, "Current workspace: "+strconv.Quote(projectA)) {
+		t.Fatalf("project B system prompt retained project A workspace %q:\n%s", projectA, sys)
+	}
+}
+
+func TestBlankTabSessionPathRejectsOtherProjectWorkspace(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectA := robustTempDir(t)
+	projectB := robustTempDir(t)
+	pathA, err := createEmptySessionFile(desktopSessionDir(projectA), "test-model")
+	if err != nil {
+		t.Fatalf("create project A empty session: %v", err)
+	}
+	tab := &WorkspaceTab{
+		ID:            "blank-project-b",
+		Scope:         "project",
+		WorkspaceRoot: projectB,
+		SessionPath:   pathA,
+	}
+
+	if blankTabSessionPathHasNoContent(tab) {
+		t.Fatalf("blank tab treated session %q from project A as reusable for project B %q", pathA, projectB)
+	}
+}
+
+func TestForkKeepsProjectWorkspacePrompt(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectA := robustTempDir(t)
+	projectB := robustTempDir(t)
+	if err := addProject(projectA, "Project A"); err != nil {
+		t.Fatalf("add project A: %v", err)
+	}
+	if err := addProject(projectB, "Project B"); err != nil {
+		t.Fatalf("add project B: %v", err)
+	}
+
+	app := NewApp()
+	first, err := app.EnsureBlankTab("project", projectA)
+	if err != nil {
+		t.Fatalf("EnsureBlankTab(project A): %v", err)
+	}
+	waitForTabReady(t, app, first.ID)
+
+	second, err := app.EnsureBlankTab("project", projectB)
+	if err != nil {
+		t.Fatalf("EnsureBlankTab(project B): %v", err)
+	}
+	tabB := waitForTabReady(t, app, second.ID)
+	ctrl := installStubControllerWithCurrentPrompt(t, app, tabB)
+	turn := submitStubTurnAndWaitForCheckpoint(t, ctrl, "project B turn")
+
+	forked, err := app.Fork(turn)
+	if err != nil {
+		t.Fatalf("Fork: %v", err)
+	}
+	if forked.ID == "" || forked.ID == second.ID {
+		t.Fatalf("forked tab ID = %q, want a fresh tab distinct from %q", forked.ID, second.ID)
+	}
+	forkTab := waitForTabReady(t, app, forked.ID)
+	if got := normalizeProjectRoot(forkTab.WorkspaceRoot); got != normalizeProjectRoot(projectB) {
+		t.Fatalf("fork tab workspace root = %q, want %q", got, normalizeProjectRoot(projectB))
+	}
+	if got := normalizeProjectRoot(forkTab.Ctrl.WorkspaceRoot()); got != normalizeProjectRoot(projectB) {
+		t.Fatalf("fork controller workspace root = %q, want %q", got, normalizeProjectRoot(projectB))
+	}
+	sys := systemPromptFrom(forkTab.Ctrl.History())
+	if !strings.Contains(sys, "Current workspace: "+strconv.Quote(projectB)) {
+		t.Fatalf("fork system prompt missing project B workspace %q:\n%s", projectB, sys)
+	}
+	if strings.Contains(sys, "Current workspace: "+strconv.Quote(projectA)) {
+		t.Fatalf("fork system prompt retained project A workspace %q:\n%s", projectA, sys)
+	}
+}
+
+func TestRewindKeepsProjectWorkspacePrompt(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectA := robustTempDir(t)
+	projectB := robustTempDir(t)
+	if err := addProject(projectA, "Project A"); err != nil {
+		t.Fatalf("add project A: %v", err)
+	}
+	if err := addProject(projectB, "Project B"); err != nil {
+		t.Fatalf("add project B: %v", err)
+	}
+
+	app := NewApp()
+	first, err := app.EnsureBlankTab("project", projectA)
+	if err != nil {
+		t.Fatalf("EnsureBlankTab(project A): %v", err)
+	}
+	waitForTabReady(t, app, first.ID)
+
+	second, err := app.EnsureBlankTab("project", projectB)
+	if err != nil {
+		t.Fatalf("EnsureBlankTab(project B): %v", err)
+	}
+	tabB := waitForTabReady(t, app, second.ID)
+	ctrl := installStubControllerWithCurrentPrompt(t, app, tabB)
+	turn := submitStubTurnAndWaitForCheckpoint(t, ctrl, "project B turn")
+
+	if err := app.Rewind(turn, "conversation"); err != nil {
+		t.Fatalf("Rewind: %v", err)
+	}
+	if got := normalizeProjectRoot(tabB.Ctrl.WorkspaceRoot()); got != normalizeProjectRoot(projectB) {
+		t.Fatalf("rewound controller workspace root = %q, want %q", got, normalizeProjectRoot(projectB))
+	}
+	sys := systemPromptFrom(tabB.Ctrl.History())
+	if !strings.Contains(sys, "Current workspace: "+strconv.Quote(projectB)) {
+		t.Fatalf("rewound system prompt missing project B workspace %q:\n%s", projectB, sys)
+	}
+	if strings.Contains(sys, "Current workspace: "+strconv.Quote(projectA)) {
+		t.Fatalf("rewound system prompt retained project A workspace %q:\n%s", projectA, sys)
+	}
+}
+
+func installStubControllerWithCurrentPrompt(t *testing.T, app *App, tab *WorkspaceTab) *control.Controller {
+	t.Helper()
+	if tab == nil || tab.Ctrl == nil {
+		t.Fatal("tab controller is required")
+	}
+	sys := systemPromptFrom(tab.Ctrl.History())
+	if strings.TrimSpace(sys) == "" {
+		t.Fatal("tab controller did not expose a system prompt")
+	}
+	sessionDir := tab.Ctrl.SessionDir()
+	sessionPath := tab.Ctrl.SessionPath()
+	workspaceRoot := tab.Ctrl.WorkspaceRoot()
+	label := tab.Ctrl.Label()
+	tab.Ctrl.Close()
+
+	sess := agent.NewSession(sys)
+	ag := agent.New(stubProvider{}, tool.NewRegistry(), sess, agent.Options{}, event.Discard)
+	ctrl := control.New(control.Options{
+		Runner:        ag,
+		Executor:      ag,
+		SessionDir:    sessionDir,
+		SessionPath:   sessionPath,
+		WorkspaceRoot: workspaceRoot,
+		Label:         label,
+		SystemPrompt:  sys,
+		Sink:          event.Discard,
+	})
+	tab.Ctrl = ctrl
+	app.bindControllerDisplayRecorder(ctrl)
+	return ctrl
+}
+
+func submitStubTurnAndWaitForCheckpoint(t *testing.T, ctrl control.SessionAPI, input string) int {
+	t.Helper()
+	ctrl.SubmitUserTurn(input, input)
+	waitNotRunning(t, ctrl)
+
+	deadline := time.Now().Add(time.Second)
+	for {
+		checkpoints := ctrl.Checkpoints()
+		if len(checkpoints) > 0 {
+			return checkpoints[len(checkpoints)-1].Turn
+		}
+		if time.Now().After(deadline) {
+			t.Fatal("controller did not record a checkpoint")
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+}
+
 func TestEnsureBlankTabResetsReusableAutoTopicTitle(t *testing.T) {
 	isolateDesktopUserDirs(t)
 

desktop/app_test.go

@@ -27,7 +27,9 @@ import (
 	"reasonix/internal/jobs"
 	"reasonix/internal/memory"
 	"reasonix/internal/plugin"
+	"reasonix/internal/pluginpkg"
 	"reasonix/internal/provider"
+	"reasonix/internal/sandbox"
 	"reasonix/internal/tool"
 )
 
@@ -955,21 +957,36 @@ api_key_env = "DEEPSEEK_API_KEY"
 	}
 
 	app := NewApp()
+	if got := app.Settings().Agent.MaxSubagentDepth; got != agent.DefaultMaxSubagentDepth {
+		t.Fatalf("default max subagent depth = %d, want %d", got, agent.DefaultMaxSubagentDepth)
+	}
 	if err := app.SetSubagentModel("deepseek/deepseek-v4-pro"); err != nil {
 		t.Fatalf("SetSubagentModel: %v", err)
 	}
 	if err := app.SetSubagentEffort("max"); err != nil {
 		t.Fatalf("SetSubagentEffort: %v", err)
 	}
+	if err := app.SetMaxSubagentDepth(1); err != nil {
+		t.Fatalf("SetMaxSubagentDepth(1): %v", err)
+	}
+	if err := app.SetMaxSubagentDepth(2); err != nil {
+		t.Fatalf("SetMaxSubagentDepth(2): %v", err)
+	}
 
 	got := app.Settings()
 	if got.SubagentModel != "deepseek/deepseek-v4-pro" || got.SubagentEffort != "max" {
 		t.Fatalf("subagent settings = model:%q effort:%q", got.SubagentModel, got.SubagentEffort)
 	}
+	if got.Agent.MaxSubagentDepth != 2 {
+		t.Fatalf("max subagent depth = %d, want 2", got.Agent.MaxSubagentDepth)
+	}
 	cfg := config.LoadForEdit(config.UserConfigPath())
 	if cfg.Agent.SubagentModel != "deepseek/deepseek-v4-pro" || cfg.Agent.SubagentEffort != "max" {
 		t.Fatalf("saved config = model:%q effort:%q", cfg.Agent.SubagentModel, cfg.Agent.SubagentEffort)
 	}
+	if cfg.Agent.MaxSubagentDepth != 2 {
+		t.Fatalf("saved max_subagent_depth = %d, want 2", cfg.Agent.MaxSubagentDepth)
+	}
 }
 
 func TestSettingsSurfacesOfficialProviderTemplatesSeparately(t *testing.T) {
@@ -1214,6 +1231,385 @@ api_key_env = "DEEPSEEK_API_KEY"
 	}
 }
 
+func TestSettingsSurfacesCuratedProviderPresets(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	view := NewApp().Settings()
+	if len(view.ProviderPresets) < 18 {
+		t.Fatalf("Settings().ProviderPresets length = %d, want curated custom presets", len(view.ProviderPresets))
+	}
+	got := map[string]ProviderPresetView{}
+	for _, preset := range view.ProviderPresets {
+		got[preset.ID] = preset
+	}
+	for _, curated := range config.CuratedProviderPresets() {
+		id := curated.ID
+		preset, ok := got[id]
+		if !ok {
+			t.Fatalf("Settings().ProviderPresets missing %q: %+v", id, view.ProviderPresets)
+		}
+		if preset.KeyEnv == "" || len(preset.ProviderNames) == 0 || len(preset.Models) == 0 {
+			t.Fatalf("preset %q view has missing fields: %+v", id, preset)
+		}
+	}
+}
+
+func providerPresetViewByID(t *testing.T, view SettingsView, id string) ProviderPresetView {
+	t.Helper()
+	for _, preset := range view.ProviderPresets {
+		if preset.ID == id {
+			return preset
+		}
+	}
+	t.Fatalf("Settings().ProviderPresets missing %q: %+v", id, view.ProviderPresets)
+	return ProviderPresetView{}
+}
+
+func TestSettingsMarksPresetAddedWhenSameNameProviderExistsWithoutAccess(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	if err := os.MkdirAll(filepath.Dir(config.UserConfigPath()), 0o755); err != nil {
+		t.Fatalf("mkdir config dir: %v", err)
+	}
+	if err := os.WriteFile(config.UserConfigPath(), []byte(`
+[desktop]
+provider_access = []
+
+[[providers]]
+name = "mimo-api"
+kind = "openai"
+base_url = "https://custom.example/v1"
+models = ["custom-model"]
+default = "custom-model"
+api_key_env = "MIMO_API_KEY"
+`), 0o644); err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	view := NewApp().Settings()
+	presetView := providerPresetViewByID(t, view, "mimo-api")
+	if !presetView.Added || presetView.Status != providerPresetStatusNameConflict || !reflect.DeepEqual(presetView.StatusProviderNames, []string{"mimo-api"}) {
+		t.Fatalf("mimo-api preset view = %+v, want name-conflict because a different same-name provider exists", presetView)
+	}
+
+	var providerView *ProviderView
+	for i := range view.Providers {
+		if view.Providers[i].Name == "mimo-api" {
+			providerView = &view.Providers[i]
+			break
+		}
+	}
+	if providerView == nil {
+		t.Fatal("mimo-api provider view missing")
+	}
+	if providerView.Added {
+		t.Fatalf("mimo-api provider Added = true, want false until provider_access explicitly enables it")
+	}
+}
+
+func TestSettingsMarksLegacyEquivalentPresetAsInstalled(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	preset, ok := config.CuratedProviderPreset("mimo-api")
+	if !ok || len(preset.Entries) == 0 {
+		t.Fatal("missing mimo-api preset")
+	}
+	legacy := preset.Entries[0]
+	legacy.PresetID = ""
+	legacy.PresetVersion = 0
+	cfg := config.Default()
+	if err := cfg.UpsertProvider(legacy); err != nil {
+		t.Fatalf("upsert legacy provider: %v", err)
+	}
+	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
+		t.Fatalf("save config: %v", err)
+	}
+
+	view := NewApp().Settings()
+	presetView := providerPresetViewByID(t, view, "mimo-api")
+	if !presetView.Added || presetView.Status != providerPresetStatusInstalled || !reflect.DeepEqual(presetView.StatusProviderNames, []string{"mimo-api"}) {
+		t.Fatalf("mimo-api preset view = %+v, want installed for legacy equivalent config", presetView)
+	}
+}
+
+func TestSettingsMarksPresetWithChangedCoreConfigAsModified(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	preset, ok := config.CuratedProviderPreset("mimo-api")
+	if !ok || len(preset.Entries) == 0 {
+		t.Fatal("missing mimo-api preset")
+	}
+	modified := preset.Entries[0]
+	modified.BaseURL = "https://custom.example/v1"
+	cfg := config.Default()
+	if err := cfg.UpsertProvider(modified); err != nil {
+		t.Fatalf("upsert modified provider: %v", err)
+	}
+	cfg.Desktop.ProviderAccess = []string{"mimo-api"}
+	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
+		t.Fatalf("save config: %v", err)
+	}
+
+	view := NewApp().Settings()
+	presetView := providerPresetViewByID(t, view, "mimo-api")
+	if !presetView.Added || presetView.Status != providerPresetStatusInstalledModified || !reflect.DeepEqual(presetView.StatusProviderNames, []string{"mimo-api"}) {
+		t.Fatalf("mimo-api preset view = %+v, want installed-modified for edited preset provider", presetView)
+	}
+}
+
+func TestSettingsMarksSimilarProviderPresetWithoutBlockingAdd(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	preset, ok := config.CuratedProviderPreset("mimo-api")
+	if !ok || len(preset.Entries) == 0 {
+		t.Fatal("missing mimo-api preset")
+	}
+	similar := preset.Entries[0]
+	similar.Name = "my-mimo"
+	similar.PresetID = ""
+	similar.PresetVersion = 0
+	cfg := config.Default()
+	if err := cfg.UpsertProvider(similar); err != nil {
+		t.Fatalf("upsert similar provider: %v", err)
+	}
+	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
+		t.Fatalf("save config: %v", err)
+	}
+
+	view := NewApp().Settings()
+	presetView := providerPresetViewByID(t, view, "mimo-api")
+	if presetView.Added || presetView.Status != providerPresetStatusSimilarExisting || !reflect.DeepEqual(presetView.StatusProviderNames, []string{"my-mimo"}) {
+		t.Fatalf("mimo-api preset view = %+v, want non-blocking similar-existing status", presetView)
+	}
+}
+
+func TestAddProviderPresetAccessSavesEditableProviderAndKey(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	t.Setenv("MIMO_API_KEY", "")
+	os.Unsetenv("MIMO_API_KEY")
+
+	if warning, err := NewApp().AddProviderPresetAccess("mimo-api", "sk-mimo"); err != nil {
+		t.Fatalf("AddProviderPresetAccess: %v", err)
+	} else if warning != "" {
+		t.Fatalf("AddProviderPresetAccess warning = %q, want none", warning)
+	}
+
+	cfg := config.LoadForEdit(config.UserConfigPath())
+	p, ok := cfg.Provider("mimo-api")
+	if !ok {
+		t.Fatal("mimo-api provider not saved")
+	}
+	if p.Kind != "openai" || p.BaseURL != "https://api.xiaomimimo.com/v1" || p.Default != "mimo-v2.5-pro" {
+		t.Fatalf("mimo-api provider after preset add = %+v", p)
+	}
+	if p.PresetID != "mimo-api" || p.PresetVersion != config.ProviderPresetVersion {
+		t.Fatalf("mimo-api preset metadata = %q/%d, want mimo-api/%d", p.PresetID, p.PresetVersion, config.ProviderPresetVersion)
+	}
+	if !p.NoProxy {
+		t.Fatal("mimo-api preset should save no_proxy = true")
+	}
+	if !p.HasVisionModel("mimo-v2.5") || p.HasVisionModel("mimo-v2.5-pro") {
+		t.Fatalf("mimo vision_models = %+v, want only vision-capable MiMo models", p.VisionModels)
+	}
+	if price := p.PriceForModel("mimo-v2.5-pro"); price == nil || price.Currency != "¥" {
+		t.Fatalf("mimo-v2.5-pro price = %+v, want RMB pricing", price)
+	}
+	if !providerAccessSet(cfg.Desktop.ProviderAccess)["mimo-api"] {
+		t.Fatalf("provider_access missing mimo-api: %+v", cfg.Desktop.ProviderAccess)
+	}
+	data, err := os.ReadFile(config.UserCredentialsPath())
+	if err != nil {
+		t.Fatalf("read saved credentials: %v", err)
+	}
+	if !strings.Contains(string(data), "MIMO_API_KEY=sk-mimo") {
+		t.Fatalf("saved credentials missing MiMo key:\n%s", data)
+	}
+
+	view := NewApp().Settings()
+	var presetView *ProviderPresetView
+	var providerView *ProviderView
+	for i := range view.ProviderPresets {
+		if view.ProviderPresets[i].ID == "mimo-api" {
+			presetView = &view.ProviderPresets[i]
+		}
+	}
+	for i := range view.Providers {
+		if view.Providers[i].Name == "mimo-api" {
+			providerView = &view.Providers[i]
+		}
+	}
+	if presetView == nil || !presetView.Added || presetView.Status != providerPresetStatusInstalled || !presetView.KeySet {
+		t.Fatalf("mimo-api preset view = %+v, want installed/key-set", presetView)
+	}
+	if providerView == nil || providerView.BuiltIn || !providerView.Added || !providerView.KeySet {
+		t.Fatalf("mimo provider view = %+v, want editable added custom provider with key", providerView)
+	}
+}
+
+func TestAddProviderPresetAccessDoesNotOverwriteExistingProvider(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	t.Setenv("MIMO_API_KEY", "")
+	os.Unsetenv("MIMO_API_KEY")
+	setDesktopTestCredential(t, "MIMO_API_KEY", "sk-original")
+
+	cfg := config.Default()
+	custom := config.ProviderEntry{
+		Name:      "mimo-api",
+		Kind:      "openai",
+		BaseURL:   "https://custom.example/v1",
+		Models:    []string{"custom-model"},
+		Default:   "custom-model",
+		APIKeyEnv: "MIMO_API_KEY",
+		Headers:   map[string]string{"X-Custom": "keep-me"},
+	}
+	if err := cfg.UpsertProvider(custom); err != nil {
+		t.Fatalf("upsert custom provider: %v", err)
+	}
+	cfg.Desktop.ProviderAccess = []string{"mimo-api"}
+	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
+		t.Fatalf("save config: %v", err)
+	}
+
+	if warning, err := NewApp().AddProviderPresetAccess("mimo-api", "sk-new"); err == nil {
+		t.Fatal("AddProviderPresetAccess unexpectedly overwrote an existing provider")
+	} else if !strings.Contains(err.Error(), "provider name(s) already exist") {
+		t.Fatalf("AddProviderPresetAccess error = %v, want name-exists guard", err)
+	} else if warning != "" {
+		t.Fatalf("AddProviderPresetAccess warning = %q, want none on rejected add", warning)
+	}
+
+	cfg = config.LoadForEdit(config.UserConfigPath())
+	got, ok := cfg.Provider("mimo-api")
+	if !ok {
+		t.Fatal("mimo-api provider missing after rejected add")
+	}
+	if got.BaseURL != custom.BaseURL || got.DefaultModel() != custom.DefaultModel() || !reflect.DeepEqual(got.ModelList(), custom.ModelList()) || !reflect.DeepEqual(got.Headers, custom.Headers) {
+		t.Fatalf("mimo-api provider was overwritten: %+v, want custom %+v", got, custom)
+	}
+	data, err := os.ReadFile(config.UserCredentialsPath())
+	if err != nil {
+		t.Fatalf("read saved credentials: %v", err)
+	}
+	if strings.Contains(string(data), "sk-new") || !strings.Contains(string(data), "MIMO_API_KEY=sk-original") {
+		t.Fatalf("credentials changed after rejected add:\n%s", data)
+	}
+}
+
+func TestResetProviderPresetAccessOverwritesSameNameProvider(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	t.Setenv("MIMO_API_KEY", "")
+	os.Unsetenv("MIMO_API_KEY")
+	setDesktopTestCredential(t, "MIMO_API_KEY", "sk-original")
+
+	cfg := config.Default()
+	custom := config.ProviderEntry{
+		Name:      "mimo-api",
+		Kind:      "openai",
+		BaseURL:   "https://custom.example/v1",
+		Models:    []string{"custom-model"},
+		Default:   "custom-model",
+		APIKeyEnv: "MIMO_API_KEY",
+		Headers:   map[string]string{"X-Custom": "remove-me"},
+	}
+	if err := cfg.UpsertProvider(custom); err != nil {
+		t.Fatalf("upsert custom provider: %v", err)
+	}
+	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
+		t.Fatalf("save config: %v", err)
+	}
+
+	if err := NewApp().ResetProviderPresetAccess("mimo-api"); err != nil {
+		t.Fatalf("ResetProviderPresetAccess: %v", err)
+	}
+
+	cfg = config.LoadForEdit(config.UserConfigPath())
+	got, ok := cfg.Provider("mimo-api")
+	if !ok {
+		t.Fatal("mimo-api provider missing after reset")
+	}
+	if got.BaseURL != "https://api.xiaomimimo.com/v1" || got.DefaultModel() != "mimo-v2.5-pro" || got.PresetID != "mimo-api" || got.PresetVersion != config.ProviderPresetVersion {
+		t.Fatalf("mimo-api provider after reset = %+v, want preset template", got)
+	}
+	if len(got.Headers) != 0 {
+		t.Fatalf("mimo-api headers after reset = %+v, want preset headers", got.Headers)
+	}
+	if !providerAccessSet(cfg.Desktop.ProviderAccess)["mimo-api"] {
+		t.Fatalf("provider_access missing mimo-api after reset: %+v", cfg.Desktop.ProviderAccess)
+	}
+	data, err := os.ReadFile(config.UserCredentialsPath())
+	if err != nil {
+		t.Fatalf("read saved credentials: %v", err)
+	}
+	if !strings.Contains(string(data), "MIMO_API_KEY=sk-original") {
+		t.Fatalf("credentials changed after reset:\n%s", data)
+	}
+
+	presetView := providerPresetViewByID(t, NewApp().Settings(), "mimo-api")
+	if !presetView.Added || presetView.Status != providerPresetStatusInstalled {
+		t.Fatalf("mimo-api preset view = %+v, want installed after reset", presetView)
+	}
+}
+
+func TestResetProviderPresetAccessRejectsMissingSameNameProvider(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	if err := NewApp().ResetProviderPresetAccess("mimo-api"); err == nil {
+		t.Fatal("ResetProviderPresetAccess unexpectedly reset a missing provider")
+	} else if !strings.Contains(err.Error(), "no same-name provider exists") {
+		t.Fatalf("ResetProviderPresetAccess error = %v, want missing same-name provider guard", err)
+	}
+}
+
+func TestAddEveryProviderPresetAccessInstallsTemplate(t *testing.T) {
+	for _, preset := range config.CuratedProviderPresets() {
+		preset := preset
+		t.Run(preset.ID, func(t *testing.T) {
+			isolateDesktopUserDirs(t)
+
+			if warning, err := NewApp().AddProviderPresetAccess(preset.ID, "sk-test"); err != nil {
+				t.Fatalf("AddProviderPresetAccess(%q): %v", preset.ID, err)
+			} else if warning != "" {
+				t.Fatalf("AddProviderPresetAccess(%q) warning = %q, want none", preset.ID, warning)
+			}
+
+			cfg := config.LoadForEdit(config.UserConfigPath())
+			access := providerAccessSet(cfg.Desktop.ProviderAccess)
+			for _, entry := range preset.Entries {
+				got, ok := cfg.Provider(entry.Name)
+				if !ok {
+					t.Fatalf("provider %q from preset %q was not saved", entry.Name, preset.ID)
+				}
+				if !access[entry.Name] {
+					t.Fatalf("provider_access for preset %q missing %q: %+v", preset.ID, entry.Name, cfg.Desktop.ProviderAccess)
+				}
+				if got.Kind != entry.Kind || got.BaseURL != entry.BaseURL || got.DefaultModel() != entry.DefaultModel() || got.APIKeyEnv != entry.APIKeyEnv || got.AuthHeader != entry.AuthHeader || got.NoProxy != entry.NoProxy {
+					t.Fatalf("provider %q core fields = %+v, want template %+v", entry.Name, got, entry)
+				}
+				if got.PresetID != preset.ID || got.PresetVersion != config.ProviderPresetVersion {
+					t.Fatalf("provider %q preset metadata = %q/%d, want %q/%d", entry.Name, got.PresetID, got.PresetVersion, preset.ID, config.ProviderPresetVersion)
+				}
+				if got.ContextWindow != entry.ContextWindow || got.Thinking != entry.Thinking || got.DefaultEffort != entry.DefaultEffort || got.ReasoningProtocol != entry.ReasoningProtocol {
+					t.Fatalf("provider %q capability fields = %+v, want template %+v", entry.Name, got, entry)
+				}
+				if !reflect.DeepEqual(got.ModelList(), entry.ModelList()) || !reflect.DeepEqual(got.VisionModels, entry.VisionModels) || !reflect.DeepEqual(got.SupportedEfforts, entry.SupportedEfforts) {
+					t.Fatalf("provider %q models/capabilities = %+v, want template %+v", entry.Name, got, entry)
+				}
+				if !reflect.DeepEqual(got.Headers, entry.Headers) || !reflect.DeepEqual(got.ExtraBody, entry.ExtraBody) {
+					t.Fatalf("provider %q request extras = %+v, want template %+v", entry.Name, got, entry)
+				}
+			}
+
+			view := NewApp().Settings()
+			var presetView *ProviderPresetView
+			for i := range view.ProviderPresets {
+				if view.ProviderPresets[i].ID == preset.ID {
+					presetView = &view.ProviderPresets[i]
+					break
+				}
+			}
+			if presetView == nil || !presetView.Added || presetView.Status != providerPresetStatusInstalled || !presetView.KeySet || !presetView.Configured {
+				t.Fatalf("preset view for %q = %+v, want installed/key-set/configured", preset.ID, presetView)
+			}
+		})
+	}
+}
+
 func TestAddOfficialProviderAccessRejectsBackgroundJobsBeforeSavingKey(t *testing.T) {
 	isolateDesktopUserDirs(t)
 	t.Setenv("DEEPSEEK_API_KEY", "")
@@ -1523,6 +1919,13 @@ func TestModelsForTabListsMimoAPIPaidAccess(t *testing.T) {
 	setDesktopTestCredential(t, "MIMO_API_KEY", "sk-test")
 
 	cfg := config.Default()
+	preset, ok := config.CuratedProviderPreset("mimo-api")
+	if !ok || len(preset.Entries) == 0 {
+		t.Fatal("mimo-api preset missing")
+	}
+	if err := cfg.UpsertProvider(preset.Entries[0]); err != nil {
+		t.Fatalf("upsert mimo-api preset: %v", err)
+	}
 	cfg.DefaultModel = "mimo-api/mimo-v2.5-pro"
 	cfg.Desktop.ProviderAccess = []string{"mimo-api"}
 	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
@@ -1534,14 +1937,13 @@ func TestModelsForTabListsMimoAPIPaidAccess(t *testing.T) {
 	for _, want := range []string{
 		"mimo-api/mimo-v2.5-pro",
 		"mimo-api/mimo-v2.5",
-		"mimo-api/mimo-v2-omni",
 	} {
 		if !refs[want] {
 			t.Fatalf("Models() refs = %+v, missing %s", models, want)
 		}
 	}
-	if len(models) != 3 {
-		t.Fatalf("Models() len = %d, want 3: %+v", len(models), models)
+	if len(models) != 2 {
+		t.Fatalf("Models() len = %d, want 2: %+v", len(models), models)
 	}
 }
 
@@ -1871,6 +2273,34 @@ func TestEnsureTabControllerWorkspaceRebuildsStaleWorkspace(t *testing.T) {
 	assertTabRebuiltToPinnedWorkspace(t, f)
 }
 
+func TestEnsureTabControllerWorkspaceWarnsWhenPinnedSessionSwitchesWorkspace(t *testing.T) {
+	f := newStaleWorkspaceBindingFixture(t, "warn_workspace_switch")
+	events := make(chan event.Event, 8)
+	f.tab.sink.SetBotSink(event.FuncSink(func(e event.Event) {
+		events <- e
+	}))
+
+	if err := f.app.ensureTabControllerWorkspace(f.tab); err != nil {
+		t.Fatalf("ensureTabControllerWorkspace: %v", err)
+	}
+	assertTabRebuiltToPinnedWorkspace(t, f)
+
+	deadline := time.After(2 * time.Second)
+	for {
+		select {
+		case e := <-events:
+			if e.Kind == event.Notice &&
+				e.Level == event.LevelWarn &&
+				strings.Contains(e.Text, f.projectA) &&
+				strings.Contains(e.Text, "switched tab") {
+				return
+			}
+		case <-deadline:
+			t.Fatal("did not receive workspace switch warning notice")
+		}
+	}
+}
+
 func TestSteerForTabReconcilesStaleWorkspaceBeforeIdleFallback(t *testing.T) {
 	f := newStaleWorkspaceBindingFixture(t, "steer_idle_fallback")
 
@@ -3676,6 +4106,17 @@ func TestDesktopSessionAPIsUseControllerSessionDir(t *testing.T) {
 	if err := app.RenameSession(pathA, "A title"); err != nil {
 		t.Fatalf("RenameSession in active session dir: %v", err)
 	}
+	meta, ok, err := agent.LoadBranchMeta(pathA)
+	if err != nil || !ok {
+		t.Fatalf("LoadBranchMeta after RenameSession ok=%v err=%v", ok, err)
+	}
+	if meta.CustomTitle != "A title" {
+		t.Fatalf("custom title should be written to branch meta, got %q", meta.CustomTitle)
+	}
+	sessions = app.ListSessions()
+	if len(sessions) != 1 || sessions[0].Title != "A title" {
+		t.Fatalf("ListSessions should return custom title from branch meta, got %+v", sessions)
+	}
 	if titles := loadSessionTitles(dirA); titles["a.jsonl"] != "A title" {
 		t.Fatalf("title should be written beside the active session, got %+v", titles)
 	}
@@ -4209,6 +4650,51 @@ args = ["-y", "@playwright/mcp"]
 	t.Fatalf("playwright MCP missing from Capabilities: %+v", view.Servers)
 }
 
+func TestCapabilitiesIncludesInstalledPlugins(t *testing.T) {
+	home := isolateDesktopUserDirs(t)
+	reasonixHome := filepath.Join(home, ".reasonix")
+	root := filepath.Join(reasonixHome, "plugins", "superpowers")
+	if err := os.MkdirAll(filepath.Join(root, "skills"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.MkdirAll(filepath.Join(root, "skills", "plan"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(root, "skills", "plan", "SKILL.md"), []byte("---\ndescription: Plan work\n---\nbody"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.MkdirAll(filepath.Join(root, ".codex-plugin"), 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(root, ".codex-plugin", "plugin.json"), []byte(`{
+  "name": "superpowers",
+  "version": "6.1.0",
+  "description": "Planning workflows",
+  "skills": "./skills/"
+}`), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := pluginpkg.Upsert(reasonixHome, pluginpkg.InstalledPlugin{
+		Name:         "superpowers",
+		Root:         "plugins/superpowers",
+		Version:      "6.1.0",
+		Description:  "Planning workflows",
+		ManifestKind: "codex",
+		Enabled:      true,
+	}); err != nil {
+		t.Fatal(err)
+	}
+
+	app := NewApp()
+	plugins := app.Capabilities().Plugins
+	if len(plugins) != 1 || plugins[0].Name != "superpowers" || plugins[0].Skills != 1 {
+		t.Fatalf("Capabilities().Plugins = %+v", plugins)
+	}
+	if len(plugins[0].SkillDetails) != 1 || plugins[0].SkillDetails[0].Invocation != "/plan" {
+		t.Fatalf("Capabilities().Plugins skill details = %+v", plugins[0].SkillDetails)
+	}
+}
+
 func TestDesktopSharedHostBackgroundMCPAutoConnectsOnBoot(t *testing.T) {
 	isolateDesktopUserDirs(t)
 	dir := robustTempDir(t)
@@ -5304,6 +5790,129 @@ func TestRunShellForTabRoutesToRequestedTab(t *testing.T) {
 	}
 }
 
+func TestRunShellForTabStaysBoundDuringRapidProjectTabSwitching(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectA := t.TempDir()
+	projectB := t.TempDir()
+	globalRoot := t.TempDir()
+	shellEvents := make(chan event.Event, 64)
+	projectEvents := make(chan event.Event, 64)
+	globalEvents := make(chan event.Event, 64)
+	shellCtrl := control.New(control.Options{
+		Sink:          event.FuncSink(func(e event.Event) { shellEvents <- e }),
+		WorkspaceRoot: projectA,
+	})
+	projectCtrl := control.New(control.Options{
+		Sink:          event.FuncSink(func(e event.Event) { projectEvents <- e }),
+		WorkspaceRoot: projectB,
+	})
+	globalCtrl := control.New(control.Options{
+		Sink:          event.FuncSink(func(e event.Event) { globalEvents <- e }),
+		WorkspaceRoot: globalRoot,
+	})
+	defer shellCtrl.Close()
+	defer projectCtrl.Close()
+	defer globalCtrl.Close()
+
+	app := &App{
+		tabs: map[string]*WorkspaceTab{
+			"shell":     {ID: "shell", Scope: "project", WorkspaceRoot: projectA, Ctrl: shellCtrl, Ready: true},
+			"project-b": {ID: "project-b", Scope: "project", WorkspaceRoot: projectB, Ctrl: projectCtrl, Ready: true},
+			"global":    {ID: "global", Scope: "global", WorkspaceRoot: globalRoot, Ctrl: globalCtrl, Ready: true},
+		},
+		tabOrder:    []string{"shell", "project-b", "global"},
+		activeTabID: "shell",
+	}
+
+	marker := "shell-route-marker.txt"
+	if err := app.RunShellForTab("shell", longRunningMarkerCommand(marker)); err != nil {
+		t.Fatalf("RunShellForTab: %v", err)
+	}
+	waitForShellDispatch(t, shellEvents, marker)
+	waitForFile(t, filepath.Join(projectA, marker), "shell")
+
+	for i := 0; i < 8; i++ {
+		if err := app.SetActiveTab("project-b"); err != nil {
+			t.Fatalf("SetActiveTab(project-b): %v", err)
+		}
+		if err := app.SetActiveTab("global"); err != nil {
+			t.Fatalf("SetActiveTab(global): %v", err)
+		}
+		if err := app.SetActiveTab("shell"); err != nil {
+			t.Fatalf("SetActiveTab(shell): %v", err)
+		}
+	}
+	if err := app.SetActiveTab("project-b"); err != nil {
+		t.Fatalf("SetActiveTab(project-b final): %v", err)
+	}
+	app.CancelTab("shell")
+
+	cancelled := false
+	deadline := time.After(15 * time.Second)
+	for {
+		select {
+		case e := <-shellEvents:
+			if e.Kind == event.ToolResult && e.Tool.Name == "bash" {
+				cancelled = e.Tool.Err != ""
+			}
+			if e.Kind == event.TurnDone {
+				if !cancelled {
+					t.Fatal("shell tab finished without a cancelled shell result")
+				}
+				if _, err := os.Stat(filepath.Join(projectB, marker)); !errors.Is(err, os.ErrNotExist) {
+					t.Fatalf("shell marker appeared in project-b workspace: %v", err)
+				}
+				if got := activeTabIDForTest(app); got != "project-b" {
+					t.Fatalf("active tab = %q, want project-b after background shell cancel", got)
+				}
+				assertNoEvents(t, projectEvents, "project-b")
+				assertNoEvents(t, globalEvents, "global")
+				return
+			}
+		case <-deadline:
+			t.Fatal("timed out waiting for shell tab cancellation")
+		}
+	}
+}
+
+func longRunningMarkerCommand(marker string) string {
+	if sandbox.ResolveShell("", "", nil).Kind == sandbox.ShellPowerShell {
+		return fmt.Sprintf("Set-Content -LiteralPath %s -Value shell; Start-Sleep -Seconds 30", marker)
+	}
+	return fmt.Sprintf("printf shell > %s; sleep 30", marker)
+}
+
+func waitForShellDispatch(t *testing.T, ch <-chan event.Event, marker string) {
+	t.Helper()
+	deadline := time.After(5 * time.Second)
+	for {
+		select {
+		case e := <-ch:
+			if e.Kind == event.ToolDispatch && strings.Contains(e.Tool.Args, marker) {
+				return
+			}
+		case <-deadline:
+			t.Fatal("timed out waiting for shell dispatch")
+		}
+	}
+}
+
+func activeTabIDForTest(app *App) string {
+	app.mu.RLock()
+	defer app.mu.RUnlock()
+	return app.activeTabID
+}
+
+func assertNoEvents(t *testing.T, ch <-chan event.Event, name string) {
+	t.Helper()
+	select {
+	case e := <-ch:
+		t.Fatalf("%s received event while shell ran in another tab: %+v", name, e)
+	default:
+	}
+}
+
 type blockingRunner struct {
 	started chan struct{}
 	release chan struct{}

desktop/bot_connection_app.go

@@ -15,6 +15,7 @@ import (
 	"reasonix/internal/bot"
 	"reasonix/internal/bot/feishu"
 	"reasonix/internal/bot/weixin"
+	"reasonix/internal/botruntime"
 	"reasonix/internal/config"
 )
 
@@ -48,6 +49,7 @@ type BotConnectionView struct {
 	Model            string                            `json:"model"`
 	ToolApprovalMode string                            `json:"toolApprovalMode"`
 	WorkspaceRoot    string                            `json:"workspaceRoot"`
+	Access           BotAccessView                     `json:"access"`
 	Credential       BotConnectionCredentialView       `json:"credential"`
 	SessionMappings  []BotConnectionSessionMappingView `json:"sessionMappings"`
 	LastError        string                            `json:"lastError"`
@@ -164,6 +166,7 @@ func (a *App) PollBotConnectionInstall(installID string) (BotInstallPollResult,
 			Label:      "微信",
 			Enabled:    true,
 			Status:     "connected",
+			Access:     botInstallAccess(result.UserID),
 			Credential: config.BotConnectionCredential{AccountID: result.AccountID, TokenEnv: "WEIXIN_BOT_TOKEN"},
 		}, func(c *config.Config) {
 			c.Bot.Enabled = true
@@ -518,6 +521,7 @@ func (a *App) pollFeishuConnectionInstall(installID string, session *botInstallS
 		Label:      label,
 		Enabled:    true,
 		Status:     "connected",
+		Access:     botInstallAccess(userID),
 		Credential: config.BotConnectionCredential{AppID: appID, AppSecretEnv: secretEnv},
 	}, func(c *config.Config) {
 		c.Bot.Enabled = true
@@ -545,6 +549,9 @@ func (a *App) upsertBotConnection(conn config.BotConnectionConfig, updateLegacy
 	if conn.Status == "" {
 		conn.Status = "connected"
 	}
+	if normalizeBotConnectionToolApprovalMode(conn.ToolApprovalMode) == "" {
+		conn.ToolApprovalMode = "ask"
+	}
 	if conn.ID == "" {
 		conn.ID = connectionID(conn.Provider, conn.Domain)
 	}
@@ -556,6 +563,9 @@ func (a *App) upsertBotConnection(conn config.BotConnectionConfig, updateLegacy
 		for i, existing := range c.Bot.Connections {
 			if existing.ID == conn.ID {
 				conn.CreatedAt = firstNonEmptyBot(existing.CreatedAt, conn.CreatedAt)
+				if !botruntime.BotAccessActive(conn.Access) && botruntime.BotAccessActive(existing.Access) {
+					conn.Access = existing.Access
+				}
 				c.Bot.Connections[i] = conn
 				replaced = true
 				break
@@ -697,6 +707,7 @@ func botConnectionView(conn config.BotConnectionConfig) BotConnectionView {
 	return BotConnectionView{
 		ID: conn.ID, Provider: conn.Provider, Domain: conn.Domain, Label: conn.Label, Enabled: conn.Enabled, Status: conn.Status,
 		Model: conn.Model, ToolApprovalMode: normalizeBotConnectionToolApprovalMode(conn.ToolApprovalMode), WorkspaceRoot: conn.WorkspaceRoot,
+		Access: botAccessViewFromConfig(conn.Access),
 		Credential: BotConnectionCredentialView{
 			AppID: conn.Credential.AppID, AppSecretEnv: conn.Credential.AppSecretEnv, AccountID: conn.Credential.AccountID, TokenEnv: conn.Credential.TokenEnv,
 			SecretSet: botCredentialSecretSet(conn),
@@ -763,8 +774,9 @@ func botConnectionConfig(view BotConnectionView) config.BotConnectionConfig {
 		Enabled:          view.Enabled,
 		Status:           strings.TrimSpace(view.Status),
 		Model:            strings.TrimSpace(view.Model),
-		ToolApprovalMode: normalizeBotConnectionToolApprovalMode(view.ToolApprovalMode),
+		ToolApprovalMode: firstNonEmptyBot(normalizeBotConnectionToolApprovalMode(view.ToolApprovalMode), "ask"),
 		WorkspaceRoot:    strings.TrimSpace(view.WorkspaceRoot),
+		Access:           botAccessConfigFromView(view.Access),
 		Credential: config.BotConnectionCredential{
 			AppID:        strings.TrimSpace(view.Credential.AppID),
 			AppSecretEnv: strings.TrimSpace(view.Credential.AppSecretEnv),
@@ -873,6 +885,15 @@ func connectionID(provider, domain string) string {
 	return strings.Trim(strings.ToLower(provider+"-"+domain), "-")
 }
 
+func botInstallAccess(userID string) config.BotAccessConfig {
+	userID = strings.TrimSpace(userID)
+	access := config.BotAccessConfig{Enabled: true, PairingEnabled: true}
+	if userID != "" {
+		access.Users = []string{userID}
+	}
+	return access
+}
+
 func randomInstallID() string {
 	var b [12]byte
 	if _, err := rand.Read(b[:]); err != nil {

desktop/bot_runtime_app.go

@@ -33,6 +33,30 @@ func newDesktopBotRuntime() *desktopBotRuntime {
 	return &desktopBotRuntime{status: BotRuntimeStatusView{Status: "stopped", Message: "bot runtime is not started"}}
 }
 
+func desktopBotChannelsWithLegacyQQ(qq config.QQBotConfig, channels map[bot.Platform]bot.ChannelConfig, connectionChannels map[string]bot.ChannelConfig) (map[bot.Platform]bot.ChannelConfig, map[string]bot.ChannelConfig) {
+	channel := bot.ChannelConfig{
+		Model:            strings.TrimSpace(qq.Model),
+		ToolApprovalMode: normalizeBotConnectionToolApprovalMode(qq.ToolApprovalMode),
+		WorkspaceRoot:    strings.TrimSpace(qq.WorkspaceRoot),
+	}
+	if channel.Model == "" && channel.ToolApprovalMode == "" && channel.WorkspaceRoot == "" {
+		return channels, connectionChannels
+	}
+	if channels == nil {
+		channels = make(map[bot.Platform]bot.ChannelConfig)
+	}
+	if _, ok := channels[bot.PlatformQQ]; !ok {
+		channels[bot.PlatformQQ] = channel
+	}
+	if connectionChannels == nil {
+		connectionChannels = make(map[string]bot.ChannelConfig)
+	}
+	if _, ok := connectionChannels[string(bot.PlatformQQ)]; !ok {
+		connectionChannels[string(bot.PlatformQQ)] = channel
+	}
+	return channels, connectionChannels
+}
+
 func (a *App) refreshBotRuntimeAsync() {
 	if a.ctx == nil {
 		return
@@ -92,13 +116,33 @@ func (r *desktopBotRuntime) apply(parent context.Context, cfg *config.Config, wo
 	logger := slog.New(slog.NewTextHandler(os.Stderr, &slog.HandlerOptions{Level: slog.LevelInfo}))
 	ctx, cancel := context.WithCancel(parent)
 	modelName := botruntime.ModelName(cfg, "")
+	channels := botruntime.ChannelConfigs(cfg.Bot.Connections, true, true)
+	connectionChannels := botruntime.ConnectionChannelConfigs(cfg.Bot.Connections, true, true)
+	channels, connectionChannels = desktopBotChannelsWithLegacyQQ(cfg.Bot.QQ, channels, connectionChannels)
 	gwCfg := bot.GatewayConfig{
 		Model:              modelName,
 		ToolApprovalMode:   cfg.Bot.ToolApprovalMode,
 		MaxSteps:           cfg.Bot.MaxSteps,
+		QueueMode:          cfg.Bot.QueueMode,
+		QueueCap:           cfg.Bot.QueueCap,
+		QueueDrop:          cfg.Bot.QueueDrop,
+		PairingEnabled:     cfg.Bot.Pairing.Enabled,
+		PairingTTL:         time.Duration(cfg.Bot.Pairing.RequestTTLMinutes) * time.Minute,
+		PairingMaxPending:  cfg.Bot.Pairing.MaxPendingPerPlatform,
+		IgnoreSelfMessages: cfg.Bot.IgnoreSelfMessages,
+		SelfUserIDs: map[bot.Platform][]string{
+			bot.PlatformQQ:     cfg.Bot.SelfUserIDs.QQ,
+			bot.PlatformFeishu: cfg.Bot.SelfUserIDs.Feishu,
+			bot.PlatformWeixin: cfg.Bot.SelfUserIDs.Weixin,
+		},
+		ControlEnabled:     cfg.Bot.Control.Enabled,
+		ControlAddr:        cfg.Bot.Control.Addr,
+		ControlToken:       os.Getenv(strings.TrimSpace(cfg.Bot.Control.TokenEnv)),
 		WorkspaceRoot:      workspaceRoot,
-		Channels:           botruntime.ChannelConfigs(cfg.Bot.Connections, true, true),
-		ConnectionChannels: botruntime.ConnectionChannelConfigs(cfg.Bot.Connections, true, true),
+		Channels:           channels,
+		ConnectionChannels: connectionChannels,
+		Routes:             botruntime.RouteConfigs(cfg.Bot.Routes, true, true),
+		ConnectionAccess:   botruntime.ConnectionAccessConfigs(cfg),
 		Enabled:            plan.Enabled,
 		Allowlist: bot.AllowlistConfig{
 			Enabled:  cfg.Bot.Allowlist.Enabled,
@@ -108,6 +152,16 @@ func (r *desktopBotRuntime) apply(parent context.Context, cfg *config.Config, wo
 				bot.PlatformFeishu: cfg.Bot.Allowlist.FeishuUsers,
 				bot.PlatformWeixin: cfg.Bot.Allowlist.WeixinUsers,
 			},
+			Approvers: map[bot.Platform][]string{
+				bot.PlatformQQ:     cfg.Bot.Allowlist.QQApprovers,
+				bot.PlatformFeishu: cfg.Bot.Allowlist.FeishuApprovers,
+				bot.PlatformWeixin: cfg.Bot.Allowlist.WeixinApprovers,
+			},
+			Admins: map[bot.Platform][]string{
+				bot.PlatformQQ:     cfg.Bot.Allowlist.QQAdmins,
+				bot.PlatformFeishu: cfg.Bot.Allowlist.FeishuAdmins,
+				bot.PlatformWeixin: cfg.Bot.Allowlist.WeixinAdmins,
+			},
 			Groups: map[bot.Platform][]string{
 				bot.PlatformQQ:     cfg.Bot.Allowlist.QQGroups,
 				bot.PlatformFeishu: cfg.Bot.Allowlist.FeishuGroups,
@@ -206,8 +260,8 @@ func desktopBotRuntimePlan(cfg *config.Config) botRuntimePlan {
 	if !cfg.Bot.Enabled {
 		return botRuntimePlan{Status: "stopped", Message: "bot is disabled"}
 	}
-	if !cfg.Bot.Allowlist.AllowAll && (!cfg.Bot.Allowlist.Enabled || botruntime.AllowlistUserCount(cfg.Bot.Allowlist) == 0) {
-		return botRuntimePlan{Status: "blocked", Message: "bot requires an allowlist or allow_all=true"}
+	if !botruntime.BotConfigHasAccessControl(cfg.Bot) {
+		return botRuntimePlan{Status: "blocked", Message: "bot requires an allowlist, pairing, per-bot access, or allow_all=true"}
 	}
 	enabled, unknown := botruntime.EnabledPlatforms(cfg, nil)
 	if len(unknown) > 0 {

desktop/bot_runtime_app_test.go

@@ -39,6 +39,7 @@ func TestDesktopBotRuntimePlanBlocksWithoutAllowlist(t *testing.T) {
 	cfg := config.Default()
 	cfg.Bot.Enabled = true
 	cfg.Bot.Allowlist.Enabled = true
+	cfg.Bot.Pairing.Enabled = false
 	cfg.Bot.Connections = []config.BotConnectionConfig{
 		{ID: "feishu-lark", Provider: "feishu", Domain: "lark", Enabled: true},
 	}
@@ -49,6 +50,44 @@ func TestDesktopBotRuntimePlanBlocksWithoutAllowlist(t *testing.T) {
 	}
 }
 
+func TestDesktopBotRuntimePlanStartsWithPairing(t *testing.T) {
+	cfg := config.Default()
+	cfg.Bot.Enabled = true
+	cfg.Bot.Allowlist.Enabled = true
+	cfg.Bot.Pairing.Enabled = true
+	cfg.Bot.Connections = []config.BotConnectionConfig{
+		{ID: "feishu-lark", Provider: "feishu", Domain: "lark", Enabled: true},
+	}
+
+	plan := desktopBotRuntimePlan(cfg)
+	if !plan.Start {
+		t.Fatalf("plan = %+v, want start with pairing enabled", plan)
+	}
+}
+
+func TestDesktopBotChannelsWithLegacyQQConfig(t *testing.T) {
+	channels, connectionChannels := desktopBotChannelsWithLegacyQQ(config.QQBotConfig{
+		Model:            "qq-model",
+		ToolApprovalMode: "auto",
+		WorkspaceRoot:    "/tmp/qq-project",
+	}, nil, nil)
+
+	channel, ok := channels[bot.PlatformQQ]
+	if !ok {
+		t.Fatalf("platform QQ channel missing: %+v", channels)
+	}
+	if channel.Model != "qq-model" || channel.ToolApprovalMode != "auto" || channel.WorkspaceRoot != "/tmp/qq-project" {
+		t.Fatalf("platform channel = %+v, want QQ-specific runtime fields", channel)
+	}
+	connectionChannel, ok := connectionChannels["qq"]
+	if !ok {
+		t.Fatalf("connection QQ channel missing: %+v", connectionChannels)
+	}
+	if connectionChannel.Model != "qq-model" || connectionChannel.ToolApprovalMode != "auto" || connectionChannel.WorkspaceRoot != "/tmp/qq-project" {
+		t.Fatalf("connection channel = %+v, want QQ-specific runtime fields", connectionChannel)
+	}
+}
+
 func TestDesktopBotRuntimePlanStopsWhenBotDisabled(t *testing.T) {
 	cfg := config.Default()
 	cfg.Bot.Enabled = false

desktop/bound_array_contract_test.go

@@ -42,9 +42,9 @@ func TestBoundArrayPayloadsAreNonNilBeforeStartup(t *testing.T) {
 	if got := app.HooksSettings("global"); got.Hooks == nil || got.Events == nil {
 		t.Fatalf("HooksSettings(global) arrays = hooks:%v events:%v, want non-nil", got.Hooks, got.Events)
 	}
-	if got := app.Settings(); got.Providers == nil || got.OfficialProviders == nil || got.ProviderKinds == nil ||
+	if got := app.Settings(); got.Providers == nil || got.OfficialProviders == nil || got.ProviderPresets == nil || got.ProviderKinds == nil ||
 		got.Permissions.Allow == nil || got.Permissions.Ask == nil || got.Permissions.Deny == nil ||
-		got.Sandbox.AllowWrite == nil ||
+		got.Sandbox.AllowWrite == nil || got.Sandbox.EffectiveWriteRoots == nil ||
 		got.Bot.Allowlist.QQUsers == nil || got.Bot.Allowlist.FeishuUsers == nil || got.Bot.Allowlist.WeixinUsers == nil ||
 		got.Bot.Allowlist.QQGroups == nil || got.Bot.Allowlist.FeishuGroups == nil || got.Bot.Allowlist.WeixinGroups == nil {
 		t.Fatalf("Settings() contains nil array fields: %+v", got)

desktop/cmd/sign/main.go

@@ -166,12 +166,12 @@ func signFiles(files []string) error {
 // release tag used in download URLs (e.g. "desktop-v1.1.0").
 func genManifest(dir, version, tag string) error {
 	repo := os.Getenv("GITHUB_REPOSITORY")
-	if repo == "" {
-		repo = "esengine/reasonix"
+	if repo == "" || repo == "esengine/reasonix" {
+		repo = "esengine/DeepSeek-Reasonix"
 	}
 	m := update.Manifest{
 		Version:      version,
-		DownloadPage: fmt.Sprintf("https://github.com/%s/releases/latest", repo),
+		DownloadPage: "https://reasonix.io/#start",
 		Platforms:    map[string]update.Asset{},
 	}
 	entries, err := os.ReadDir(dir)

desktop/cmd/sign/main_test.go

@@ -85,6 +85,9 @@ func TestGenManifest(t *testing.T) {
 	if m.Version != "v1.2.0" {
 		t.Fatalf("version = %q, want v1.2.0", m.Version)
 	}
+	if m.DownloadPage != "https://reasonix.io/#start" {
+		t.Fatalf("download_page = %q, want official install page", m.DownloadPage)
+	}
 	if len(m.Platforms) != 5 {
 		t.Fatalf("want 5 platforms, got %d: %v", len(m.Platforms), m.Platforms)
 	}
@@ -92,7 +95,7 @@ func TestGenManifest(t *testing.T) {
 	if !ok {
 		t.Fatal("windows-amd64 missing")
 	}
-	wantURL := "https://github.com/esengine/reasonix/releases/download/desktop-v1.2.0/Reasonix-windows-amd64-installer.exe"
+	wantURL := "https://github.com/esengine/DeepSeek-Reasonix/releases/download/desktop-v1.2.0/Reasonix-windows-amd64-installer.exe"
 	if win.URL != wantURL {
 		t.Fatalf("windows url = %q, want %q", win.URL, wantURL)
 	}

desktop/frontend/package.json

@@ -10,9 +10,9 @@
     "check:css": "node scripts/check-css-syntax.mjs src/styles.css && node scripts/check-z-index-tokens.mjs src/styles.css",
     "test:todo-visibility": "node scripts/test-todo-visibility.mjs",
     "typecheck": "tsc --noEmit",
-    "test": "node scripts/test-todo-visibility.mjs && tsx src/__tests__/anchored-popover-scroll.test.tsx && tsx src/__tests__/math-golden.test.ts && tsx src/__tests__/mermaid-rendering.test.tsx && tsx src/__tests__/text-size.test.ts && tsx src/__tests__/font-availability.test.ts && tsx src/__tests__/theme-auto-background.test.ts && tsx src/__tests__/typography-overflow-contract.test.ts && tsx src/__tests__/ask-card-layout.test.ts && tsx src/__tests__/approval-modal-file-reference.test.tsx && tsx src/__tests__/provider-model-refresh.test.ts && tsx src/__tests__/composer-profile.test.ts && tsx src/__tests__/composer-history.test.ts && tsx src/__tests__/session-title-contract.test.ts && tsx src/__tests__/composer-keyboard.test.ts && tsx src/__tests__/composer-goal-toggle.test.tsx && tsx src/__tests__/composer-session-draft.test.tsx && tsx src/__tests__/composer-image-capability.test.tsx && tsx src/__tests__/use-controller-meta.test.ts && tsx src/__tests__/use-controller-cancel-reconcile.test.tsx && tsx src/__tests__/new-session-load-race.test.tsx && tsx src/__tests__/tab-switch-hydration.test.tsx && tsx src/__tests__/workspace-layout.test.ts && tsx src/__tests__/workspace-changes-errors.test.tsx && tsx src/__tests__/workspace-split.test.ts && tsx src/__tests__/workspace-preview-css.test.ts && tsx src/__tests__/resize-drag.test.ts && tsx src/__tests__/tool-approval-mode.test.ts && tsx src/__tests__/heartbeat-next-run.test.ts && tsx src/__tests__/keyboard-shortcuts.test.ts && tsx src/__tests__/send-failed.test.ts && tsx src/__tests__/edit-replay.test.ts && tsx src/__tests__/history-tool-status.test.ts && tsx src/__tests__/message-selection-copy.test.ts && tsx src/__tests__/code-block-copy-position.test.tsx && tsx src/__tests__/reasoning-display.test.ts && tsx src/__tests__/attachment-display.test.ts && tsx src/__tests__/crash-reporting.test.ts && tsx src/__tests__/bridge-drag-rejection.test.ts && tsx src/__tests__/startup-settings-contract.test.ts && tsx src/__tests__/bundle-contract.test.ts && tsx src/__tests__/command-palette-css.test.ts && tsx src/__tests__/command-palette-interactions.test.tsx && tsx src/__tests__/context-panel-breakdown.test.ts && tsx src/__tests__/statusbar-workspace.test.tsx && tsx src/__tests__/settings-refresh-snapshot.test.tsx && tsx src/__tests__/capabilities-panel-actions.test.ts && tsx src/__tests__/tool-data-archive.test.ts && tsx src/__tests__/tool-subject.test.ts && tsx src/__tests__/diff-rendering.test.ts && tsx src/__tests__/render-optimization.test.ts && tsx src/__tests__/transcript-grouping.test.ts && tsx src/__tests__/project-tree-runtime.test.ts && tsx src/__tests__/app-chrome-tabs.test.ts && tsx src/__tests__/open-topic-coalescing.test.tsx && tsx src/__tests__/memory-compiler-display.test.ts && tsx src/__tests__/memory-citation-visibility.test.ts",
+    "test": "node scripts/test-todo-visibility.mjs && tsx src/__tests__/anchored-popover-scroll.test.tsx && tsx src/__tests__/math-golden.test.ts && tsx src/__tests__/mermaid-rendering.test.tsx && tsx src/__tests__/text-size.test.ts && tsx src/__tests__/font-availability.test.ts && tsx src/__tests__/theme-auto-background.test.ts && tsx src/__tests__/typography-overflow-contract.test.ts && tsx src/__tests__/ask-card-layout.test.ts && tsx src/__tests__/approval-modal-file-reference.test.tsx && tsx src/__tests__/provider-model-refresh.test.ts && tsx src/__tests__/composer-profile.test.ts && tsx src/__tests__/composer-history.test.ts && tsx src/__tests__/session-title-contract.test.ts && tsx src/__tests__/composer-keyboard.test.ts && tsx src/__tests__/composer-goal-toggle.test.tsx && tsx src/__tests__/composer-session-draft.test.tsx && tsx src/__tests__/composer-image-capability.test.tsx && tsx src/__tests__/use-controller-meta.test.ts && tsx src/__tests__/use-controller-cancel-reconcile.test.tsx && tsx src/__tests__/new-session-load-race.test.tsx && tsx src/__tests__/tab-switch-hydration.test.tsx && tsx src/__tests__/workspace-layout.test.ts && tsx src/__tests__/workspace-changes-errors.test.tsx && tsx src/__tests__/workspace-split.test.ts && tsx src/__tests__/workspace-preview-css.test.ts && tsx src/__tests__/resize-drag.test.ts && tsx src/__tests__/tool-approval-mode.test.ts && tsx src/__tests__/heartbeat-next-run.test.ts && tsx src/__tests__/keyboard-shortcuts.test.ts && tsx src/__tests__/send-failed.test.ts && tsx src/__tests__/edit-replay.test.ts && tsx src/__tests__/message-pasted-blocks.test.ts && tsx src/__tests__/history-tool-status.test.ts && tsx src/__tests__/message-selection-copy.test.ts && tsx src/__tests__/code-block-copy-position.test.tsx && tsx src/__tests__/reasoning-display.test.ts && tsx src/__tests__/attachment-display.test.ts && tsx src/__tests__/crash-reporting.test.ts && tsx src/__tests__/bridge-drag-rejection.test.ts && tsx src/__tests__/startup-settings-contract.test.ts && tsx src/__tests__/bundle-contract.test.ts && tsx src/__tests__/command-palette-css.test.ts && tsx src/__tests__/command-palette-interactions.test.tsx && tsx src/__tests__/context-panel-breakdown.test.ts && tsx src/__tests__/statusbar-workspace.test.tsx && tsx src/__tests__/settings-refresh-snapshot.test.tsx && tsx src/__tests__/capabilities-panel-actions.test.ts && tsx src/__tests__/tool-data-archive.test.ts && tsx src/__tests__/tool-subject.test.ts && tsx src/__tests__/diff-rendering.test.ts && tsx src/__tests__/render-optimization.test.ts && tsx src/__tests__/transcript-grouping.test.ts && tsx src/__tests__/project-tree-runtime.test.ts && tsx src/__tests__/app-chrome-tabs.test.ts && tsx src/__tests__/open-topic-coalescing.test.tsx && tsx src/__tests__/memory-compiler-display.test.ts && tsx src/__tests__/memory-citation-visibility.test.ts",
     "test:typecheck": "tsc --noEmit -p tsconfig.test.json",
-    "test:all": "tsc --noEmit -p tsconfig.test.json && node scripts/test-todo-visibility.mjs && tsx src/__tests__/anchored-popover-scroll.test.tsx && tsx src/__tests__/math-golden.test.ts && tsx src/__tests__/mermaid-rendering.test.tsx && tsx src/__tests__/text-size.test.ts && tsx src/__tests__/font-availability.test.ts && tsx src/__tests__/theme-auto-background.test.ts && tsx src/__tests__/typography-overflow-contract.test.ts && tsx src/__tests__/ask-card-layout.test.ts && tsx src/__tests__/approval-modal-file-reference.test.tsx && tsx src/__tests__/provider-model-refresh.test.ts && tsx src/__tests__/composer-profile.test.ts && tsx src/__tests__/composer-history.test.ts && tsx src/__tests__/session-title-contract.test.ts && tsx src/__tests__/composer-keyboard.test.ts && tsx src/__tests__/composer-goal-toggle.test.tsx && tsx src/__tests__/composer-session-draft.test.tsx && tsx src/__tests__/composer-image-capability.test.tsx && tsx src/__tests__/use-controller-meta.test.ts && tsx src/__tests__/use-controller-cancel-reconcile.test.tsx && tsx src/__tests__/new-session-load-race.test.tsx && tsx src/__tests__/tab-switch-hydration.test.tsx && tsx src/__tests__/workspace-layout.test.ts && tsx src/__tests__/workspace-changes-errors.test.tsx && tsx src/__tests__/workspace-split.test.ts && tsx src/__tests__/workspace-preview-css.test.ts && tsx src/__tests__/resize-drag.test.ts && tsx src/__tests__/tool-approval-mode.test.ts && tsx src/__tests__/heartbeat-next-run.test.ts && tsx src/__tests__/keyboard-shortcuts.test.ts && tsx src/__tests__/send-failed.test.ts && tsx src/__tests__/edit-replay.test.ts && tsx src/__tests__/history-tool-status.test.ts && tsx src/__tests__/message-selection-copy.test.ts && tsx src/__tests__/code-block-copy-position.test.tsx && tsx src/__tests__/reasoning-display.test.ts && tsx src/__tests__/attachment-display.test.ts && tsx src/__tests__/crash-reporting.test.ts && tsx src/__tests__/bridge-drag-rejection.test.ts && tsx src/__tests__/startup-settings-contract.test.ts && tsx src/__tests__/bundle-contract.test.ts && tsx src/__tests__/command-palette-css.test.ts && tsx src/__tests__/command-palette-interactions.test.tsx && tsx src/__tests__/context-panel-breakdown.test.ts && tsx src/__tests__/statusbar-workspace.test.tsx && tsx src/__tests__/settings-refresh-snapshot.test.tsx && tsx src/__tests__/capabilities-panel-actions.test.ts && tsx src/__tests__/tool-data-archive.test.ts && tsx src/__tests__/tool-subject.test.ts && tsx src/__tests__/diff-rendering.test.ts && tsx src/__tests__/render-optimization.test.ts && tsx src/__tests__/transcript-grouping.test.ts && tsx src/__tests__/project-tree-runtime.test.ts && tsx src/__tests__/app-chrome-tabs.test.ts && tsx src/__tests__/open-topic-coalescing.test.tsx"
+    "test:all": "tsc --noEmit -p tsconfig.test.json && node scripts/test-todo-visibility.mjs && tsx src/__tests__/anchored-popover-scroll.test.tsx && tsx src/__tests__/math-golden.test.ts && tsx src/__tests__/mermaid-rendering.test.tsx && tsx src/__tests__/text-size.test.ts && tsx src/__tests__/font-availability.test.ts && tsx src/__tests__/theme-auto-background.test.ts && tsx src/__tests__/typography-overflow-contract.test.ts && tsx src/__tests__/ask-card-layout.test.ts && tsx src/__tests__/approval-modal-file-reference.test.tsx && tsx src/__tests__/provider-model-refresh.test.ts && tsx src/__tests__/composer-profile.test.ts && tsx src/__tests__/composer-history.test.ts && tsx src/__tests__/session-title-contract.test.ts && tsx src/__tests__/composer-keyboard.test.ts && tsx src/__tests__/composer-goal-toggle.test.tsx && tsx src/__tests__/composer-session-draft.test.tsx && tsx src/__tests__/composer-image-capability.test.tsx && tsx src/__tests__/use-controller-meta.test.ts && tsx src/__tests__/use-controller-cancel-reconcile.test.tsx && tsx src/__tests__/new-session-load-race.test.tsx && tsx src/__tests__/tab-switch-hydration.test.tsx && tsx src/__tests__/workspace-layout.test.ts && tsx src/__tests__/workspace-changes-errors.test.tsx && tsx src/__tests__/workspace-split.test.ts && tsx src/__tests__/workspace-preview-css.test.ts && tsx src/__tests__/resize-drag.test.ts && tsx src/__tests__/tool-approval-mode.test.ts && tsx src/__tests__/heartbeat-next-run.test.ts && tsx src/__tests__/keyboard-shortcuts.test.ts && tsx src/__tests__/send-failed.test.ts && tsx src/__tests__/edit-replay.test.ts && tsx src/__tests__/message-pasted-blocks.test.ts && tsx src/__tests__/history-tool-status.test.ts && tsx src/__tests__/message-selection-copy.test.ts && tsx src/__tests__/code-block-copy-position.test.tsx && tsx src/__tests__/reasoning-display.test.ts && tsx src/__tests__/attachment-display.test.ts && tsx src/__tests__/crash-reporting.test.ts && tsx src/__tests__/bridge-drag-rejection.test.ts && tsx src/__tests__/startup-settings-contract.test.ts && tsx src/__tests__/bundle-contract.test.ts && tsx src/__tests__/command-palette-css.test.ts && tsx src/__tests__/command-palette-interactions.test.tsx && tsx src/__tests__/context-panel-breakdown.test.ts && tsx src/__tests__/statusbar-workspace.test.tsx && tsx src/__tests__/settings-refresh-snapshot.test.tsx && tsx src/__tests__/capabilities-panel-actions.test.ts && tsx src/__tests__/tool-data-archive.test.ts && tsx src/__tests__/tool-subject.test.ts && tsx src/__tests__/diff-rendering.test.ts && tsx src/__tests__/render-optimization.test.ts && tsx src/__tests__/transcript-grouping.test.ts && tsx src/__tests__/project-tree-runtime.test.ts && tsx src/__tests__/app-chrome-tabs.test.ts && tsx src/__tests__/open-topic-coalescing.test.tsx"
   },
   "dependencies": {
     "@gsap/react": "^2.1.2",

desktop/frontend/src/App.tsx

@@ -38,7 +38,7 @@ import { useToast } from "./lib/toast";
 import { asArray } from "./lib/array";
 import { clearLegacyLangPref, normalizeLangPref, readLegacyLangPref, useI18n, useT, type Translator } from "./lib/i18n";
 import { useController, type Item, type LiveStream } from "./lib/useController";
-import { app, onEvent, onProjectTreeChanged } from "./lib/bridge";
+import { app, onEvent, onProjectTreeChanged, onSessionRecovered, onSessionRecoveryFailed } from "./lib/bridge";
 import { generativeMusic, isGenerativeMusicEnabled } from "./lib/generative-music";
 import { playSuccessChime } from "./lib/sound";
 import { Transcript } from "./components/Transcript";
@@ -1030,6 +1030,9 @@ export default function App() {
   const rightDockMode = useLayoutStore((s) => s.rightDockMode);
   const setRightDockMode = useLayoutStore((s) => s.setRightDockMode);
   const [dockRefreshKey, setDockRefreshKey] = useState(0);
+  const [fileRefRefreshKey, setFileRefRefreshKey] = useState(0);
+  const refreshComposerFileRefs = useCallback(() => setFileRefRefreshKey((value) => value + 1), []);
+  const composerFileRefRefreshKey = `${dockRefreshKey}:${fileRefRefreshKey}`;
   const [projectRevision, setProjectRevision] = useState(0);
   const [activeTopicTurns, setActiveTopicTurns] = useState<number | undefined>(undefined);
   const [composerInsertRequest, setComposerInsertRequest] = useState<ComposerInsertRequest | null>(null);
@@ -2588,6 +2591,37 @@ export default function App() {
     enqueueNavigation({ kind: "blank", scope, workspaceRoot: scope === "project" ? workspaceRoot : "" }),
   [enqueueNavigation]);
 
+  useEffect(() => {
+    return onSessionRecovered((event) => {
+      const scope = event.scope === "project" ? "project" : "global";
+      const workspaceRoot = scope === "project" ? event.workspaceRoot || "" : "";
+      setProjectRevision((value) => value + 1);
+      void refreshTabMetas();
+      showToast(t("recovery.toast", { title: event.topicTitle || t("recovery.branch") }), "warn", event.topicId
+        ? {
+            actionLabel: t("recovery.open"),
+            durationMs: 9000,
+            onAction: () => {
+              void enqueueNavigation({
+                kind: "topic",
+                scope,
+                workspaceRoot,
+                topicId: event.topicId || "",
+                sessionPath: event.recoveryPath || "",
+              });
+            },
+          }
+        : { durationMs: 9000 });
+    });
+  }, [enqueueNavigation, refreshTabMetas, showToast, t]);
+
+  useEffect(() => {
+    return onSessionRecoveryFailed((event) => {
+      const key = event.reason === "lease_held" ? "recovery.failedLease" : "recovery.failedUnavailable";
+      showToast(t(key), "error", { durationMs: 9000 });
+    });
+  }, [showToast, t]);
+
   const handleNewTab = useCallback(async () => {
     closeTransientOverlays();
     setSidebarImDetailConnectionId("");
@@ -2848,6 +2882,13 @@ export default function App() {
     : [topicbarWorkspacePath || topicbarWorkspaceLabel, topicbarImSourceLabel].filter(Boolean).join(" · ");
   const topicbarCanRename = !sidebarImDetailConnection && Boolean(activeTab?.topicId);
   const topicbarTitleEditSize = Math.min(56, Math.max(4, topicTitleDraft.length || topicbarTitle.length || 1));
+  const recoveryBannerTitle = activeTab?.recovered
+    ? [
+        activeTab.recoveryReason ? t("recovery.reason", { reason: activeTab.recoveryReason }) : "",
+        activeTab.recoveryDigest ? t("recovery.digest", { digest: activeTab.recoveryDigest.slice(0, 12) }) : "",
+        activeTab.recoveryParentId ? t("recovery.parent", { parent: activeTab.recoveryParentId }) : "",
+      ].filter(Boolean).join(" · ")
+    : "";
   const sidebarWorkbench = desktopLayoutStyle === "workbench";
   const windowsFramelessChrome = desktopPlatform === "windows";
   const handleWindowsTitlebarDoubleClick = useCallback((event: ReactMouseEvent<HTMLDivElement>) => {
@@ -3399,6 +3440,13 @@ export default function App() {
             <div className="banner banner--error">{t("topbar.startupError", { msg: state.meta.startupErr })}</div>
           )}
 
+          {activeTab?.recovered && !sidebarImDetailConnection && (
+            <div className="banner banner--recovery" title={recoveryBannerTitle || undefined}>
+              <span className="banner__badge">{t("recovery.branch")}</span>
+              <span>{t("recovery.banner")}</span>
+            </div>
+          )}
+
           <UpdateBanner enabled={startupUpdateChecksEnabled === true} />
 
           <main className="main">
@@ -3538,6 +3586,7 @@ export default function App() {
               retry={state.retry}
               transientDismissSignal={transientOverlayDismissSignal}
               sessionKey={composerSessionKey}
+              fileRefRefreshKey={composerFileRefRefreshKey}
               guidanceConsumedKey={latestGuidanceConsumed?.key}
               guidanceConsumedText={latestGuidanceConsumed?.text}
               guidanceQueuePreviewItems={guidanceQueueMockItems}
@@ -3656,6 +3705,7 @@ export default function App() {
                   onPreviewModeChange={handleWorkspacePreviewModeChange}
                   onAddToChat={addWorkspaceTextToComposer}
                   onRequestPanelWidth={ensureWorkspacePanelWidth}
+                  onFileTreeRefresh={refreshComposerFileRefs}
                   refreshKey={dockRefreshKey}
                   initialViewMode={rightDockMode === "changed" ? "changed" : "files"}
                   showViewTabs={false}

desktop/frontend/src/__tests__/app-chrome-tabs.test.ts

@@ -69,6 +69,12 @@ ok(
   "AppChrome markup includes classic tab strip containers",
 );
 
+ok(
+  /const titlebarDragRail = darwinChrome \|\| platform === "windows";/.test(appChromeSource) &&
+    /\{titlebarDragRail && <span className="app-chrome__drag-rail"/.test(appChromeSource),
+  "AppChrome exposes the classic drag rail on macOS and Windows",
+);
+
 ok(
   /const WORKSPACE_PANEL_DEFAULT_OPEN = false;/.test(layoutStoreSource) &&
     /workspacePanelOpen:\s*WORKSPACE_PANEL_DEFAULT_OPEN/.test(layoutStoreSource),
@@ -276,6 +282,13 @@ for (const selector of [
   );
 }
 
+ok(
+  finalDeclaration(".app--windows-frameless:not(.app--workbench):not(.app--creation) .app-chrome--native-tabs .app-chrome__drag-rail", "--wails-draggable") === "drag" &&
+    finalDeclaration(".app--windows-frameless:not(.app--workbench):not(.app--creation) .app-chrome--native-tabs .app-chrome__drag-rail", "right")?.includes("--windows-window-controls-safe") &&
+    finalDeclaration(".app--windows .app-chrome--native-tabs .tabbar", "--wails-draggable") === "no-drag",
+  "Windows classic chrome keeps a draggable rail while tabs remain clickable",
+);
+
 for (const selector of [
   ".layout--workbench-chrome-hidden",
   ":root[data-theme-style] .layout--workbench-chrome-hidden",

desktop/frontend/src/__tests__/approval-modal-file-reference.test.tsx

@@ -4,6 +4,7 @@ import { JSDOM } from "jsdom";
 import React from "react";
 import { act } from "react";
 import { createRoot } from "react-dom/client";
+import gsap from "gsap";
 import { ApprovalModal } from "../components/ApprovalModal";
 import { LocaleProvider } from "../lib/i18n";
 import type { AppBindings } from "../lib/bridge";
@@ -12,6 +13,17 @@ import type { WireApproval } from "../lib/types";
 let passed = 0;
 let failed = 0;
 
+type GsapToOptions = { onComplete?: () => void };
+const gsapForTests = (typeof gsap.to === "function" ? gsap : (gsap as unknown as { default?: typeof gsap }).default) as unknown as {
+  to?: (target: unknown, vars: GsapToOptions) => unknown;
+};
+if (typeof gsapForTests.to === "function") {
+  gsapForTests.to = (_target: unknown, vars: GsapToOptions) => {
+    vars.onComplete?.();
+    return {};
+  };
+}
+
 function ok(value: boolean, label: string) {
   if (value) {
     process.stdout.write(`  PASS  ${label}\n`);
@@ -225,5 +237,48 @@ console.log("\napproval modal file references");
   dom.window.close();
 }
 
+{
+  const dom = installDom();
+  mockApp({
+    ListDir: async () => [],
+    SearchFileRefs: async () => [],
+  });
+  const answers: Array<[boolean, boolean, boolean]> = [];
+  const { root } = await renderApproval({
+    approval: {
+      id: "memory-approval",
+      tool: "remember",
+      subject: "Save/update memory \"prefers-vitest\": Preferred test framework",
+    },
+    onAnswer: (allow, session, persist) => answers.push([allow, session, persist]),
+  });
+
+  const text = document.body.textContent ?? "";
+  ok(text.includes("Allow once"), "fresh-human approval shows allow once");
+  ok(text.includes("Deny"), "fresh-human approval shows deny");
+  ok(!text.includes("Allow for session"), "fresh-human approval hides session grant");
+  ok(!text.includes("Always allow"), "fresh-human approval hides persistent grant");
+  eq(
+    Array.from(document.querySelectorAll(".prompt-shelf__actions button")).map((button) => button.textContent).join("|"),
+    "1Allow once|4Deny",
+    "fresh-human approval keeps allow/deny shortcut keys",
+  );
+
+  const allowButton = Array.from(document.querySelectorAll("button")).find((button) => button.textContent?.includes("Allow once")) as HTMLButtonElement | undefined;
+  if (!allowButton) throw new Error("allow once button did not render");
+
+  await act(async () => {
+    allowButton.click();
+    await flushTimers();
+  });
+
+  eq(JSON.stringify(answers), JSON.stringify([[true, false, false]]), "fresh-human approval allows only once");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
 console.log(`\n${passed} passed, ${failed} failed, ${passed + failed} total`);
 if (failed > 0) process.exit(1);

desktop/frontend/src/__tests__/bundle-contract.test.ts

@@ -43,7 +43,7 @@ ok(
   "App loads secondary drawers on demand",
 );
 ok(
-  !/import\s+\{[^}]*\b(?:MCPServersSettingsPage|SkillsSettingsPage)\b[^}]*\}\s+from\s+["']\.\/CapabilitiesPanel["']/.test(settingsSource) &&
+  !/import\s+\{[^}]*\b(?:MCPServersSettingsPage|SkillsSettingsPage|PluginsSettingsPage)\b[^}]*\}\s+from\s+["']\.\/CapabilitiesPanel["']/.test(settingsSource) &&
     !/import\s+\{[^}]*\bMemorySettingsPage\b[^}]*\}\s+from\s+["']\.\/MemoryPanel["']/.test(settingsSource),
   "SettingsPanel keeps secondary settings pages out of the first settings chunk",
 );

desktop/frontend/src/__tests__/capabilities-panel-actions.test.ts

@@ -2,11 +2,11 @@ import { JSDOM } from "jsdom";
 import React from "react";
 import { act } from "react";
 import { createRoot } from "react-dom/client";
-import { MCPServersSettingsPage } from "../components/CapabilitiesPanel";
+import { MCPServersSettingsPage, PluginsSettingsPage } from "../components/CapabilitiesPanel";
 import type { AppBindings } from "../lib/bridge";
 import { LocaleProvider } from "../lib/i18n";
 import { mcpServerLifecycleActions, mcpServerRetryableFromAvailableList } from "../lib/mcpServerLifecycle";
-import type { Meta, ServerView, TabMeta } from "../lib/types";
+import type { Meta, PluginInstallOptions, PluginView, ServerView, TabMeta } from "../lib/types";
 
 function ok(value: unknown, message: string) {
   if (!value) throw new Error(message);
@@ -80,6 +80,7 @@ function installDom() {
   globalThis.Node = dom.window.Node;
   globalThis.HTMLElement = dom.window.HTMLElement;
   globalThis.HTMLButtonElement = dom.window.HTMLButtonElement;
+  globalThis.HTMLInputElement = dom.window.HTMLInputElement;
   globalThis.Event = dom.window.Event;
   globalThis.KeyboardEvent = dom.window.KeyboardEvent;
   globalThis.MouseEvent = dom.window.MouseEvent;
@@ -106,6 +107,16 @@ function findButton(label: string): HTMLButtonElement | undefined {
   return Array.from(document.querySelectorAll("button")).find((button) => button.textContent?.trim() === label) as HTMLButtonElement | undefined;
 }
 
+function setInputValue(input: HTMLInputElement, value: string) {
+  const win = input.ownerDocument.defaultView;
+  const setter = Object.getOwnPropertyDescriptor((win?.HTMLInputElement ?? HTMLInputElement).prototype, "value")?.set;
+  setter?.call(input, value);
+  const eventCtor = win?.Event ?? Event;
+  const inputEventCtor = win?.InputEvent ?? eventCtor;
+  input.dispatchEvent(new inputEventCtor("input", { bubbles: true, data: value, inputType: "insertText" } as InputEventInit));
+  input.dispatchEvent(new eventCtor("change", { bubbles: true }));
+}
+
 console.log("capabilities panel MCP actions");
 
 {
@@ -233,3 +244,221 @@ console.log("capabilities panel MCP actions");
   });
   dom.window.close();
 }
+
+console.log("capabilities panel plugin actions");
+
+{
+  const dom = installDom();
+  const rootEl = document.getElementById("root");
+  if (!rootEl) throw new Error("missing root");
+  const root = createRoot(rootEl);
+  const meta: Meta = { label: "test", ready: true, eventChannel: "plugin-channel", cwd: "/tmp/reasonix-test", workspaceRoot: "/tmp/reasonix-test" };
+  const tabs: TabMeta[] = [{
+    id: "tab-plugin",
+    scope: "project",
+    workspaceRoot: "/tmp/reasonix-test",
+    workspaceName: "reasonix-test",
+    topicId: "topic-plugin",
+    topicTitle: "Plugins",
+    label: "Plugins",
+    ready: true,
+    running: false,
+    mode: "normal",
+    toolApprovalMode: "auto",
+    active: true,
+    cwd: "/tmp/reasonix-test",
+  }];
+  let planCalls = 0;
+  let installCalls = 0;
+  let toggleCalls = 0;
+  let updateCalls = 0;
+  let doctorCalls = 0;
+  let removeCalls = 0;
+  let pickFolderCalls = 0;
+  const plannedSources: string[] = [];
+  const installedSources: string[] = [];
+  let plugins: PluginView[] = [{
+    name: "superpowers",
+    version: "0.1.0",
+    description: "Shared agent skills and hooks.",
+    source: "git:github.com/obra/superpowers",
+    root: "~/.reasonix/plugins/superpowers",
+    manifestKind: "reasonix",
+    enabled: true,
+    skills: 2,
+    hooks: 1,
+    mcpServers: 0,
+  }];
+  window.go = {
+    main: {
+      App: {
+        Meta: async () => meta,
+        ListTabs: async () => tabs,
+        Plugins: async () => plugins.map((plugin) => ({ ...plugin, warnings: [...(plugin.warnings ?? [])] })),
+        PlanPluginInstall: async (source: string, options: PluginInstallOptions) => {
+          planCalls += 1;
+          plannedSources.push(source);
+          ok(options.dryRun === true, "plugin preview asks for dry-run planning");
+          return JSON.stringify({
+            ok: true,
+            status: "planned",
+            name: "superpowers",
+            actions: [{ kind: "plugin", action: "install_plugin_package", name: "superpowers", source, status: "planned" }],
+          });
+        },
+        InstallPlugin: async (source: string, _options: PluginInstallOptions) => {
+          installCalls += 1;
+          installedSources.push(source);
+          const next: PluginView = {
+            name: "superpowers",
+            version: "0.1.1",
+            description: "Shared agent skills and hooks.",
+            source,
+            root: "~/.reasonix/plugins/superpowers",
+            manifestKind: "reasonix",
+            enabled: true,
+            skills: 3,
+            hooks: 1,
+            mcpServers: 1,
+            skillDetails: [{ name: "plan", description: "Plan work before implementation.", invocation: "/plan", runAs: "inline" }],
+            hookDetails: [{ event: "SessionStart", contextFile: "CLAUDE.md", description: "Load startup context." }],
+            mcpServerDetails: [{ name: "context", transport: "stdio", command: "node server.js" }],
+          };
+          plugins = plugins.filter((plugin) => plugin.name !== next.name).concat(next);
+          return JSON.stringify({ ok: true, status: "done", actions: [{ action: "install_plugin_package", name: next.name, status: "done" }] });
+        },
+        SetPluginEnabled: async (name: string, enabled: boolean) => {
+          toggleCalls += 1;
+          plugins = plugins.map((plugin) => plugin.name === name ? { ...plugin, enabled } : plugin);
+        },
+        UpdatePlugin: async (name: string) => {
+          updateCalls += 1;
+          plugins = plugins.map((plugin) => plugin.name === name ? { ...plugin, version: "0.1.2" } : plugin);
+          return JSON.stringify({ ok: true, status: "done", name });
+        },
+        PluginDoctor: async (name: string) => {
+          doctorCalls += 1;
+          return { ...(plugins.find((plugin) => plugin.name === name) ?? plugins[0]), warnings: ["manifest exports no MCP auth metadata"] };
+        },
+        RemovePlugin: async (name: string) => {
+          removeCalls += 1;
+          plugins = plugins.filter((plugin) => plugin.name !== name);
+        },
+        PickPluginFolder: async () => {
+          pickFolderCalls += 1;
+          return "/tmp/superpowers-plugin";
+        },
+      } as Partial<AppBindings> as AppBindings,
+    },
+  };
+
+  await act(async () => {
+    root.render(React.createElement(LocaleProvider, null, React.createElement(PluginsSettingsPage)));
+    await flush();
+  });
+  await waitFor("superpowers plugin row", () => Boolean(document.querySelector(".cap-row__name")?.textContent?.includes("superpowers")));
+  ok(Boolean(document.querySelector(".cap-plugin-form-grid .cap-plugin-fields--local")), "local plugin install mode uses the shared form grid");
+  const localOptionTexts = Array.from(document.querySelectorAll(".cap-plugin-installer__options > .cap-plugin-option-block"))
+    .map((option) => option.textContent ?? "");
+  ok(localOptionTexts[0]?.includes("Overwrite same-name plugin"), "local install mode shows overwrite before link mode");
+  ok(localOptionTexts[1]?.includes("Developer mode: link source folder"), "local install mode shows link mode after overwrite");
+
+  const chooseFolder = findButton("Choose plugin folder");
+  if (!chooseFolder) throw new Error("missing plugin folder picker button");
+  await act(async () => {
+    chooseFolder.click();
+    await flush();
+  });
+  await waitFor("picked plugin folder source", () => document.body.textContent?.includes("/tmp/superpowers-plugin") ?? false);
+  ok(pickFolderCalls === 1, "clicking Choose folder invokes the plugin folder picker once");
+
+  const gitMode = findButton("Git repository");
+  if (!gitMode) throw new Error("missing Git repository install mode");
+  await act(async () => {
+    gitMode.click();
+    await flush();
+  });
+  ok(Boolean(document.querySelector(".cap-plugin-form-grid .cap-plugin-fields--git")), "Git plugin install mode uses the shared form grid");
+  const sourceInput = document.querySelector<HTMLInputElement>('input[aria-label="Git repository URL"]');
+  if (!sourceInput) throw new Error("missing plugin git source input");
+  await act(async () => {
+    setInputValue(sourceInput, "git:github.com/obra/superpowers");
+    await flush();
+  });
+  await waitFor("plugin preview enabled", () => findButton("Preview")?.disabled === false);
+
+  const preview = findButton("Preview");
+  if (!preview) throw new Error("missing plugin preview button");
+  await act(async () => {
+    preview.click();
+    await flush();
+  });
+  await waitFor("plugin install plan", () => document.body.textContent?.includes("install_plugin_package") ?? false);
+  ok(planCalls === 1, "clicking Preview invokes plugin install planning once");
+  ok(plannedSources[0] === "git:github.com/obra/superpowers", "plugin preview receives the entered Git source");
+
+  const install = findButton("Install plugin");
+  if (!install) throw new Error("missing plugin install button");
+  await act(async () => {
+    install.click();
+    await flush();
+  });
+  await waitFor("plugin install result", () => installCalls === 1 && plugins[0]?.version === "0.1.1");
+  ok(installedSources[0] === "git:github.com/obra/superpowers", "plugin install receives the entered Git source");
+
+  const disclosure = document.querySelector<HTMLButtonElement>(".cap-plugin-entry .cap-disclosure");
+  if (!disclosure) throw new Error("missing plugin disclosure");
+  await act(async () => {
+    disclosure.click();
+    await flush();
+  });
+  await waitFor("plugin update action", () => Boolean(findButton("Update")));
+  ok(document.body.textContent?.includes("How to use") ?? false, "expanded plugin details explain how to use the plugin");
+  ok(document.body.textContent?.includes("/plan") ?? false, "expanded plugin details list exported skill invocations");
+  ok(document.body.textContent?.includes("SessionStart") ?? false, "expanded plugin details list exported hooks");
+  ok(document.body.textContent?.includes("context") ?? false, "expanded plugin details list exported MCP servers");
+
+  const update = findButton("Update");
+  if (!update) throw new Error("missing plugin update button");
+  await act(async () => {
+    update.click();
+    await flush();
+  });
+  await waitFor("plugin update call", () => updateCalls === 1 && plugins[0]?.version === "0.1.2");
+
+  const doctor = findButton("Doctor");
+  if (!doctor) throw new Error("missing plugin doctor button");
+  await act(async () => {
+    doctor.click();
+    await flush();
+  });
+  await waitFor("plugin diagnostic warning", () => document.body.textContent?.includes("manifest exports no MCP auth metadata") ?? false);
+  ok(doctorCalls === 1, "clicking Doctor invokes plugin diagnostics once");
+
+  const toggle = document.querySelector<HTMLInputElement>(".cap-plugin-entry .cap-switch input");
+  if (!toggle) throw new Error("missing plugin enable toggle");
+  await act(async () => {
+    toggle.click();
+    await flush();
+  });
+  await waitFor("plugin disabled", () => toggleCalls === 1 && plugins[0]?.enabled === false);
+
+  const remove = findButton("Remove plugin");
+  if (!remove) throw new Error("missing plugin remove button");
+  await act(async () => {
+    remove.click();
+    await flush();
+  });
+  const confirmRemove = findButton("Confirm remove");
+  if (!confirmRemove) throw new Error("missing plugin confirm remove button");
+  await act(async () => {
+    confirmRemove.click();
+    await flush();
+  });
+  await waitFor("plugin removed", () => removeCalls === 1 && plugins.length === 0);
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}

desktop/frontend/src/__tests__/composer-goal-toggle.test.tsx

@@ -8,7 +8,7 @@ import { Composer, composerPickFileEntry } from "../components/Composer";
 import { LocaleProvider } from "../lib/i18n";
 import { ToastProvider } from "../lib/toast";
 import type { AppBindings } from "../lib/bridge";
-import type { CollaborationMode, ToolApprovalMode, TokenMode } from "../lib/types";
+import type { CollaborationMode, DirEntry, ToolApprovalMode, TokenMode } from "../lib/types";
 
 let passed = 0;
 let failed = 0;
@@ -200,6 +200,16 @@ async function waitFor(label: string, predicate: () => boolean) {
   throw new Error(`timed out waiting for ${label}`);
 }
 
+type RenderedComposer = Awaited<ReturnType<typeof renderComposer>>;
+
+function fileEntry(name: string): DirEntry {
+  return { name, isDir: false };
+}
+
+async function replaceComposerDraft(rerender: RenderedComposer["rerender"], id: number, text: string) {
+  await rerender({ insertRequest: { id, text, mode: "replace" } });
+}
+
 console.log("\ncomposer goal toggle");
 
 {
@@ -739,5 +749,156 @@ console.log("\ncomposer goal toggle");
   dom.window.close();
 }
 
+{
+  const dom = installDom();
+  let listDirCalls = 0;
+  mockApp({
+    ListDir: async () => {
+      listDirCalls += 1;
+      return listDirCalls === 1 ? [fileEntry("cached-dir.txt")] : [fileEntry("fresh-dir.txt")];
+    },
+    SearchFileRefs: async () => [],
+  });
+  const { root, rerender } = await renderComposer();
+
+  await replaceComposerDraft(rerender, 101, "@");
+  await waitFor("initial @ directory load", () => listDirCalls === 1);
+
+  await replaceComposerDraft(rerender, 102, "");
+  await replaceComposerDraft(rerender, 103, "@");
+  await waitFor("@ directory revalidation call", () => listDirCalls === 2);
+
+  eq(listDirCalls, 2, "@ directory cache hit still revalidates ListDir");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
+{
+  const dom = installDom();
+  let listDirCalls = 0;
+  mockApp({
+    ListDir: async () => {
+      listDirCalls += 1;
+      return listDirCalls === 1 ? [fileEntry("manual-refresh-stale.txt")] : [fileEntry("manual-refresh-fresh.txt")];
+    },
+    SearchFileRefs: async () => [],
+  });
+  const { root, rerender } = await renderComposer({ fileRefRefreshKey: "0" });
+
+  await replaceComposerDraft(rerender, 201, "@");
+  await waitFor("initial @ directory load before refresh key", () => listDirCalls === 1);
+
+  await rerender({ fileRefRefreshKey: "1" });
+  await waitFor("@ directory reload after refresh key", () => listDirCalls === 2);
+
+  eq(listDirCalls, 2, "fileRefRefreshKey refreshes @ directory cache while the menu is open");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
+{
+  const dom = installDom();
+  const realDateNow = Date.now;
+  let now = 1000;
+  let searchCalls = 0;
+  Date.now = () => now;
+  mockApp({
+    ListDir: async () => [],
+    SearchFileRefs: async () => {
+      searchCalls += 1;
+      return searchCalls === 1 ? [fileEntry("alpha-old.ts")] : [fileEntry("alpha-new.ts")];
+    },
+  });
+  const { root, rerender } = await renderComposer();
+
+  try {
+    await replaceComposerDraft(rerender, 301, "@alpha");
+    await waitFor("initial @ search request", () => searchCalls === 1);
+    eq(searchCalls, 1, "@ search fetches the first query");
+
+    await replaceComposerDraft(rerender, 302, "");
+    now = 2000;
+    await replaceComposerDraft(rerender, 303, "@alpha");
+    await act(async () => {
+      await flushTimers();
+    });
+    eq(searchCalls, 1, "@ search cache is reused inside the TTL");
+
+    await replaceComposerDraft(rerender, 304, "");
+    now = 7001;
+    await replaceComposerDraft(rerender, 305, "@alpha");
+    await waitFor("expired @ search cache refresh", () => searchCalls === 2);
+    eq(searchCalls, 2, "@ search cache revalidates after the TTL");
+  } finally {
+    Date.now = realDateNow;
+  }
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
+{
+  const dom = installDom();
+  let staleListDirResolve: ((entries: DirEntry[]) => void) | undefined;
+  let thirdListDirResolve: ((entries: DirEntry[]) => void) | undefined;
+  let listDirCalls = 0;
+  mockApp({
+    ListDir: async () => {
+      listDirCalls += 1;
+      if (listDirCalls === 1) {
+        return new Promise<DirEntry[]>((resolve) => {
+          staleListDirResolve = resolve;
+        });
+      }
+      if (listDirCalls === 2) return [fileEntry("cache-live.txt")];
+      return new Promise<DirEntry[]>((resolve) => {
+        thirdListDirResolve = resolve;
+      });
+    },
+    SearchFileRefs: async () => [],
+  });
+  const { root, rerender } = await renderComposer({ fileRefRefreshKey: "0" });
+
+  const textarea = document.querySelector("textarea") as HTMLTextAreaElement | null;
+  if (!textarea) throw new Error("composer textarea did not render");
+
+  await replaceComposerDraft(rerender, 401, "@cache");
+  await waitFor("initial stale @ directory request", () => listDirCalls === 1);
+
+  await rerender({ fileRefRefreshKey: "1" });
+  await waitFor("fresh @ directory request after refresh key", () => listDirCalls === 2);
+  await act(async () => {
+    await flushTimers();
+  });
+
+  staleListDirResolve?.([fileEntry("cache-stale.txt")]);
+  await act(async () => {
+    await flushTimers();
+  });
+
+  await replaceComposerDraft(rerender, 402, "");
+  await replaceComposerDraft(rerender, 403, "@cache");
+  await waitFor("second fresh @ directory request", () => listDirCalls === 3);
+  await act(async () => {
+    textarea.dispatchEvent(new KeyboardEvent("keydown", { key: "Enter", bubbles: true, cancelable: true }));
+    await flushTimers();
+  });
+  eq(textarea.value, "@cache-live.txt ", "stale @ directory request cannot repopulate cache after refresh");
+  thirdListDirResolve?.([fileEntry("cache-later.txt")]);
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
 console.log(`\n${passed} passed, ${failed} failed, ${passed + failed} total`);
 if (failed > 0) process.exit(1);

desktop/frontend/src/__tests__/composer-image-capability.test.tsx

@@ -5,6 +5,7 @@ import React from "react";
 import { act } from "react";
 import { createRoot } from "react-dom/client";
 import { Composer } from "../components/Composer";
+import { UserMessage } from "../components/Message";
 import { LocaleProvider } from "../lib/i18n";
 import { ToastProvider } from "../lib/toast";
 import type { CollaborationMode, TokenMode, ToolApprovalMode } from "../lib/types";
@@ -137,7 +138,9 @@ async function renderComposer(props: Partial<Parameters<typeof Composer>[0]> = {
       root.render(
         <LocaleProvider>
           <ToastProvider>
-            <Composer {...currentProps} />
+            <div className="chat-pane">
+              <Composer {...currentProps} />
+            </div>
           </ToastProvider>
         </LocaleProvider>,
       );
@@ -183,6 +186,29 @@ function imagePasteEvent(file: File): Event {
   return event;
 }
 
+function imageViewerOpen(): boolean {
+  return Boolean(document.querySelector(".image-viewer-backdrop .image-viewer__image"));
+}
+
+function renderUserMessage(text: string, props: Partial<Parameters<typeof UserMessage>[0]> = {}) {
+  const rootEl = document.getElementById("root");
+  if (!rootEl) throw new Error("missing root");
+  const root = createRoot(rootEl);
+  const paint = async () => {
+    await act(async () => {
+      root.render(
+        <LocaleProvider>
+          <div className="chat-pane">
+            <UserMessage text={text} {...props} />
+          </div>
+        </LocaleProvider>,
+      );
+      await flushTimers();
+    });
+  };
+  return { root, paint };
+}
+
 console.log("\ncomposer image capability");
 
 {
@@ -255,5 +281,107 @@ console.log("\ncomposer image capability");
   dom.window.close();
 }
 
+{
+  const dom = installDom();
+  installBridgeApp({
+    SavePastedImage: async () => ".reasonix/attachments/mock.png",
+    AttachmentDataURL: async () => "data:image/png;base64,iVBORw0KGgo=",
+  });
+  const { root } = await renderComposer({ imageInputEnabled: true });
+  const file = new File(["img"], "photo.png", { type: "image/png", lastModified: 1 });
+
+  await act(async () => {
+    textarea().dispatchEvent(imagePasteEvent(file));
+    await flushTimers();
+    await flushTimers();
+  });
+  await waitFor(() => Boolean(document.querySelector(".composer-context__thumb img")));
+  const thumb = document.querySelector(".composer-context__thumb") as HTMLElement | null;
+  if (!thumb) throw new Error("missing composer image thumbnail");
+  await act(async () => {
+    thumb.click();
+    await flushTimers();
+  });
+  await waitFor(imageViewerOpen);
+  ok(imageViewerOpen(), "composer image thumbnail opens the image viewer");
+
+  await act(async () => {
+    document.dispatchEvent(new KeyboardEvent("keydown", { key: "Escape", bubbles: true }));
+    await flushTimers();
+  });
+  await waitFor(() => !document.querySelector(".image-viewer-backdrop"));
+  ok(!document.querySelector(".image-viewer-backdrop"), "composer image viewer closes on Escape");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
+{
+  const dom = installDom();
+  installBridgeApp({
+    AttachmentDataURL: async () => "data:image/png;base64,iVBORw0KGgo=",
+  });
+  const { root, paint } = renderUserMessage("check @[photo.png](.reasonix/attachments/mock.png)");
+  await paint();
+  await waitFor(() => Boolean(document.querySelector(".msg-attachment--image img")));
+  const thumb = document.querySelector(".msg-attachment--image") as HTMLElement | null;
+  if (!thumb) throw new Error("missing message image thumbnail");
+  await act(async () => {
+    thumb.click();
+    await flushTimers();
+  });
+  await waitFor(() => Boolean(document.querySelector(".chat-pane > .image-viewer-backdrop")));
+  ok(Boolean(document.querySelector(".chat-pane > .image-viewer-backdrop")), "sent message image preview portals into the chat pane");
+
+  const close = document.querySelector(".image-viewer__close") as HTMLButtonElement | null;
+  if (!close) throw new Error("missing image viewer close button");
+  await act(async () => {
+    close.click();
+    await flushTimers();
+  });
+  await waitFor(() => !document.querySelector(".image-viewer-backdrop"));
+  ok(!document.querySelector(".image-viewer-backdrop"), "sent message image viewer closes from the close button");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
+{
+  const dom = installDom();
+  installBridgeApp({
+    AttachmentDataURL: async () => "data:image/png;base64,iVBORw0KGgo=",
+  });
+  const { root, paint } = renderUserMessage("check @[photo.png](.reasonix/attachments/mock.png)", {
+    turn: 1,
+    onEdit: () => true,
+  });
+  await paint();
+  await waitFor(() => Boolean(document.querySelector(".msg-attachment--image img")));
+  const edit = document.querySelector("button.msg-meta__btn:not(.msg-meta__copy)") as HTMLButtonElement | null;
+  if (!edit) throw new Error("missing message edit button");
+  await act(async () => {
+    edit.click();
+    await flushTimers();
+  });
+  await waitFor(() => Boolean(document.querySelector(".msg-edit .composer-context__thumb img")));
+  const thumb = document.querySelector(".msg-edit .composer-context__thumb") as HTMLElement | null;
+  if (!thumb) throw new Error("missing edit image thumbnail");
+  await act(async () => {
+    thumb.click();
+    await flushTimers();
+  });
+  await waitFor(imageViewerOpen);
+  ok(imageViewerOpen(), "edit message image thumbnail opens the image viewer");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
 console.log(`\n${passed} passed, ${failed} failed, ${passed + failed} total`);
 if (failed > 0) process.exit(1);

desktop/frontend/src/__tests__/context-panel-breakdown.test.ts

@@ -1,6 +1,6 @@
 // Run: tsx src/__tests__/context-panel-breakdown.test.ts
 
-import { cacheHitTone, contextBreakdown, contextCostDisplay, contextSourceRows, contextWindowStatus, formatCacheHitRate, formatMetricTokens } from "../components/ContextPanel";
+import { cacheHitTone, contextBreakdown, contextCostDisplay, contextSourceRows, contextUsageRefreshKey, contextWindowStatus, formatCacheHitRate, formatMetricTokens } from "../components/ContextPanel";
 import { currencySymbol, formatMoney, formatMoneyLocalized } from "../lib/money";
 
 let passed = 0;
@@ -130,6 +130,28 @@ eq(cacheHitTone(6000, 4000), "notice", "mid cache hit rate uses notice tone");
 eq(cacheHitTone(5999, 4001), "warn", "low cache hit rate uses warning tone");
 eq(cacheHitTone(0, 0), undefined, "missing cache data stays uncolored");
 
+console.log("\ncontext panel usage refresh key");
+
+eq(contextUsageRefreshKey(undefined), "", "missing usage does not request a streaming refresh");
+ok(
+  contextUsageRefreshKey({
+    totalTokens: 10,
+    promptTokens: 10,
+    completionTokens: 0,
+    reasoningTokens: 0,
+    sessionCacheHitTokens: 0,
+    sessionCacheMissTokens: 0,
+  }) !== contextUsageRefreshKey({
+    totalTokens: 11,
+    promptTokens: 10,
+    completionTokens: 1,
+    reasoningTokens: 0,
+    sessionCacheHitTokens: 0,
+    sessionCacheMissTokens: 0,
+  }),
+  "general token changes refresh even when cache counters stay unchanged",
+);
+
 console.log("\ncontext panel source rows");
 
 const sourceRows = contextSourceRows({

desktop/frontend/src/__tests__/message-pasted-blocks.test.ts

@@ -0,0 +1,39 @@
+// Run: tsx src/__tests__/message-pasted-blocks.test.ts
+
+import { parsePastedBlocks } from "../components/Message";
+
+let passed = 0;
+let failed = 0;
+
+function eq(a: unknown, b: unknown, label: string) {
+  if (JSON.stringify(a) === JSON.stringify(b)) {
+    process.stdout.write(`  PASS  ${label}\n`);
+    passed += 1;
+  } else {
+    process.stdout.write(`  FAIL  ${label}: expected ${JSON.stringify(b)}, got ${JSON.stringify(a)}\n`);
+    failed += 1;
+  }
+}
+
+function wrapped(label: string, content: string): string {
+  return `${label}\n\n--- Begin ${label} ---\n${content}\n--- End ${label} ---`;
+}
+
+console.log("\nmessage pasted blocks");
+
+for (const [label, name] of [
+  ["[已粘贴文本 #2 · 31 行]", "Simplified Chinese"],
+  ["[已貼上文字 #2 · 31 行]", "Traditional Chinese"],
+  ["[Pasted text #2 · 31 lines]", "English"],
+] as const) {
+  eq(
+    parsePastedBlocks(`before\n${label}\nafter`, wrapped(label, "line 1\nline 2")),
+    [{ label, content: "line 1\nline 2\n" }],
+    `${name} pasted text labels are parsed from submit text`,
+  );
+}
+
+eq(parsePastedBlocks("[unknown paste #1]", "--- Begin [unknown paste #1] ---\nnope\n--- End [unknown paste #1] ---"), [], "unknown labels are ignored");
+
+console.log(`\n${passed} passed, ${failed} failed`);
+if (failed > 0) process.exit(1);

desktop/frontend/src/__tests__/new-session-load-race.test.tsx

@@ -74,7 +74,7 @@ function tabMeta(overrides: Partial<TabMeta> = {}): TabMeta {
   };
 }
 
-function meta(): Meta {
+function meta(overrides: Partial<Meta> = {}): Meta {
   return {
     label: "model",
     ready: true,
@@ -91,6 +91,7 @@ function meta(): Meta {
     tokenMode: "full",
     goal: "",
     goalStatus: "stopped",
+    ...overrides,
   };
 }
 
@@ -211,10 +212,14 @@ await act(async () => {
 });
 
 const guardedStartupTabs = deferred<TabMeta[]>();
-let ensureBlankSurfaceCalls = 0;
+const staleProjectA = "/repo/project-a";
+const targetProjectB = "/repo/project-b";
+const ensureBlankSurfaceCalls: Array<{ scope: string; workspaceRoot: string }> = [];
 window.go.main.App = {
   ListTabs: async () => guardedStartupTabs.promise,
-  MetaForTab: async () => meta(),
+  MetaForTab: async (tabID: string) => tabID === "tab-new"
+    ? meta({ cwd: targetProjectB, workspaceRoot: targetProjectB, workspaceName: "project-b", workspacePath: targetProjectB })
+    : meta({ cwd: staleProjectA, workspaceRoot: staleProjectA, workspaceName: "project-a", workspacePath: staleProjectA }),
   ContextUsageForTab: async () => context,
   EffortForTab: async () => effort,
   BalanceForTab: async () => balance,
@@ -224,9 +229,17 @@ window.go.main.App = {
   HistoryPageForTab: async () => ({ messages: [], startTurn: 0, endTurn: 0, totalTurns: 0, hasOlder: false }),
   HistoryCheckpointTurnsForTab: async () => [],
   ReplayPendingPrompts: async () => {},
-  EnsureBlankSurface: async () => {
-    ensureBlankSurfaceCalls += 1;
-    return tabMeta({ id: "tab-new", topicId: "topic-new", topicTitle: "New session" });
+  EnsureBlankSurface: async (scope: string, workspaceRoot: string) => {
+    ensureBlankSurfaceCalls.push({ scope, workspaceRoot });
+    return tabMeta({
+      id: "tab-new",
+      topicId: "topic-new",
+      topicTitle: "New session",
+      workspaceRoot: targetProjectB,
+      workspaceName: "project-b",
+      workspacePath: targetProjectB,
+      cwd: targetProjectB,
+    });
   },
 } as Partial<AppBindings> as AppBindings;
 
@@ -239,20 +252,31 @@ await act(async () => {
 });
 
 await act(async () => {
-  await controller?.ensureBlankSurface("project", "/repo");
+  await controller?.ensureBlankSurface("project", targetProjectB);
   await flushPromises();
 });
 
-eq(ensureBlankSurfaceCalls, 1, "EnsureBlankSurface is called once");
+eq(ensureBlankSurfaceCalls.length, 1, "EnsureBlankSurface is called once");
+eq(ensureBlankSurfaceCalls[0]?.workspaceRoot, targetProjectB, "EnsureBlankSurface keeps the requested project root");
 eq(controller?.activeTabId, "tab-new", "blank surface becomes active before startup sync resolves");
+eq(controller?.state.meta?.workspaceRoot, targetProjectB, "blank surface exposes the new project root");
 
 await act(async () => {
-  guardedStartupTabs.resolve([tabMeta({ id: "tab-old", topicId: "topic-old", topicTitle: "Old session" })]);
+  guardedStartupTabs.resolve([tabMeta({
+    id: "tab-old",
+    topicId: "topic-old",
+    topicTitle: "Old session",
+    workspaceRoot: staleProjectA,
+    workspaceName: "project-a",
+    workspacePath: staleProjectA,
+    cwd: staleProjectA,
+  })]);
   await guardedStartupTabs.promise;
   await flushPromises();
 });
 
 eq(controller?.activeTabId, "tab-new", "guarded startup sync cannot restore an older active tab");
+eq(controller?.state.meta?.workspaceRoot, targetProjectB, "guarded startup sync cannot restore the old project root");
 
 await act(async () => {
   guardRoot.unmount();

desktop/frontend/src/__tests__/provider-model-refresh.test.ts

@@ -47,14 +47,14 @@ eq(
 );
 
 eq(
-  mergedFetchedProviderModels(["mimo-v2.5-pro"], ["mimo-v2-flash", "mimo-v2-omni", "mimo-v2.5-pro"], { preserveCurated: true }),
+  mergedFetchedProviderModels(["mimo-v2.5-pro"], ["mimo-v2.5", "mimo-v2.5-pro"], { preserveCurated: true }),
   ["mimo-v2.5-pro"],
   "manual access refresh preserves selected MiMo model instead of importing provider catalog",
 );
 
 eq(
-  providerModelCandidates(["mimo-v2.5-pro"], ["mimo-v2-flash", "mimo-v2-omni", "mimo-v2.5-pro"]),
-  ["mimo-v2.5-pro", "mimo-v2-flash", "mimo-v2-omni"],
+  providerModelCandidates(["mimo-v2.5-pro"], ["mimo-v2.5", "mimo-v2.5-pro"]),
+  ["mimo-v2.5-pro", "mimo-v2.5"],
   "manual access refresh can show provider catalog as unsaved candidates",
 );
 

desktop/frontend/src/__tests__/settings-refresh-snapshot.test.tsx

@@ -6,6 +6,8 @@ import { act } from "react";
 import { createRoot } from "react-dom/client";
 import {
   SettingsPanel,
+  formatProviderExtraBody,
+  parseProviderExtraBody,
   providerBaseURLFromChatURL,
   providerChatURLPreview,
   providerEditorEffectiveKind,
@@ -58,18 +60,52 @@ function baseSettings(displayMode: "standard" | "compact" = "standard"): Setting
     autoPlan: "off",
     providers: [],
     officialProviders: [],
+    providerPresets: [],
     permissions: { mode: "ask", allow: [], ask: [], deny: [] },
-    sandbox: { bash: "enforce", network: false, workspaceRoot: "", allowWrite: [], shell: "auto" },
+    sandbox: { bash: "enforce", network: false, workspaceRoot: "", allowWrite: [], effectiveWorkspaceRoot: "/work", effectiveWriteRoots: ["/work"], shell: "auto" },
     network: { proxyMode: "auto", proxyUrl: "", noProxy: "", proxy: { type: "socks5", server: "", port: 0, username: "", password: "" } },
-    agent: { temperature: 0, maxSteps: 0, plannerMaxSteps: 0, systemPrompt: "", coldResumePrune: true, reasoningLanguage: "auto" },
+    agent: { temperature: 0, maxSteps: 0, plannerMaxSteps: 0, maxSubagentDepth: 2, systemPrompt: "", coldResumePrune: true, reasoningLanguage: "auto" },
     bot: {
       enabled: false,
       model: "",
-      toolApprovalMode: "",
-      maxSteps: 0,
-      debounceMs: 0,
-      allowlist: { enabled: false, allowAll: false, qqUsers: [], feishuUsers: [], weixinUsers: [], qqGroups: [], feishuGroups: [], weixinGroups: [] },
-      qq: { enabled: false, appId: "", appSecretEnv: "", secretSet: false, sandbox: false },
+	      toolApprovalMode: "",
+	      maxSteps: 0,
+	      debounceMs: 0,
+	      queueMode: "steer",
+	      queueCap: 20,
+	      queueDrop: "summarize",
+	      ignoreSelfMessages: true,
+	      selfUserIds: { qq: [], feishu: [], weixin: [] },
+	      control: { enabled: false, addr: "127.0.0.1:37913", tokenEnv: "REASONIX_BOT_CONTROL_TOKEN" },
+	      pairing: { enabled: true, requestTtlMinutes: 60, maxPendingPerPlatform: 3 },
+	      routes: [],
+	      allowlist: {
+	        enabled: false,
+	        allowAll: false,
+	        qqUsers: [],
+	        feishuUsers: [],
+	        weixinUsers: [],
+	        qqApprovers: [],
+	        feishuApprovers: [],
+	        weixinApprovers: [],
+	        qqAdmins: [],
+	        feishuAdmins: [],
+	        weixinAdmins: [],
+	        qqGroups: [],
+	        feishuGroups: [],
+	        weixinGroups: [],
+	      },
+      qq: {
+        enabled: false,
+        appId: "",
+        appSecretEnv: "",
+        secretSet: false,
+        sandbox: false,
+        model: "",
+        toolApprovalMode: "ask",
+        workspaceRoot: "",
+        access: { enabled: true, allowAll: false, pairingEnabled: true, users: [], groups: [], approvers: [], admins: [] },
+      },
       feishu: { enabled: false, domain: "feishu", appId: "", appSecretEnv: "", secretSet: false, verificationToken: "", mode: "webhook", webhookPort: 0, requireMention: false },
       weixin: { enabled: false, accountId: "", tokenEnv: "", tokenSet: false, apiBase: "" },
       connections: [],
@@ -96,11 +132,20 @@ function baseSettings(displayMode: "standard" | "compact" = "standard"): Setting
 
 console.log("\nsettings refresh snapshot");
 
-eq(providerEditorEffectiveKind(true, "anthropic", ["anthropic", "openai"]), "openai", "new custom providers ignore sorted providerKinds and default to OpenAI");
+eq(providerEditorEffectiveKind(true, "anthropic", ["anthropic", "openai"]), "anthropic", "new custom providers keep the selected Anthropic-compatible kind");
 eq(providerEditorEffectiveKind(false, "anthropic", ["anthropic", "openai"]), "anthropic", "existing providers preserve their stored kind");
 eq(providerChatURLPreview("https://proxy.example.com/v1", "", false), "https://proxy.example.com/v1/chat/completions", "base URL mode previews chat completions URL");
 eq(providerChatURLPreview("", "https://proxy.example.com/custom/chat", true), "https://proxy.example.com/custom/chat", "full URL mode previews configured URL");
 eq(providerBaseURLFromChatURL("https://proxy.example.com/v1/chat/completions"), "https://proxy.example.com/v1", "chat URL derives base URL for model discovery");
+eq(formatProviderExtraBody({ top_p: 0.7, enable_thinking: true }), "{\n  \"enable_thinking\": true,\n  \"top_p\": 0.7\n}", "extra body editor formats stable JSON");
+eq(JSON.stringify(parseProviderExtraBody('{ "enable_thinking": true, "top_p": 0.7 }')), "{\"enable_thinking\":true,\"top_p\":0.7}", "extra body editor parses JSON object");
+let extraBodyRejected = false;
+try {
+  parseProviderExtraBody("[true]");
+} catch {
+  extraBodyRejected = true;
+}
+ok(extraBodyRejected, "extra body editor rejects non-object JSON");
 
 const dom = new JSDOM("<!doctype html><html><body><div id=\"root\"></div></body></html>", {
   pretendToBeVisual: true,
@@ -220,6 +265,88 @@ ok(document.body.textContent?.includes("Settings could not be loaded.") === fals
 await act(async () => {
   retryRoot.unmount();
 });
+
+// Bots tab: direct four-channel bot manager.
+const botsRootEl = document.createElement("div");
+document.body.appendChild(botsRootEl);
+const botsRoot = createRoot(botsRootEl);
+const botsSettings = baseSettings("standard");
+botsSettings.bot.connections = [
+  {
+    id: "conn-feishu-1",
+    provider: "feishu",
+    domain: "feishu",
+    label: "kun",
+    enabled: true,
+    status: "connected",
+    model: "",
+    toolApprovalMode: "",
+    workspaceRoot: "",
+    credential: { appId: "cli_mock", appSecretEnv: "FEISHU_BOT_APP_SECRET", accountId: "", tokenEnv: "", secretSet: true },
+    sessionMappings: [],
+    lastError: "",
+    createdAt: "",
+	    updatedAt: "",
+	    access: { enabled: true, allowAll: false, pairingEnabled: true, users: ["ou_mock_user_001"], groups: [], approvers: [], admins: [] },
+	  },
+	];
+window.go = {
+  main: {
+    App: {
+      Settings: async () => botsSettings,
+    } as Partial<AppBindings> as AppBindings,
+  },
+};
+
+await act(async () => {
+  botsRoot.render(
+    <LocaleProvider>
+      <SettingsPanel initialTab="bots" onClose={() => {}} onChanged={() => {}} />
+    </LocaleProvider>,
+  );
+  await flushPromises();
+});
+await waitFor("bot channel manager", () => Boolean(document.querySelector(".bot-channel-manager")));
+
+ok(!document.querySelector(".bot-overview-grid"), "bots tab does not render the removed entry overview");
+ok(!document.getElementById("bot-mobile-remote"), "bots tab no longer renders the mobile remote entry card");
+ok(!document.querySelector(".bot-channel-entry"), "bots tab no longer renders the Bot Channel entry panel");
+ok(!document.getElementById("bot-step-access"), "bots tab omits the old global access step card");
+ok(!document.getElementById("bot-step-behavior"), "bots tab omits global default behavior card");
+eq(document.querySelectorAll(".bot-step-chip").length, 0, "hero no longer shows the old two-step chips");
+
+eq(document.querySelectorAll(".bot-channel-tabs [role=\"tab\"]").length, 4, "bot manager uses four fixed channel tabs on the left");
+ok(document.querySelector(".bot-channel-setup-card")?.textContent?.includes("Configure QQ") === true, "unconfigured QQ tab shows key setup on the right");
+ok(document.body.textContent?.includes("Back to entry") === false, "bot manager does not show a return-to-entry action");
+
+const feishuTab = Array.from(document.querySelectorAll(".bot-channel-tabs [role=\"tab\"]")).find((button) => button.textContent?.includes("Feishu")) as HTMLButtonElement | undefined;
+if (!feishuTab) throw new Error("Feishu channel tab did not render");
+await act(async () => {
+  feishuTab.click();
+  await flushPromises();
+});
+await waitFor("selected Feishu detail", () => Boolean(document.querySelector(".bot-channel-manager__detail .bot-detail-card")));
+
+ok(Boolean(document.querySelector(".bot-channel-manager__detail .bot-detail-card")), "configured channel renders selected bot detail on the right");
+ok(Boolean(document.querySelector(".bot-channel-manager__detail .bot-detail-section--access")), "selected bot detail owns its access control");
+ok(document.body.textContent?.includes("Access control") === true, "selected bot detail labels per-bot access control");
+const selectedBotDetailText = document.querySelector(".bot-channel-manager__detail .bot-detail-card")?.textContent ?? "";
+const connectionSummaryIndex = selectedBotDetailText.indexOf("Connection summary");
+const enableBotIndex = selectedBotDetailText.indexOf("Enable bot");
+const toolApprovalIndex = selectedBotDetailText.indexOf("Tool approval");
+const modelIndex = selectedBotDetailText.indexOf("Model");
+const accessControlIndex = selectedBotDetailText.indexOf("Access control");
+ok(
+  connectionSummaryIndex >= 0 && enableBotIndex > connectionSummaryIndex && toolApprovalIndex > enableBotIndex && modelIndex > toolApprovalIndex && accessControlIndex > modelIndex,
+  "selected bot detail places enable, approval, and model controls between summary and access control",
+);
+ok(document.body.textContent?.includes("ou_mock_user_001") === true, "selected bot detail shows its trusted user");
+ok(document.body.textContent?.includes("Legacy global allowlist") === true, "advanced area keeps the legacy global allowlist");
+ok(document.querySelector(".bot-simple-advanced")?.textContent?.includes("local control API") === false, "advanced area no longer owns mobile/control API setup");
+
+await act(async () => {
+  botsRoot.unmount();
+});
 dom.window.close();
 
 console.log(`\n${passed} passed, ${failed} failed, ${passed + failed} total`);

desktop/frontend/src/__tests__/tab-switch-hydration.test.tsx

@@ -130,25 +130,34 @@ const tabA = tabMeta("tab-a", { active: true });
 const tabB = tabMeta("tab-b");
 const tabC = tabMeta("tab-c");
 const tabD = tabMeta("tab-d");
+const tabE = tabMeta("tab-e");
+const tabF = tabMeta("tab-f");
+const tabG = tabMeta("tab-g");
 let backendActiveId = "tab-a";
 const historyB = deferred<HistoryMessage[]>();
 const historyD = deferred<HistoryMessage[]>();
 const contextDGate = deferred<ContextInfo>();
 const setActiveBGate = deferred<void>();
+const setActiveEGate = deferred<void>();
+const setActiveFGate = deferred<void>();
 const submitTabCGate = deferred<void>();
 const historyCalls: string[] = [];
+const cancelCalls: string[] = [];
 let contextDCalls = 0;
 let holdNextContextForD = false;
 let setActiveCalls = 0;
 let newSessionCalls = 0;
 let failSetActiveFor = "";
 const runningTabs = new Set<string>();
-const tabsById = new Map([tabA, tabB, tabC, tabD].map((tab) => [tab.id, tab]));
+const tabsById = new Map([tabA, tabB, tabC, tabD, tabE, tabF, tabG].map((tab) => [tab.id, tab]));
 const eventHandlers: Array<(e: WireEvent) => void> = [];
 const readyHandlers: Array<(tabId?: string) => void> = [];
 
 function currentTabs(): TabMeta[] {
-  return [tabA, tabB, tabC, tabD].map((tab) => ({ ...tab, active: tab.id === backendActiveId, running: runningTabs.has(tab.id) }));
+  return Array.from(tabsById.values()).map((tab) => {
+    const running = runningTabs.has(tab.id);
+    return { ...tab, active: tab.id === backendActiveId, running, cancellable: running };
+  });
 }
 
 window.runtime = {
@@ -181,6 +190,7 @@ window.go = {
         historyCalls.push(tabID);
         if (tabID === "tab-b") return historyB.promise;
         if (tabID === "tab-d") return historyD.promise;
+        if (tabID === "tab-e") return [userMessage("fork E")];
         return [userMessage("cached A")];
       },
       HistoryPageForTab: async (tabID: string) => {
@@ -195,13 +205,25 @@ window.go = {
       NewSession: async () => {
         newSessionCalls += 1;
       },
+      Fork: async () => {
+        tabsById.set("tab-e", tabE);
+        backendActiveId = "tab-e";
+        runningTabs.add("tab-e");
+        return { ...tabE, active: true, running: true };
+      },
       ReplayPendingPrompts: async () => {},
       SetActiveTab: async (tabID: string) => {
         setActiveCalls += 1;
         if (tabID === "tab-b") await setActiveBGate.promise;
+        if (tabID === "tab-e") await setActiveEGate.promise;
+        if (tabID === "tab-f") await setActiveFGate.promise;
         if (tabID === failSetActiveFor) throw new Error("persist failed");
         backendActiveId = tabID;
       },
+      CancelTab: async (tabID: string) => {
+        cancelCalls.push(tabID);
+        runningTabs.delete(tabID);
+      },
       SubmitToTab: async (tabID: string) => {
         runningTabs.add(tabID);
         if (tabID === "tab-c") await submitTabCGate.promise;
@@ -280,6 +302,10 @@ ok(controller?.state.items.some((item) => item.kind === "user" && item.text ===
 ok(!(controller?.state.items.some((item) => item.kind === "user" && item.text === "late B") ?? false), "late history stays scoped to its tab state");
 
 const historyCallsBeforeFallbackSync = historyCalls.length;
+await act(async () => {
+  for (const handler of eventHandlers) handler({ kind: "approval_request", tabId: "tab-b", approval: { id: "stale-fallback-approval", tool: "bash", subject: "stale fallback approval" } });
+  await flushPromises();
+});
 backendActiveId = "tab-b";
 await act(async () => {
   await controller?.syncActiveTab(false);
@@ -288,12 +314,69 @@ await act(async () => {
 eq(controller?.activeTabId, "tab-b", "backend fallback sync activates the backend-selected cached tab");
 ok(controller?.state.items.some((item) => item.kind === "user" && item.text === "late B") ?? false, "backend fallback sync keeps the cached transcript");
 eq(historyCalls.length, historyCallsBeforeFallbackSync, "backend fallback sync preserves cached history instead of reloading it");
+eq(controller?.state.approval?.id, undefined, "backend fallback sync reconciles stale approval state");
+eq(controller?.state.running, false, "backend fallback sync reconciles stale running state");
 await act(async () => {
   await controller?.switchTab("tab-a", tabA);
   await flushPromises();
 });
 await waitFor("tab-a restored after fallback sync", () => controller?.activeTabId === "tab-a" && controller.state.items.some((item) => item.kind === "user" && item.text === "cached A"));
 
+runningTabs.add("tab-e");
+let switchToE: Promise<TabMeta[] | undefined> | undefined;
+await act(async () => {
+  switchToE = controller?.switchTab("tab-e", { ...tabE, running: true, cancellable: true });
+  await flushPromises();
+});
+eq(controller?.activeTabId, "tab-e", "switching to a backend-running tab updates the active tab immediately");
+eq(controller?.state.running, true, "backend-running tab restores the stop state before backend activation settles");
+eq(controller?.state.cancellable, true, "backend-running tab remains cancellable before backend activation settles");
+await act(async () => {
+  controller?.cancel();
+  await flushPromises();
+});
+eq(cancelCalls.join(","), "tab-e", "cancel targets the backend-running tab while activation is pending");
+await act(async () => {
+  setActiveEGate.resolve();
+  await switchToE;
+  await flushPromises();
+});
+eq(controller?.state.running, false, "cancelled backend-running tab reconciles to idle after activation");
+await act(async () => {
+  await controller?.switchTab("tab-a", tabA);
+  await flushPromises();
+});
+await waitFor("tab-a restored after backend-running switch", () => controller?.activeTabId === "tab-a" && controller.state.items.some((item) => item.kind === "user" && item.text === "cached A"));
+
+let switchToF: Promise<TabMeta[] | undefined> | undefined;
+await act(async () => {
+  switchToF = controller?.switchTab("tab-f", tabF);
+  await flushPromises();
+});
+eq(controller?.activeTabId, "tab-f", "first rapid switch activates the slow target optimistically");
+let switchToG: Promise<TabMeta[] | undefined> | undefined;
+await act(async () => {
+  switchToG = controller?.switchTab("tab-g", tabG);
+  await switchToG;
+  await flushPromises();
+});
+eq(controller?.activeTabId, "tab-g", "second rapid switch wins immediately");
+eq(backendActiveId, "tab-g", "second rapid switch activates the backend");
+await act(async () => {
+  setActiveFGate.resolve();
+  await switchToF;
+  await flushPromises();
+});
+eq(controller?.activeTabId, "tab-g", "late completion from the first rapid switch does not replace the visible tab");
+eq(backendActiveId, "tab-g", "late completion from the first rapid switch reasserts the last-clicked backend tab");
+ok(!historyCalls.includes("tab-f"), "late completion from the first rapid switch does not hydrate the stale target");
+
+await act(async () => {
+  await controller?.switchTab("tab-a", tabA);
+  await flushPromises();
+});
+await waitFor("tab-a restored after rapid switch", () => controller?.activeTabId === "tab-a" && controller.state.items.some((item) => item.kind === "user" && item.text === "cached A"));
+
 failSetActiveFor = "tab-b";
 const historyCallsBeforeFailedSwitch = historyCalls.length;
 await act(async () => {
@@ -382,6 +465,11 @@ ok(controller?.state.items.some((item) => item.kind === "user" && item.text ===
 eq(controller?.state.hydratePlaceholderItems?.length ?? 0, 0, "topic history clears the hydration placeholder");
 
 const historyCallsBeforeReopenD = historyCalls.length;
+await act(async () => {
+  for (const handler of eventHandlers) handler({ kind: "approval_request", tabId: "tab-d", approval: { id: "stale-approval", tool: "bash", subject: "stale approval" } });
+  await controller?.switchTab("tab-a", tabA);
+  await flushPromises();
+});
 await act(async () => {
   await controller?.openProjectTab(tabD.workspaceRoot, tabD.topicId || "");
   await flushPromises();
@@ -389,6 +477,17 @@ await act(async () => {
 eq(controller?.activeTabId, "tab-d", "reopening an already hydrated topic keeps it active");
 ok(controller?.state.items.some((item) => item.kind === "user" && item.text === "history D") ?? false, "reopened cached topic keeps its transcript");
 eq(historyCalls.length, historyCallsBeforeReopenD, "reopening an already hydrated topic skips history hydration");
+eq(controller?.state.approval?.id, undefined, "reopening a topic reconciles stale approval state");
+eq(controller?.state.running, false, "reopening a topic reconciles stale running state");
+
+await act(async () => {
+  await controller?.rewind(0, "fork");
+  await flushPromises();
+});
+eq(controller?.activeTabId, "tab-e", "fork activates the forked tab");
+ok(controller?.state.items.some((item) => item.kind === "user" && item.text === "fork E") ?? false, "fork loads the forked transcript");
+eq(controller?.state.running, true, "fork reconciles backend running state after reset hydration");
+runningTabs.delete("tab-e");
 
 const contextCallsBeforeInactiveD = contextDCalls;
 await act(async () => {

desktop/frontend/src/__tests__/typography-overflow-contract.test.ts

@@ -113,6 +113,10 @@ eq(
   "var(--statusbar-dock-height)",
   "status bar min-height follows the scaled dock token",
 );
+eq(finalDeclaration(".provider-template-grid", "grid-auto-rows"), "92px", "provider preset cards use compact equal-height grid rows");
+eq(finalDeclaration(".provider-template-card", "height"), "100%", "provider preset cards stretch to the grid row height");
+eq(finalDeclaration(".provider-template-card strong", "-webkit-line-clamp"), "1", "provider preset card titles clamp to one line");
+eq(finalDeclaration(".provider-template-card span", "-webkit-line-clamp"), "2", "provider preset card descriptions clamp to two lines");
 
 eq(finalDeclaration(".statusbar", "white-space"), "nowrap", "status bar keeps metrics on one row");
 eq(finalDeclaration(".statusbar", "overflow"), "hidden", "status bar clips instead of overflowing");
@@ -160,6 +164,16 @@ ok(
 );
 
 eq(finalDeclaration(".composer-modebar", "overflow"), "hidden", "chat mode switcher contains enlarged labels");
+eq(
+  finalDeclaration(".app--creation .tool:not(.tool--open) > .tool__body", "height"),
+  "0 !important",
+  "collapsed creation tool bodies keep mounted content clipped",
+);
+eq(
+  finalDeclaration(".app--creation .tool:not(.tool--open) > .tool__body", "visibility"),
+  "hidden",
+  "collapsed creation tool bodies do not paint hidden tool text",
+);
 ok(
   /@container\s*\(max-width:\s*760px\)[\s\S]*?\.composer-meta__control--model\s*\{[\s\S]*?flex\s*:\s*0 1 auto[\s\S]*?width\s*:\s*fit-content[\s\S]*?max-width\s*:\s*min\(240px,\s*42vw\)[\s\S]*?\.composer-meta--has-intent-chip\s+\.composer-meta__control--model\s*\{[\s\S]*?flex\s*:\s*0 1 auto[\s\S]*?width\s*:\s*fit-content[\s\S]*?max-width\s*:\s*min\(220px,\s*38vw\)[\s\S]*?\.composer-meta__control--effort\s*\{[\s\S]*?display\s*:\s*none[\s\S]*?\.composer-meta__control--more\s*\{[\s\S]*?display\s*:\s*inline-flex/.test(styles),
   "composer compact controls activate at the capped theme width",

desktop/frontend/src/__tests__/use-controller-send-fallback.test.tsx

@@ -0,0 +1,163 @@
+// Run: tsx src/__tests__/use-controller-send-fallback.test.tsx
+
+import { JSDOM } from "jsdom";
+import React, { act } from "react";
+import { createRoot } from "react-dom/client";
+import type { AppBindings } from "../lib/bridge";
+import { useController } from "../lib/useController";
+import type { BalanceInfo, CheckpointMeta, ContextInfo, EffortInfo, HistoryMessage, JobView, Meta, TabMeta } from "../lib/types";
+
+let passed = 0;
+let failed = 0;
+
+function ok(value: boolean, label: string) {
+  if (value) {
+    process.stdout.write(`  PASS  ${label}\n`);
+    passed += 1;
+  } else {
+    process.stdout.write(`  FAIL  ${label}\n`);
+    failed += 1;
+  }
+}
+
+function eq(actual: unknown, expected: unknown, label: string) {
+  ok(actual === expected, `${label}${actual === expected ? "" : `: expected ${JSON.stringify(expected)}, got ${JSON.stringify(actual)}`}`);
+}
+
+function flushPromises(): Promise<void> {
+  return new Promise((resolve) => setTimeout(resolve, 0));
+}
+
+function tabMeta(overrides: Partial<TabMeta> = {}): TabMeta {
+  return {
+    id: "tab-send",
+    scope: "project",
+    workspaceRoot: "/repo/send",
+    workspaceName: "send",
+    workspacePath: "/repo/send",
+    gitBranch: "main",
+    topicId: "topic-send",
+    topicTitle: "Send",
+    label: "model-send",
+    ready: true,
+    running: false,
+    mode: "normal",
+    toolApprovalMode: "ask",
+    tokenMode: "full",
+    active: true,
+    cwd: "/repo/send",
+    ...overrides,
+  };
+}
+
+function metaFor(tab: TabMeta): Meta {
+  return {
+    label: tab.label,
+    ready: tab.ready,
+    startupErr: tab.startupErr,
+    eventChannel: "agent:event",
+    cwd: tab.cwd || tab.workspaceRoot,
+    workspaceRoot: tab.workspaceRoot,
+    workspaceName: tab.workspaceName,
+    workspacePath: tab.workspacePath,
+    gitBranch: tab.gitBranch,
+    autoApproveTools: false,
+    bypass: false,
+    collaborationMode: tab.collaborationMode ?? "normal",
+    toolApprovalMode: tab.toolApprovalMode ?? "ask",
+    tokenMode: tab.tokenMode ?? "full",
+    goal: "",
+    goalStatus: "stopped",
+  };
+}
+
+console.log("\nuse controller send fallback");
+
+const dom = new JSDOM("<!doctype html><html><body><div id=\"root\"></div></body></html>", {
+  pretendToBeVisual: true,
+  url: "http://localhost/",
+});
+(globalThis as typeof globalThis & { IS_REACT_ACT_ENVIRONMENT: boolean }).IS_REACT_ACT_ENVIRONMENT = true;
+globalThis.window = dom.window as unknown as Window & typeof globalThis;
+globalThis.document = dom.window.document;
+Object.defineProperty(globalThis, "navigator", { configurable: true, value: dom.window.navigator });
+globalThis.Node = dom.window.Node;
+globalThis.HTMLElement = dom.window.HTMLElement;
+globalThis.Event = dom.window.Event;
+globalThis.CustomEvent = dom.window.CustomEvent;
+globalThis.KeyboardEvent = dom.window.KeyboardEvent;
+globalThis.MouseEvent = dom.window.MouseEvent;
+globalThis.localStorage = dom.window.localStorage;
+globalThis.requestAnimationFrame = dom.window.requestAnimationFrame.bind(dom.window);
+globalThis.cancelAnimationFrame = dom.window.cancelAnimationFrame.bind(dom.window);
+
+const backendTab = tabMeta({ backgroundJobs: 2 });
+const context: ContextInfo = { used: 0, window: 100, sessionTokens: 0 };
+const effort: EffortInfo = { supported: true, current: "auto", default: "auto", levels: ["auto"] };
+const balance: BalanceInfo = { available: false, display: "" };
+const jobs: JobView[] = [];
+const checkpoints: CheckpointMeta[] = [];
+let tabsAvailable = false;
+let submitCalls = 0;
+
+window.runtime = {
+  EventsOn: () => () => {},
+  BrowserOpenURL: () => {},
+};
+window.go = {
+  main: {
+    App: {
+      ListTabs: async () => (tabsAvailable ? [backendTab] : []),
+      MetaForTab: async () => metaFor(backendTab),
+      ContextUsageForTab: async () => context,
+      EffortForTab: async () => effort,
+      BalanceForTab: async () => balance,
+      JobsForTab: async () => jobs,
+      CheckpointsForTab: async () => checkpoints,
+      HistoryForTab: async (): Promise<HistoryMessage[]> => [],
+      HistoryPageForTab: async () => ({ messages: [], startTurn: 0, endTurn: 0, totalTurns: 0, hasOlder: false }),
+      HistoryCheckpointTurnsForTab: async () => [],
+      ReplayPendingPrompts: async () => {},
+      SubmitToTab: async (tabId: string) => {
+        submitCalls += tabId === "tab-send" ? 1 : 0;
+      },
+    } as Partial<AppBindings> as AppBindings,
+  },
+};
+
+type Controller = ReturnType<typeof useController>;
+let controller: Controller | undefined;
+
+function Probe() {
+  controller = useController();
+  return null;
+}
+
+const rootEl = document.getElementById("root");
+if (!rootEl) throw new Error("missing root");
+const root = createRoot(rootEl);
+
+await act(async () => {
+  root.render(<Probe />);
+  await flushPromises();
+});
+eq(controller?.activeTabId, undefined, "startup has no active tab when backend has no tabs");
+
+tabsAvailable = true;
+await act(async () => {
+  await controller?.send("hello from fallback");
+  await flushPromises();
+});
+
+eq(controller?.activeTabId, "tab-send", "send fallback activates the backend-selected tab");
+eq(controller?.state.backgroundJobs, 2, "send fallback reconciles backend runtime metadata");
+ok(controller?.state.items.some((item) => item.kind === "user" && item.text === "hello from fallback") ?? false, "send fallback keeps the optimistic user turn");
+eq(submitCalls, 1, "send fallback submits to the activated tab");
+
+await act(async () => {
+  root.unmount();
+});
+dom.window.close();
+
+console.log(`\n${passed} passed, ${failed} failed, ${passed + failed} total`);
+if (failed > 0) process.exit(1);

desktop/frontend/src/__tests__/workspace-changes-errors.test.tsx

@@ -7,7 +7,7 @@ import { createRoot } from "react-dom/client";
 import { WorkspacePanel } from "../components/WorkspacePanel";
 import { LocaleProvider } from "../lib/i18n";
 import type { AppBindings } from "../lib/bridge";
-import type { WorkspaceChangesView } from "../lib/types";
+import type { DirEntry, WorkspaceChangesView } from "../lib/types";
 
 let passed = 0;
 let failed = 0;
@@ -101,6 +101,47 @@ async function renderWorkspace(changes: WorkspaceChangesView) {
   return { dom, root };
 }
 
+async function renderFilesWorkspace(methods: Partial<AppBindings>, props: Partial<Parameters<typeof WorkspacePanel>[0]> = {}) {
+  const dom = installDom();
+  window.go = {
+    main: {
+      App: {
+        ListDir: async () => [],
+        SearchFileRefs: async () => [],
+        WorkspaceGitHistory: async () => [],
+        WorkspaceChanges: async () => ({ files: [], gitAvailable: true }),
+        ...methods,
+      } as Partial<AppBindings> as AppBindings,
+    },
+  };
+  const rootEl = document.getElementById("root");
+  if (!rootEl) throw new Error("missing root");
+  const root = createRoot(rootEl);
+  let currentProps: Parameters<typeof WorkspacePanel>[0] = {
+    open: true,
+    tabId: "tab-a",
+    cwd: "/repo",
+    maximized: false,
+    initialViewMode: "files",
+    onClose: () => {},
+    onToggleMaximized: () => {},
+    ...props,
+  };
+  const rerender = async (nextProps: Partial<Parameters<typeof WorkspacePanel>[0]> = {}) => {
+    currentProps = { ...currentProps, ...nextProps };
+    await act(async () => {
+      root.render(
+        <LocaleProvider>
+          <WorkspacePanel {...currentProps} />
+        </LocaleProvider>,
+      );
+      await flushPromises();
+    });
+  };
+  await rerender();
+  return { dom, root, rerender };
+}
+
 console.log("\nworkspace changes git errors");
 
 {
@@ -151,5 +192,28 @@ console.log("\nworkspace changes git errors");
   dom.window.close();
 }
 
+{
+  const calls: string[] = [];
+  const listDir = async (dir: string): Promise<DirEntry[]> => {
+    calls.push(dir);
+    return [];
+  };
+  const { dom, root, rerender } = await renderFilesWorkspace(
+    { ListDir: listDir },
+    { fileListRequest: { id: 1, paths: ["src/app.ts"] } },
+  );
+
+  await waitFor("initial referenced file dirs", () => calls.filter((dir) => dir === "src/").length === 1);
+  await rerender({ fileListRequest: { id: 2, paths: ["src/app.ts"] } });
+  await waitFor("referenced file dirs revalidated", () => calls.filter((dir) => dir === "src/").length === 2);
+
+  ok(calls.filter((dir) => dir === "src/").length === 2, "workspace file tree revalidates cached directories for repeated file-list requests");
+
+  await act(async () => {
+    root.unmount();
+  });
+  dom.window.close();
+}
+
 console.log(`\n${passed} passed, ${failed} failed, ${passed + failed} total`);
 if (failed > 0) process.exit(1);

desktop/frontend/src/components/AppChrome.tsx

@@ -58,6 +58,7 @@ export function AppChrome({
 }: AppChromeProps) {
   const t = useT();
   const darwinChrome = platform === "darwin";
+  const titlebarDragRail = darwinChrome || platform === "windows";
   const chromeClassName = [
     "app-chrome",
     "app-chrome--tabs",
@@ -90,7 +91,7 @@ export function AppChrome({
           <span />
         </div>
       )}
-      {darwinChrome && <span className="app-chrome__drag-rail" aria-hidden="true" />}
+      {titlebarDragRail && <span className="app-chrome__drag-rail" aria-hidden="true" />}
       <button
         className={[
           "app-chrome__panel-toggle",

desktop/frontend/src/components/ApprovalModal.tsx

@@ -27,6 +27,10 @@ function animateShelfExit(
   options.onComplete();
 }
 
+function requiresFreshHumanApproval(tool: string): boolean {
+  return tool === "remember" || tool === "forget" || tool === "exit_plan_mode";
+}
+
 export function ApprovalModal({
   approval,
   onAnswer,
@@ -48,6 +52,9 @@ export function ApprovalModal({
 }) {
   const t = useT();
   const isPlanApproval = approval.tool === "exit_plan_mode";
+  const isPlanModeReadOnlyCommand = approval.tool === "plan_mode_read_only_command";
+  const toolLabel = isPlanModeReadOnlyCommand ? "bash" : approval.tool;
+  const isFreshHumanApproval = requiresFreshHumanApproval(approval.tool);
   const subject = approval.subject.trim();
   const subjectSummary = subject.split(/\r?\n/).find((line) => line.trim())?.trim() ?? "";
   const toolMeta = approval.reason?.trim() || subjectSummary || approval.tool;
@@ -93,9 +100,10 @@ export function ApprovalModal({
 
   const chooseToolAction = (key: string) => {
     if (key === "1") answerWithExit(() => onAnswer(true, false, false));
-    else if (key === "2") answerWithExit(() => onAnswer(true, true, false));
-    else if (key === "3") answerWithExit(() => onAnswer(true, true, true));
-    else if (key === "4") answerWithExit(() => onAnswer(false, false, false));
+    else if (isFreshHumanApproval && key === "4") answerWithExit(() => onAnswer(false, false, false));
+    else if (!isFreshHumanApproval && key === "2") answerWithExit(() => onAnswer(true, true, false));
+    else if (!isFreshHumanApproval && key === "3") answerWithExit(() => onAnswer(true, true, true));
+    else if (!isFreshHumanApproval && key === "4") answerWithExit(() => onAnswer(false, false, false));
     else if (key === "Escape") answerWithExit(onStop);
   };
 
@@ -108,7 +116,7 @@ export function ApprovalModal({
     playAttentionChime();
   }, [approval.id, isPlanApproval, showToolDetailsByDefault]);
 
-  const actionCount = isPlanApproval ? 3 : 4;
+  const actionCount = isPlanApproval ? 3 : isFreshHumanApproval ? 2 : 4;
   const selectedIndexRef = useRef(selectedIndex);
   selectedIndexRef.current = selectedIndex;
 
@@ -129,6 +137,7 @@ export function ApprovalModal({
         event.preventDefault();
         const key = String(selectedIndexRef.current + 1);
         if (isPlanApproval) choosePlanAction(key);
+        else if (isFreshHumanApproval) chooseToolAction(selectedIndexRef.current === 0 ? "1" : "4");
         else chooseToolAction(key);
       } else if (event.key === "1" || event.key === "2" || event.key === "3" || event.key === "4" || event.key === "Escape") {
         event.preventDefault();
@@ -138,7 +147,7 @@ export function ApprovalModal({
     };
     document.addEventListener("keydown", onKeyDown);
     return () => document.removeEventListener("keydown", onKeyDown);
-  }, [isPlanApproval, onAnswer, onExitPlan, onStop, actionCount]);
+  }, [isPlanApproval, isFreshHumanApproval, onAnswer, onExitPlan, onStop, actionCount]);
 
   useEffect(() => {
     if (revisionOpen) {
@@ -291,7 +300,7 @@ export function ApprovalModal({
         barRef={cardRef}
         titleId="tool-approval-title"
         title={t("approval.toolPending")}
-        badges={<PromptBadge>{approval.tool}</PromptBadge>}
+        badges={<PromptBadge>{toolLabel}</PromptBadge>}
         meta={toolMeta}
         headerActions={
           <>
@@ -308,9 +317,15 @@ export function ApprovalModal({
         actions={
           <>
             <PromptAction keyLabel="1" label={t("approval.allowOnce")} onClick={() => answerWithExit(() => onAnswer(true, false, false))} selected={selectedIndex === 0} />
-            <PromptAction keyLabel="2" label={t("approval.allowRuleSession")} onClick={() => answerWithExit(() => onAnswer(true, true, false))} selected={selectedIndex === 1} />
-            <PromptAction keyLabel="3" label={t("approval.allowRulePersistent")} onClick={() => answerWithExit(() => onAnswer(true, true, true))} selected={selectedIndex === 2} />
-            <PromptAction keyLabel="4" label={t("approval.deny")} onClick={() => answerWithExit(() => onAnswer(false, false, false))} selected={selectedIndex === 3} />
+            {isFreshHumanApproval ? (
+              <PromptAction keyLabel="4" label={t("approval.deny")} onClick={() => answerWithExit(() => onAnswer(false, false, false))} selected={selectedIndex === 1} />
+            ) : (
+              <>
+                <PromptAction keyLabel="2" label={t("approval.allowRuleSession")} onClick={() => answerWithExit(() => onAnswer(true, true, false))} selected={selectedIndex === 1} />
+                <PromptAction keyLabel="3" label={t("approval.allowRulePersistent")} onClick={() => answerWithExit(() => onAnswer(true, true, true))} selected={selectedIndex === 2} />
+                <PromptAction keyLabel="4" label={t("approval.deny")} onClick={() => answerWithExit(() => onAnswer(false, false, false))} selected={selectedIndex === 3} />
+              </>
+            )}
           </>
         }
       >

desktop/frontend/src/components/CapabilitiesPanel.tsx

@@ -4,7 +4,7 @@ import { asArray } from "../lib/array";
 import { app, openExternal } from "../lib/bridge";
 import { useT } from "../lib/i18n";
 import { mcpServerLifecycleActions, mcpServerRetryableFromAvailableList } from "../lib/mcpServerLifecycle";
-import type { CapabilitiesView, MCPServerInput, ServerView, SkillRootSkillView, SkillRootView, SkillsSettingsView, SkillView, TabMeta } from "../lib/types";
+import type { CapabilitiesView, MCPServerInput, PluginHookView, PluginInstallOptions, PluginMCPServerView, PluginSkillView, PluginView, ServerView, SkillRootSkillView, SkillRootView, SkillsSettingsView, SkillView, TabMeta } from "../lib/types";
 import { InlineConfirmButton } from "./InlineConfirmButton";
 import { ResizableDrawer } from "./ResizableDrawer";
 import { Tooltip } from "./Tooltip";
@@ -20,6 +20,7 @@ type SettingsSnapshot<T> = { key: string; value: T };
 
 let mcpSettingsSnapshot: SettingsSnapshot<ServerView[]> | null = null;
 let skillsSettingsSnapshot: SettingsSnapshot<SkillsSettingsView> | null = null;
+let pluginsSettingsSnapshot: SettingsSnapshot<PluginView[]> | null = null;
 
 function settingsSnapshotKey(meta: Awaited<ReturnType<typeof app.Meta>> | null | undefined, tabs: TabMeta[] | null | undefined): string {
   const active = tabs?.find((tab) => tab.active);
@@ -50,7 +51,7 @@ export function CapabilitiesPanel({
   const [expandedServerTools, setExpandedServerTools] = useState<Set<string>>(() => new Set());
 
   const reload = useCallback(async () => {
-    setView(normalizeCapabilitiesView(await app.Capabilities().catch(() => ({ servers: [], skills: [], skillRoots: [] }))));
+    setView(normalizeCapabilitiesView(await app.Capabilities().catch(() => ({ servers: [], skills: [], skillRoots: [], plugins: [] }))));
   }, []);
   useEffect(() => {
     void reload();
@@ -314,6 +315,7 @@ export function CapabilitiesPanel({
 function normalizeCapabilitiesView(view: CapabilitiesView | null | undefined): CapabilitiesView {
   return {
     servers: normalizeServerViews(view?.servers),
+    plugins: asArray(view?.plugins),
     ...normalizeSkillsSettingsView(view),
   };
 }
@@ -1590,6 +1592,619 @@ function AddServerForm({
   );
 }
 
+type PluginInstallPlanAction = {
+  action?: string;
+  kind?: string;
+  name?: string;
+  source?: string;
+  status?: string;
+  message?: string;
+  error?: string;
+};
+
+type PluginInstallPlanView = {
+  raw: string;
+  ok?: boolean;
+  status?: string;
+  name?: string;
+  actions: PluginInstallPlanAction[];
+  warnings: string[];
+  error?: string;
+};
+
+type PluginInstallMode = "local" | "git";
+
+// PluginsSettingsPage is the desktop plugin package manager embedded inside
+// Settings. It mirrors the MCP/Skills density: install planning on top, package
+// rows below, and diagnostics/details only when a row is expanded.
+export function PluginsSettingsPage() {
+	const t = useT();
+	const [snapshotKey, setSnapshotKey] = useState("");
+	const [plugins, setPlugins] = useState<PluginView[] | null>(null);
+	const [busy, setBusy] = useState(false);
+	const [err, setErr] = useState<string | null>(null);
+	const [installMode, setInstallMode] = useState<PluginInstallMode>("local");
+	const [localSource, setLocalSource] = useState("");
+	const [gitSource, setGitSource] = useState("");
+	const [name, setName] = useState("");
+	const [link, setLink] = useState(false);
+	const [replace, setReplace] = useState(false);
+	const [plan, setPlan] = useState<PluginInstallPlanView | null>(null);
+	const [notice, setNotice] = useState<string | null>(null);
+	const [expanded, setExpanded] = useState<Set<string>>(() => new Set());
+	const [diagnostics, setDiagnostics] = useState<Record<string, PluginView>>({});
+
+	const reload = useCallback(async () => {
+		const [meta, tabs] = await Promise.all([
+			app.Meta().catch(() => null),
+			app.ListTabs().catch(() => []),
+		]);
+		const key = settingsSnapshotKey(meta, tabs);
+		setSnapshotKey(key);
+		const cached = key ? pluginsSettingsSnapshot : null;
+		if (cached?.key === key) {
+			setPlugins(cached.value);
+		} else {
+			setPlugins(null);
+		}
+		const next = normalizePluginViews(await app.Plugins().catch(() => []));
+		pluginsSettingsSnapshot = { key, value: next };
+		setPlugins(next);
+	}, []);
+	useEffect(() => { void reload(); }, [reload]);
+
+	const run = async (fn: () => Promise<unknown>, reloadAfter = true) => {
+		setBusy(true);
+		setErr(null);
+		setNotice(null);
+		try {
+			const result = await fn();
+			if (typeof result === "string" && result.trim()) {
+				const parsed = parsePluginInstallPlan(result);
+				setNotice(pluginPlanNotice(parsed, t));
+			}
+			if (reloadAfter) await reload();
+			return true;
+		} catch (e) {
+			setErr(String((e as Error)?.message ?? e));
+			if (reloadAfter) await reload();
+			return false;
+		} finally {
+			setBusy(false);
+		}
+	};
+
+	const sourceValue = (installMode === "local" ? localSource : gitSource).trim();
+	const installOptions = (): PluginInstallOptions => ({
+		dryRun: false,
+		link: installMode === "local" ? link : false,
+		replace,
+		name: installMode === "git" ? name.trim() || undefined : undefined,
+	});
+	const actionBusy = busy || !snapshotKey || !plugins;
+	const canPlan = sourceValue.length > 0 && !actionBusy;
+	const summary = plugins ? pluginListSummary(plugins, t) : "";
+	const togglePlugin = useCallback((pluginName: string) => {
+		setExpanded((prev) => { const next = new Set(prev); if (next.has(pluginName)) next.delete(pluginName); else next.add(pluginName); return next; });
+	}, []);
+	const setMode = (mode: PluginInstallMode) => {
+		setInstallMode(mode);
+		setPlan(null);
+	};
+	const previewInstall = () => {
+		if (!sourceValue) return;
+		void run(async () => {
+			const raw = await app.PlanPluginInstall(sourceValue, { ...installOptions(), dryRun: true });
+			setPlan(parsePluginInstallPlan(raw));
+		}, false);
+	};
+	const install = () => {
+		if (!sourceValue) return;
+		void run(async () => {
+			const raw = await app.InstallPlugin(sourceValue, installOptions());
+			setPlan(parsePluginInstallPlan(raw));
+			return raw;
+		});
+	};
+	const runDoctor = (pluginName: string) => {
+		void run(async () => {
+			const view = normalizePluginView(await app.PluginDoctor(pluginName));
+			setDiagnostics((prev) => ({ ...prev, [pluginName]: view }));
+			setExpanded((prev) => {
+				const next = new Set(prev);
+				next.add(pluginName);
+				return next;
+			});
+		}, false);
+	};
+	const updateLocalSource = (value: string) => {
+		setLocalSource(value);
+		setPlan(null);
+	};
+	const updateGitSource = (value: string) => {
+		setGitSource(value);
+		setPlan(null);
+	};
+	const pickPluginFolder = () => {
+		void run(async () => {
+			const path = await app.PickPluginFolder();
+			if (path) {
+				setInstallMode("local");
+				updateLocalSource(path);
+			}
+		}, false);
+	};
+
+	return (
+		<section className="mem-section">
+			{err && <div className="banner banner--error">{err}</div>}
+			{notice && !err && <div className="banner banner--success">{notice}</div>}
+			<div className="cap-plugin-installer">
+				<div className="cap-plugin-installer__head">
+					<div className="cap-plugin-installer__copy">
+						<div className="cap-plugin-installer__title">{t("caps.pluginInstallTitle")}</div>
+						<div className="cap-plugin-installer__hint">{t("caps.pluginInstallHint")}</div>
+					</div>
+					<div className="cap-tabs cap-plugin-installer__mode" role="group" aria-label={t("caps.pluginInstallMethod")}>
+						<button
+							className={`cap-tab${installMode === "local" ? " cap-tab--active" : ""}`}
+							type="button"
+							aria-pressed={installMode === "local"}
+							onClick={() => setMode("local")}
+						>
+							{t("caps.pluginInstallLocal")}
+						</button>
+						<button
+							className={`cap-tab${installMode === "git" ? " cap-tab--active" : ""}`}
+							type="button"
+							aria-pressed={installMode === "git"}
+							onClick={() => setMode("git")}
+						>
+							{t("caps.pluginInstallGit")}
+						</button>
+					</div>
+				</div>
+				<div className="cap-plugin-form-grid">
+					{installMode === "local" ? (
+						<div className="cap-plugin-fields cap-plugin-fields--local">
+							<div className="cap-plugin-folder-field">
+								<button className="btn btn--small" disabled={actionBusy} type="button" onClick={pickPluginFolder}>
+									{t("caps.pluginChooseLocalFolder")}
+								</button>
+								<div
+									className={`cap-plugin-path${localSource ? "" : " cap-plugin-path--empty"}`}
+									aria-label={t("caps.pluginLocalFolder")}
+								>
+									{localSource || t("caps.pluginNoLocalFolder")}
+								</div>
+							</div>
+						</div>
+					) : (
+						<div className="cap-plugin-fields cap-plugin-fields--git">
+							<input
+								className="mem-input"
+								aria-label={t("caps.pluginGitSource")}
+								placeholder={t("caps.pluginSourcePlaceholder")}
+								value={gitSource}
+								onInput={(e) => updateGitSource(e.currentTarget.value)}
+								onChange={(e) => updateGitSource(e.target.value)}
+							/>
+							<div className="cap-plugin-field">
+								<input
+									className="mem-input"
+									aria-label={t("caps.pluginInstallName")}
+									placeholder={t("caps.pluginInstallNamePlaceholder")}
+									value={name}
+									onChange={(e) => setName(e.target.value)}
+								/>
+							</div>
+						</div>
+					)}
+					<div className="cap-plugin-installer__options">
+						<div className="cap-plugin-option-block">
+							<label className="cap-plugin-option">
+								<input type="checkbox" checked={replace} disabled={actionBusy} onChange={(e) => setReplace(e.target.checked)} />
+								<span>{t("caps.pluginReplace")}</span>
+							</label>
+							<div className="cap-plugin-option-hint">{t("caps.pluginReplaceHint")}</div>
+						</div>
+						{installMode === "local" && (
+							<div className="cap-plugin-option-block">
+								<label className="cap-plugin-option">
+									<input type="checkbox" checked={link} disabled={actionBusy} onChange={(e) => setLink(e.target.checked)} />
+									<span>{t("caps.pluginLink")}</span>
+								</label>
+								<div className="cap-plugin-option-hint">{t("caps.pluginLinkHint")}</div>
+							</div>
+						)}
+					</div>
+					<div className="cap-plugin-installer__actions">
+						<button className="btn btn--small" type="button" disabled={!canPlan} onClick={previewInstall}>
+							{t("caps.pluginPreview")}
+						</button>
+						<button className="btn btn--primary btn--small" type="button" disabled={!canPlan} onClick={install}>
+							{t("caps.pluginInstall")}
+						</button>
+					</div>
+				</div>
+			</div>
+			{plan && <PluginPlanPreview plan={plan} />}
+			<div className="cap-server-section cap-plugin-section">
+				<div className="cap-server-section__head">
+					<div className="cap-server-section__copy">
+						<div className="cap-server-section__title">{t("caps.installedPlugins")}</div>
+						{plugins && plugins.length > 0 && <div className="drawer__summary">{summary}</div>}
+					</div>
+					<button className="btn btn--small" disabled={actionBusy} type="button" onClick={() => void reload()}>
+						{t("caps.pluginRefresh")}
+					</button>
+				</div>
+				{!plugins ? (
+					<div className="mem-empty">{t("caps.loading")}</div>
+				) : plugins.length === 0 ? (
+					<div className="mem-empty mem-empty--cta">
+						<strong>{t("caps.noPluginsTitle")}</strong>
+						<span>{t("caps.noPluginsHint")}</span>
+					</div>
+				) : (
+					<div className="cap-server-group">
+						{plugins.map((plugin) => (
+							<PluginRow
+								key={plugin.name}
+								plugin={plugin}
+								diagnostic={diagnostics[plugin.name]}
+								busy={actionBusy}
+								expanded={expanded.has(plugin.name)}
+								onToggleDetails={() => togglePlugin(plugin.name)}
+								onToggleEnabled={(enabled) => void run(() => app.SetPluginEnabled(plugin.name, enabled))}
+								onUpdate={() => void run(() => app.UpdatePlugin(plugin.name))}
+								onDoctor={() => runDoctor(plugin.name)}
+								onRemove={() => void run(() => app.RemovePlugin(plugin.name))}
+							/>
+						))}
+					</div>
+				)}
+			</div>
+		</section>
+	);
+}
+
+function PluginPlanPreview({ plan }: { plan: PluginInstallPlanView }) {
+	const t = useT();
+	return (
+		<div className={`cap-plugin-plan${plan.error ? " cap-plugin-plan--error" : ""}`}>
+			<div className="cap-plugin-plan__head">
+				<div className="cap-plugin-plan__title">{plan.error ? t("caps.pluginPlanError") : t("caps.pluginPlanReady")}</div>
+				{plan.status && <span className="cap-source-badge">{plan.status}</span>}
+			</div>
+			{plan.name && <div className="cap-plugin-plan__meta">{plan.name}</div>}
+			{plan.error && <div className="cap-plugin-plan__warning">{plan.error}</div>}
+			{plan.warnings.map((warning, idx) => (
+				<div className="cap-plugin-plan__warning" key={`${warning}-${idx}`}>{warning}</div>
+			))}
+			{plan.actions.length > 0 ? (
+				<div className="cap-plugin-actions">
+					{plan.actions.map((action, idx) => (
+						<div className="cap-plugin-action" key={`${action.action || action.kind || "action"}-${idx}`}>
+							<span className="cap-plugin-action__name">{pluginPlanActionLabel(action, t)}</span>
+							{action.status && <span className="cap-source-badge">{action.status}</span>}
+							{action.source && <span className="cap-plugin-action__source">{action.source}</span>}
+							{action.message && <span className="cap-plugin-action__source">{action.message}</span>}
+							{action.error && <span className="cap-plugin-plan__warning">{action.error}</span>}
+						</div>
+					))}
+				</div>
+			) : (
+				<pre className="cap-plugin-plan__raw">{plan.raw}</pre>
+			)}
+		</div>
+	);
+}
+
+function PluginRow({
+	plugin,
+	diagnostic,
+	busy,
+	expanded,
+	onToggleDetails,
+	onToggleEnabled,
+	onUpdate,
+	onDoctor,
+	onRemove,
+}: {
+	plugin: PluginView;
+	diagnostic?: PluginView;
+	busy: boolean;
+	expanded: boolean;
+	onToggleDetails: () => void;
+	onToggleEnabled: (enabled: boolean) => void;
+	onUpdate: () => void;
+	onDoctor: () => void;
+	onRemove: () => void;
+}) {
+	const t = useT();
+	const status = plugin.error ? "failed" : plugin.enabled ? "connected" : "disabled";
+	const warnings = pluginWarnings(plugin, diagnostic);
+	const sub = plugin.error || pluginCapabilitiesSummary(plugin, t);
+	return (
+		<div className={`cap-server-entry cap-plugin-entry${plugin.enabled ? "" : " cap-server-entry--disabled"}`}>
+			<Tooltip label={plugin.error} disabled={!plugin.error} fill block>
+				<div className={`cap-row${plugin.enabled ? "" : " cap-row--disabled"}`}>
+					<Tooltip label={expanded ? t("caps.collapseDetails") : t("caps.expandDetails")}>
+						<button
+							className="cap-disclosure"
+							aria-expanded={expanded}
+							type="button"
+							onClick={onToggleDetails}
+						>
+							{expanded ? "⌄" : "›"}
+						</button>
+					</Tooltip>
+					<span className={`cap-dot cap-dot--${status}`} />
+					<div className="cap-row__text">
+						<div className="cap-row__head">
+							<span className="cap-row__name">{plugin.name}</span>
+							{plugin.manifestKind && <span className="cap-row__transport">{plugin.manifestKind}</span>}
+							{plugin.version && <span className="cap-source-badge">{plugin.version}</span>}
+							{warnings.length > 0 && <span className="cap-row__update cap-row__update--error">{t("caps.pluginWarnings", { count: warnings.length })}</span>}
+						</div>
+						<div className="cap-row__sub">{sub}</div>
+					</div>
+					<div className="cap-row__actions">
+						<Tooltip label={plugin.enabled ? t("caps.pluginDisable") : t("caps.pluginEnable")}>
+							<label className="cap-switch">
+								<input
+									type="checkbox"
+									checked={plugin.enabled}
+									disabled={busy}
+									onChange={(e) => onToggleEnabled(e.target.checked)}
+								/>
+								<span className="cap-switch__track" />
+							</label>
+						</Tooltip>
+					</div>
+				</div>
+			</Tooltip>
+			{expanded && (
+				<div className="cap-server-details">
+					<div className="cap-detail-grid">
+						<div className="cap-detail">
+							<span className="cap-detail__label">{t("caps.status")}</span>
+							<span className="cap-detail__value">{plugin.enabled ? t("caps.pluginEnabled") : t("caps.pluginDisabled")}</span>
+						</div>
+						{plugin.version && (
+							<div className="cap-detail">
+								<span className="cap-detail__label">{t("caps.pluginVersion")}</span>
+								<span className="cap-detail__value">{plugin.version}</span>
+							</div>
+						)}
+						{plugin.source && (
+							<div className="cap-detail cap-detail--wide">
+								<span className="cap-detail__label">{t("caps.pluginSource")}</span>
+								<span className="cap-detail__code">{plugin.source}</span>
+							</div>
+						)}
+						{plugin.root && (
+							<div className="cap-detail cap-detail--wide">
+								<span className="cap-detail__label">{t("caps.pluginRoot")}</span>
+								<span className="cap-detail__code">{plugin.root}</span>
+							</div>
+						)}
+					</div>
+					{plugin.description && <div className="cap-plugin-description">{plugin.description}</div>}
+					<PluginUsageDetails plugin={plugin} />
+					{diagnostic?.error && <div className="cap-source__warning">{diagnostic.error}</div>}
+					{warnings.map((warning, idx) => (
+						<div className="cap-source__warning" key={`${plugin.name}-warning-${idx}`}>{warning}</div>
+					))}
+					<div className="cap-detail-actions">
+						<button className="btn btn--small" disabled={busy} type="button" onClick={onUpdate}>
+							{t("caps.pluginUpdate")}
+						</button>
+						<button className="btn btn--small" disabled={busy} type="button" onClick={onDoctor}>
+							{t("caps.pluginDoctor")}
+						</button>
+						<InlineConfirmButton
+							label={t("caps.pluginRemove")}
+							confirmLabel={t("caps.pluginConfirmRemove")}
+							cancelLabel={t("common.cancel")}
+							disabled={busy}
+							danger
+							onConfirm={onRemove}
+						/>
+					</div>
+				</div>
+			)}
+		</div>
+	);
+}
+
+function PluginUsageDetails({ plugin }: { plugin: PluginView }) {
+	const t = useT();
+	const skills = asArray(plugin.skillDetails);
+	const hooks = asArray(plugin.hookDetails);
+	const mcps = asArray(plugin.mcpServerDetails);
+	const hasDetails = skills.length > 0 || hooks.length > 0 || mcps.length > 0;
+	return (
+		<div className="cap-plugin-usage">
+			<div className="cap-plugin-usage__title">{t("caps.pluginUsageTitle")}</div>
+			<div className="cap-plugin-usage__hint">
+				{plugin.enabled ? t("caps.pluginUsageEnabledHint") : t("caps.pluginUsageDisabledHint")}
+			</div>
+			{hasDetails ? (
+				<div className="cap-plugin-capabilities">
+					{skills.length > 0 && <PluginSkillList skills={skills} />}
+					{hooks.length > 0 && <PluginHookList hooks={hooks} />}
+					{mcps.length > 0 && <PluginMCPList servers={mcps} />}
+				</div>
+			) : (
+				<div className="cap-plugin-usage__empty">{t("caps.pluginNoCapabilityDetails")}</div>
+			)}
+		</div>
+	);
+}
+
+function PluginSkillList({ skills }: { skills: PluginSkillView[] }) {
+	const t = useT();
+	return (
+		<div className="cap-plugin-capability">
+			<div className="cap-plugin-capability__head">{t("caps.pluginSkillsTitle")}</div>
+			<div className="cap-plugin-capability__hint">{t("caps.pluginSkillsHint")}</div>
+			<div className="cap-plugin-capability__list">
+				{skills.map((skill) => (
+					<div className="cap-plugin-capability__item" key={`${skill.name}-${skill.path || skill.invocation || ""}`}>
+						<div className="cap-plugin-capability__line">
+							<span className="cap-plugin-capability__name">{skill.invocation || `/${skill.name}`}</span>
+							{skill.runAs && <span className="cap-source-badge">{skill.runAs}</span>}
+						</div>
+						<div className="cap-plugin-capability__desc">{skill.description || t("caps.pluginNoDescription")}</div>
+					</div>
+				))}
+			</div>
+		</div>
+	);
+}
+
+function PluginHookList({ hooks }: { hooks: PluginHookView[] }) {
+	const t = useT();
+	return (
+		<div className="cap-plugin-capability">
+			<div className="cap-plugin-capability__head">{t("caps.pluginHooksTitle")}</div>
+			<div className="cap-plugin-capability__hint">{t("caps.pluginHooksHint")}</div>
+			<div className="cap-plugin-capability__list">
+				{hooks.map((hook, idx) => {
+					const target = hook.command || hook.contextFile || t("caps.pluginHookNoTarget");
+					return (
+						<div className="cap-plugin-capability__item" key={`${hook.event}-${hook.match || "*"}-${target}-${idx}`}>
+							<div className="cap-plugin-capability__line">
+								<span className="cap-plugin-capability__name">{hook.event}</span>
+								<span className="cap-source-badge">{hook.match || "*"}</span>
+							</div>
+							<div className="cap-plugin-capability__desc">{hook.description || target}</div>
+						</div>
+					);
+				})}
+			</div>
+		</div>
+	);
+}
+
+function PluginMCPList({ servers }: { servers: PluginMCPServerView[] }) {
+	const t = useT();
+	return (
+		<div className="cap-plugin-capability">
+			<div className="cap-plugin-capability__head">{t("caps.pluginMCPTitle")}</div>
+			<div className="cap-plugin-capability__hint">{t("caps.pluginMCPHint")}</div>
+			<div className="cap-plugin-capability__list">
+				{servers.map((server) => (
+					<div className="cap-plugin-capability__item" key={server.name}>
+						<div className="cap-plugin-capability__line">
+							<span className="cap-plugin-capability__name">{server.name}</span>
+							{server.transport && <span className="cap-source-badge">{server.transport}</span>}
+						</div>
+						<div className="cap-plugin-capability__desc">{server.command || server.url || t("caps.pluginMCPNoTarget")}</div>
+					</div>
+				))}
+			</div>
+		</div>
+	);
+}
+
+function normalizePluginViews(plugins: PluginView[] | null | undefined): PluginView[] {
+	return sortPluginsForDisplay(asArray(plugins).map(normalizePluginView));
+}
+
+function normalizePluginView(plugin: PluginView): PluginView {
+	return {
+		...plugin,
+		name: plugin.name || "plugin",
+		root: plugin.root || "",
+		enabled: Boolean(plugin.enabled),
+		skills: Number.isFinite(plugin.skills) ? plugin.skills : 0,
+		hooks: Number.isFinite(plugin.hooks) ? plugin.hooks : 0,
+		mcpServers: Number.isFinite(plugin.mcpServers) ? plugin.mcpServers : 0,
+		skillDetails: asArray(plugin.skillDetails),
+		hookDetails: asArray(plugin.hookDetails),
+		mcpServerDetails: asArray(plugin.mcpServerDetails),
+		warnings: asArray(plugin.warnings),
+	};
+}
+
+function sortPluginsForDisplay(plugins: PluginView[]): PluginView[] {
+	return [...plugins].sort((a, b) => {
+		const priority = pluginDisplayPriority(a) - pluginDisplayPriority(b);
+		if (priority !== 0) return priority;
+		return a.name.localeCompare(b.name, undefined, { sensitivity: "base" });
+	});
+}
+
+function pluginDisplayPriority(plugin: PluginView): number {
+	if (plugin.error) return 0;
+	if (plugin.enabled) return 1;
+	return 2;
+}
+
+function pluginListSummary(plugins: PluginView[], t: ReturnType<typeof useT>): string {
+	const enabled = plugins.filter((plugin) => plugin.enabled && !plugin.error).length;
+	const issues = plugins.filter((plugin) => Boolean(plugin.error) || asArray(plugin.warnings).length > 0).length;
+	return t("caps.pluginsSummary", { enabled, total: plugins.length, issues });
+}
+
+function pluginCapabilitiesSummary(plugin: PluginView, t: ReturnType<typeof useT>): string {
+	if (plugin.skills === 0 && plugin.hooks === 0 && plugin.mcpServers === 0) return t("caps.pluginNoCapabilities");
+	return t("caps.pluginCounts", { skills: plugin.skills, hooks: plugin.hooks, mcps: plugin.mcpServers });
+}
+
+function pluginWarnings(plugin: PluginView, diagnostic?: PluginView): string[] {
+	const warnings = [...asArray(plugin.warnings), ...asArray(diagnostic?.warnings)];
+	return Array.from(new Set(warnings.filter((warning) => warning.trim().length > 0)));
+}
+
+function parsePluginInstallPlan(raw: string): PluginInstallPlanView {
+	try {
+		const value = JSON.parse(raw) as Record<string, unknown>;
+		const actions = (Array.isArray(value.actions) ? value.actions : []).flatMap((action) => {
+			if (!action || typeof action !== "object") return [];
+			const item = action as Record<string, unknown>;
+			return [{
+				action: stringValue(item.action),
+				kind: stringValue(item.kind),
+				name: stringValue(item.name),
+				source: stringValue(item.source),
+				status: stringValue(item.status),
+				message: stringValue(item.message),
+				error: stringValue(item.error),
+			}];
+		});
+		return {
+			raw,
+			ok: typeof value.ok === "boolean" ? value.ok : undefined,
+			status: stringValue(value.status),
+			name: stringValue(value.name),
+			actions,
+			warnings: (Array.isArray(value.warnings) ? value.warnings : []).flatMap((warning) => typeof warning === "string" ? [warning] : []),
+			error: stringValue(value.error),
+		};
+	} catch {
+		return { raw, actions: [], warnings: [] };
+	}
+}
+
+function stringValue(value: unknown): string | undefined {
+	return typeof value === "string" && value.trim() ? value.trim() : undefined;
+}
+
+function pluginPlanActionLabel(action: PluginInstallPlanAction, t: ReturnType<typeof useT>): string {
+	const verb = action.action || action.kind || t("caps.pluginAction");
+	return [verb, action.name].filter(Boolean).join(" · ");
+}
+
+function pluginPlanNotice(plan: PluginInstallPlanView, t: ReturnType<typeof useT>): string {
+	if (plan.error) return plan.error;
+	if (plan.status === "done" || plan.status === "applied" || plan.status === "complete") return t("caps.pluginPlanInstalled");
+	return plan.status ? t("caps.pluginPlanStatus", { status: plan.status }) : t("caps.pluginPlanComplete");
+}
+
 // MCPServersSettingsPage is a self-contained MCP servers management page
 // embedded inside the settings centre.
 export function MCPServersSettingsPage() {

desktop/frontend/src/components/Composer.tsx

@@ -26,6 +26,7 @@ import { EffortSwitcher } from "./EffortSwitcher";
 import { ModelSwitcher } from "./ModelSwitcher";
 import { Tooltip } from "./Tooltip";
 import { ComposerContextCard } from "./ComposerContextCard";
+import { ImageViewer } from "./ImageViewer";
 import { VirtualMenu } from "./VirtualMenu";
 import { dirEntryMenuLabel, dirEntrySubmitPath } from "./FileReferenceMenu";
 
@@ -57,6 +58,7 @@ const PROMPT_HISTORY_PREFETCH_REMAINING = 3;
 // it; the real gap is a few ms, so keep it short or a deliberate quick second
 // Enter (submit) gets eaten too.
 const IME_CONFIRM_GRACE_MS = 100;
+const FILE_REF_SEARCH_CACHE_TTL_MS = 5000;
 
 type PastedBlock = {
   label: string;
@@ -69,6 +71,11 @@ type PendingGuidance = {
   submitText: string;
 };
 
+type FileRefSearchCacheEntry = {
+  entries: DirEntry[];
+  cachedAt: number;
+};
+
 type ComposerDraft = {
   text: string;
   attachments: Attachment[];
@@ -444,6 +451,7 @@ export function Composer({
   retry,
   transientDismissSignal,
   sessionKey,
+  fileRefRefreshKey,
   guidanceConsumedKey,
   guidanceConsumedText,
   guidanceQueuePreviewItems,
@@ -486,6 +494,7 @@ export function Composer({
   retry?: { attempt: number; max: number };
   transientDismissSignal?: number;
   sessionKey?: string;
+  fileRefRefreshKey?: number | string;
   guidanceConsumedKey?: string;
   guidanceConsumedText?: string;
   guidanceQueuePreviewItems?: readonly string[];
@@ -496,6 +505,15 @@ export function Composer({
   const now = useTick(running);
   const [text, setText] = useState("");
   const [attachments, setAttachments] = useState<Attachment[]>([]);
+  const [imageViewer, setImageViewer] = useState<{ open: boolean; url: string; name: string }>({ open: false, url: "", name: "" });
+  const openComposerImageViewer = useCallback((url: string, name: string) => {
+    setImageViewer({ open: true, url, name });
+  }, []);
+
+  const closeComposerImageViewer = useCallback(() => {
+    setImageViewer((prev) => (prev.open ? { ...prev, open: false } : prev));
+  }, []);
+
   const [workspaceRefs, setWorkspaceRefs] = useState<WorkspaceReference[]>([]);
   const [pastedBlocks, setPastedBlocks] = useState<PastedBlock[]>([]);
   const [openPastedLabels, setOpenPastedLabels] = useState<string[]>([]);
@@ -743,16 +761,9 @@ export function Composer({
   const [entries, setEntries] = useState<DirEntry[]>([]);
   const [searchEntries, setSearchEntries] = useState<DirEntry[]>([]);
   const dirCache = useRef<Record<string, DirEntry[]>>({});
-  const searchCache = useRef<Record<string, DirEntry[]>>({});
+  const searchCache = useRef<Record<string, FileRefSearchCacheEntry>>({});
 
-  // When the workspace/project changes (cwd prop), invalidate all @ mention
-  // state so the picker reloads candidates for the new project. Without this,
-  // dirCache/searchCache retain entries from the old project and the picker
-  // shows stale results (issue #3601).
-  const prevCwdRef = useRef(cwd);
-  useEffect(() => {
-    if (prevCwdRef.current === cwd) return; // skip mount — state already initial
-    prevCwdRef.current = cwd;
+  const clearFileRefState = useCallback(() => {
     dirCache.current = {};
     searchCache.current = {};
     setEntries([]);
@@ -763,29 +774,50 @@ export function Composer({
     setLoadingPastChats(false);
     setActive(0);
     setDismissed(false);
-  }, [cwd]);
+  }, []);
+
+  // When the workspace/project changes (cwd prop), invalidate all @ mention
+  // state so the picker reloads candidates for the new project. Without this,
+  // dirCache/searchCache retain entries from the old project and the picker
+  // shows stale results (issue #3601).
+  const prevCwdRef = useRef(cwd);
+  useEffect(() => {
+    if (prevCwdRef.current === cwd) return; // skip mount — state already initial
+    prevCwdRef.current = cwd;
+    clearFileRefState();
+  }, [clearFileRefState, cwd]);
+
+  const prevFileRefRefreshKeyRef = useRef(fileRefRefreshKey);
+  useEffect(() => {
+    if (prevFileRefRefreshKeyRef.current === fileRefRefreshKey) return;
+    prevFileRefRefreshKeyRef.current = fileRefRefreshKey;
+    clearFileRefState();
+  }, [clearFileRefState, fileRefRefreshKey]);
 
   useEffect(() => {
     if (atRaw === null) return;
     const cached = dirCache.current[atDir];
     if (cached) {
       setEntries(cached);
-      return;
+    } else {
+      setEntries([]);
     }
     let live = true;
     app
       .ListDir(atDir)
       .then((es) => {
         const list = asArray(es);
+        if (!live) return;
         dirCache.current[atDir] = list;
-        if (live) setEntries(list);
+        setEntries(list);
       })
       .catch(() => {});
     return () => {
       live = false;
     };
-    // re-fetch when the menu opens or the directory level changes
-  }, [atRaw === null, atDir, cwd]);
+    // Re-fetch when the menu opens, the directory level changes, or the
+    // workspace tree refreshes; cached data is only a fast first paint.
+  }, [atRaw === null, atDir, cwd, fileRefRefreshKey]);
   useEffect(() => {
     if (atRaw === null || atDir !== "" || atFrag === "") {
       setSearchEntries([]);
@@ -793,23 +825,25 @@ export function Composer({
     }
     const cached = searchCache.current[atFrag];
     if (cached) {
-      setSearchEntries(cached);
-      return;
+      setSearchEntries(cached.entries);
+      if (Date.now() - cached.cachedAt < FILE_REF_SEARCH_CACHE_TTL_MS) return;
+    } else {
+      setSearchEntries([]);
     }
-    setSearchEntries([]);
     let live = true;
     app
       .SearchFileRefs(atFrag)
       .then((es) => {
-        const list = es ?? [];
-        searchCache.current[atFrag] = list;
-        if (live) setSearchEntries(list);
+        const list = asArray(es);
+        if (!live) return;
+        searchCache.current[atFrag] = { entries: list, cachedAt: Date.now() };
+        setSearchEntries(list);
       })
       .catch(() => {});
     return () => {
       live = false;
     };
-  }, [atRaw === null, atDir, atFrag, cwd]);
+  }, [atRaw === null, atDir, atFrag, cwd, fileRefRefreshKey]);
   const atMatches = useMemo(
     () => {
       if (atRaw === null) return [];
@@ -2362,10 +2396,11 @@ export function Composer({
               <ComposerContextCard
                 key={a.path}
                 variant="attachment"
-                tooltipLabel={a.path}
+                tooltipLabel={a.previewUrl ? `${t("imageViewer.clickToPreview")} — ${a.path}` : a.path}
                 removeLabel={t("composer.removeImage")}
                 onRemove={() => removeAttachment(a.path)}
                 previewUrl={a.previewUrl}
+                onImageClick={a.previewUrl ? () => openComposerImageViewer(a.previewUrl!, attachmentName(a)) : undefined}
                 imageOnly={imageOnly}
                 name={attachmentName(a)}
                 meta={attachmentExt(attachmentName(a)) || t("msg.fileAttachment")}
@@ -2409,6 +2444,12 @@ export function Composer({
           ))}
         </div>
       )}
+      <ImageViewer
+        open={imageViewer.open}
+        imageUrl={imageViewer.url}
+        imageName={imageViewer.name}
+        onClose={closeComposerImageViewer}
+      />
       {activePastedBlocks.length > 0 && (
         <div className="composer__pasted">
           {activePastedBlocks.map((block) => {

desktop/frontend/src/components/ComposerContextCard.tsx

@@ -12,6 +12,7 @@ export function ComposerContextCard({
   removeDisabled = false,
   removeIconSize,
   previewUrl,
+  onImageClick,
   imageOnly = false,
   folder = false,
   name,
@@ -26,6 +27,8 @@ export function ComposerContextCard({
   removeDisabled?: boolean;
   removeIconSize?: number;
   previewUrl?: string;
+  /** called when the image thumbnail is clicked (only for image attachments with previewUrl) */
+  onImageClick?: () => void;
   imageOnly?: boolean;
   folder?: boolean;
   name?: string;
@@ -46,7 +49,13 @@ export function ComposerContextCard({
       <Tooltip label={tooltipLabel}>
         <span className="composer-context__label">
           {previewUrl ? (
-            <span className="composer-context__thumb">
+            <span
+              className={`composer-context__thumb${onImageClick ? " composer-context__thumb--interactive" : ""}`}
+              onClick={onImageClick}
+              role={onImageClick ? "button" : undefined}
+              tabIndex={onImageClick ? 0 : undefined}
+              onKeyDown={onImageClick ? (e) => { if (e.key === "Enter" || e.key === " ") { e.preventDefault(); onImageClick(); } } : undefined}
+            >
               <img src={previewUrl} alt="" draggable={false} />
             </span>
           ) : variant === "attachment" ? (

desktop/frontend/src/components/ContextPanel.tsx

@@ -77,6 +77,22 @@ export function formatCacheHitRate(hitTokens: number, missTokens: number): strin
 
 type MetricTone = "accent" | "good" | "notice" | "warn";
 type UsageAnalysisView = "source" | "type";
+type ContextUsageRefreshFields = Pick<
+  WireUsage,
+  "totalTokens" | "promptTokens" | "completionTokens" | "reasoningTokens" | "sessionCacheHitTokens" | "sessionCacheMissTokens"
+>;
+
+export function contextUsageRefreshKey(usage?: ContextUsageRefreshFields): string {
+  if (!usage) return "";
+  return [
+    usage.totalTokens ?? 0,
+    usage.promptTokens ?? 0,
+    usage.completionTokens ?? 0,
+    usage.reasoningTokens ?? 0,
+    usage.sessionCacheHitTokens ?? 0,
+    usage.sessionCacheMissTokens ?? 0,
+  ].join(":");
+}
 
 export function cacheHitTone(hitTokens: number, missTokens: number): MetricTone | undefined {
   const denom = hitTokens + missTokens;
@@ -287,6 +303,8 @@ export function ContextPanel({
   const [info, setInfo] = useState<ContextPanelInfo | null>(null);
   const [analysisView, setAnalysisView] = useState<UsageAnalysisView>("source");
   const refreshSeq = useRef(0);
+  const lastRefreshTime = useRef(0);
+  const usageRefreshKey = contextUsageRefreshKey(usage);
 
   const refresh = useCallback(async () => {
     if (!tabId) return;
@@ -311,19 +329,23 @@ export function ContextPanel({
     void refresh();
   }, [refresh, refreshKey]);
 
-  const hasPanelUsage = Boolean(
-    (info?.requestCount ?? 0) > 0 ||
-    (info?.promptTokens ?? 0) > 0 ||
-    (info?.completionTokens ?? 0) > 0 ||
-    (info?.totalTokens ?? 0) > 0 ||
-    (info?.reasoningTokens ?? 0) > 0 ||
-    (info?.cacheHitTokens ?? 0) > 0 ||
-    (info?.cacheMissTokens ?? 0) > 0
-  );
+  // Refresh the panel snapshot while usage events stream. The key includes
+  // general token fields so providers without cache telemetry still tick.
+  useEffect(() => {
+    if (!usageRefreshKey) return;
+    const now = Date.now();
+    if (now - lastRefreshTime.current >= 1000) {
+      lastRefreshTime.current = now;
+      void refresh();
+    }
+  }, [usageRefreshKey, refresh]);
+
   const usedTokens = context?.used && context.used > 0 ? context.used : info?.usedTokens ?? 0;
   const windowTokens = context?.window && context.window > 0 ? context.window : info?.windowTokens ?? 0;
-  const promptTokens = hasPanelUsage ? info?.promptTokens ?? 0 : usage?.promptTokens ?? 0;
-  const completionTokens = hasPanelUsage ? info?.completionTokens ?? 0 : usage?.completionTokens ?? 0;
+  // Prefer live usage props (updated in real-time by the reducer during streaming)
+  // over the async-fetched info snapshot (only refreshed on turn_done).
+  const promptTokens = usage?.promptTokens ?? info?.promptTokens ?? 0;
+  const completionTokens = usage?.completionTokens ?? info?.completionTokens ?? 0;
   const totalTokens = info?.totalTokens && info.totalTokens > 0
     ? info.totalTokens
     : sessionTokens && sessionTokens > 0
@@ -331,10 +353,12 @@ export function ContextPanel({
       : usage?.totalTokens && usage.totalTokens > 0
         ? usage.totalTokens
         : promptTokens + completionTokens;
-  const reasoningTokens = hasPanelUsage ? info?.reasoningTokens ?? 0 : usage?.reasoningTokens ?? 0;
+  const reasoningTokens = usage?.reasoningTokens ?? info?.reasoningTokens ?? 0;
   // Session-cumulative values for the top summary.
-  const sessionCacheHit = info?.sessionCacheHitTokens ?? usage?.sessionCacheHitTokens ?? context?.cacheHitTokens ?? 0;
-  const sessionCacheMiss = info?.sessionCacheMissTokens ?? usage?.sessionCacheMissTokens ?? context?.cacheMissTokens ?? 0;
+  // Prefer usage.sessionCacheHitTokens — it is the session-cumulative value from
+  // the Go agent (includes background tasks) and arrives with every usage event.
+  const sessionCacheHit = usage?.sessionCacheHitTokens ?? info?.sessionCacheHitTokens ?? context?.cacheHitTokens ?? 0;
+  const sessionCacheMiss = usage?.sessionCacheMissTokens ?? info?.sessionCacheMissTokens ?? context?.cacheMissTokens ?? 0;
   const totalTokensMetric = formatMetricTokens(totalTokens, locale);
   const cost = contextCostDisplay({ info, sessionCost, sessionCurrency, usage });
   const sourceUsageRows = contextSourceRows(info, sessionCurrency);

desktop/frontend/src/components/FileReferenceMenu.tsx

@@ -6,6 +6,13 @@ import { app } from "../lib/bridge";
 import type { DirEntry } from "../lib/types";
 import { VirtualMenu } from "./VirtualMenu";
 
+const FILE_REF_SEARCH_CACHE_TTL_MS = 5000;
+
+type FileRefSearchCacheEntry = {
+  entries: DirEntry[];
+  cachedAt: number;
+};
+
 export function dirEntrySubmitPath(entry: DirEntry, atDir: string): string {
   return entry.path || atDir + entry.name;
 }
@@ -58,7 +65,7 @@ export function useFileReferenceMenu(text: string, cwd?: string) {
   const [active, setActive] = useState(0);
   const [dismissed, setDismissed] = useState(false);
   const dirCache = useRef<Record<string, DirEntry[]>>({});
-  const searchCache = useRef<Record<string, DirEntry[]>>({});
+  const searchCache = useRef<Record<string, FileRefSearchCacheEntry>>({});
   const prevCwdRef = useRef(cwd);
 
   useEffect(() => {
@@ -82,15 +89,17 @@ export function useFileReferenceMenu(text: string, cwd?: string) {
     const cached = dirCache.current[atDir];
     if (cached) {
       setEntries(cached);
-      return;
+    } else {
+      setEntries([]);
     }
     let live = true;
     app
       .ListDir(atDir)
       .then((next) => {
         const list = asArray(next);
+        if (!live) return;
         dirCache.current[atDir] = list;
-        if (live) setEntries(list);
+        setEntries(list);
       })
       .catch(() => {});
     return () => {
@@ -105,17 +114,19 @@ export function useFileReferenceMenu(text: string, cwd?: string) {
     }
     const cached = searchCache.current[atFrag];
     if (cached) {
-      setSearchEntries(cached);
-      return;
+      setSearchEntries(cached.entries);
+      if (Date.now() - cached.cachedAt < FILE_REF_SEARCH_CACHE_TTL_MS) return;
+    } else {
+      setSearchEntries([]);
     }
-    setSearchEntries([]);
     let live = true;
     app
       .SearchFileRefs(atFrag)
       .then((next) => {
-        const list = next ?? [];
-        searchCache.current[atFrag] = list;
-        if (live) setSearchEntries(list);
+        const list = asArray(next);
+        if (!live) return;
+        searchCache.current[atFrag] = { entries: list, cachedAt: Date.now() };
+        setSearchEntries(list);
       })
       .catch(() => {});
     return () => {

desktop/frontend/src/components/ImageViewer.tsx

@@ -0,0 +1,99 @@
+import { useCallback, useEffect, useState } from "react";
+import { createPortal } from "react-dom";
+import { X } from "lucide-react";
+import { useT } from "../lib/i18n";
+
+export interface ImageViewerProps {
+  open: boolean;
+  /** data-URL of the image to display */
+  imageUrl: string;
+  /** optional filename shown at the bottom */
+  imageName?: string;
+  onClose: () => void;
+}
+
+/**
+ * ImageViewer renders a full-size image preview in a portal overlay.
+ *
+ * It follows the same portal pattern as MermaidDiagram fullscreen:
+ * the overlay is portaled into .chat-pane (fallback document.body) so it
+ * covers the chat area without being clipped by overflow containers.
+ */
+export function ImageViewer({ open, imageUrl, imageName, onClose }: ImageViewerProps) {
+  const t = useT();
+  const [portalTarget, setPortalTarget] = useState<Element | null>(null);
+  const [visible, setVisible] = useState(false);
+
+  // Resolve portal target when opening.
+  useEffect(() => {
+    if (!open) {
+      setVisible(false);
+      setPortalTarget(null);
+      return;
+    }
+    const target = document.querySelector(".chat-pane") ?? document.body;
+    setPortalTarget(target);
+    // Trigger enter animation on the next frame.
+    const raf = requestAnimationFrame(() => setVisible(true));
+    return () => cancelAnimationFrame(raf);
+  }, [open]);
+
+  // Close on Escape.
+  useEffect(() => {
+    if (!open) return;
+    const onKeyDown = (event: KeyboardEvent) => {
+      if (event.key === "Escape") onClose();
+    };
+    document.addEventListener("keydown", onKeyDown);
+    return () => document.removeEventListener("keydown", onKeyDown);
+  }, [open, onClose]);
+
+  // Prevent body scroll while open.
+  useEffect(() => {
+    if (!open) return;
+    const prev = document.body.style.overflow;
+    document.body.style.overflow = "hidden";
+    return () => {
+      document.body.style.overflow = prev;
+    };
+  }, [open]);
+
+  const handleBackdropClick = useCallback(
+    (e: React.MouseEvent) => {
+      if (e.target === e.currentTarget) onClose();
+    },
+    [onClose],
+  );
+
+  if (!open || !portalTarget || !imageUrl) return null;
+
+  const overlay = (
+    <div
+      className={`image-viewer-backdrop${visible ? " image-viewer--enter" : ""}`}
+      onClick={handleBackdropClick}
+      role="dialog"
+      aria-modal="true"
+      aria-label={imageName || t("imageViewer.title")}
+    >
+      <button
+        className="image-viewer__close"
+        type="button"
+        onClick={onClose}
+        aria-label={t("imageViewer.closePreview")}
+      >
+        <X size={22} />
+      </button>
+      <div className="image-viewer__content">
+        <img
+          className="image-viewer__image"
+          src={imageUrl}
+          alt={imageName || ""}
+          draggable={false}
+        />
+        {imageName && <div className="image-viewer__name">{imageName}</div>}
+      </div>
+    </div>
+  );
+
+  return createPortal(overlay, portalTarget);
+}

desktop/frontend/src/components/Message.tsx

@@ -1,4 +1,4 @@
-import { memo, useEffect, useRef, useState } from "react";
+import { memo, useCallback, useEffect, useMemo, useRef, useState } from "react";
 import type { FormEvent, KeyboardEvent as ReactKeyboardEvent } from "react";
 import { BrainCircuit, ChevronDown, ChevronRight, FileText, Folder, GitBranch, Image, MessageSquare, Pencil, RotateCcw, ScrollText } from "lucide-react";
 import { Markdown } from "./Markdown";
@@ -10,6 +10,7 @@ import type { DisplayAttachment } from "../lib/attachmentDisplay";
 import { app } from "../lib/bridge";
 import { replaySubmitText } from "../lib/editReplay";
 import { useT } from "../lib/i18n";
+import { ImageViewer } from "./ImageViewer";
 import { Tooltip } from "./Tooltip";
 import { useGSAPCollapse } from "../lib/useGSAPCollapse";
 import { displayReasoningText } from "../lib/reasoningDisplay";
@@ -81,6 +82,31 @@ function mergeDisplayAttachments(existing: DisplayAttachment[], incoming: Displa
   return merged;
 }
 
+type PastedBlockInfo = {
+  label: string;
+  content: string;
+};
+
+const PASTE_LABEL_RE = /\[(?:已粘贴文本|已貼上文字|Pasted text) #\d+ · \d+ (?:行|lines)\]/g;
+
+export function parsePastedBlocks(text: string, submitText?: string): PastedBlockInfo[] {
+  const labels = text.match(PASTE_LABEL_RE);
+  if (!labels || labels.length === 0 || !submitText) return [];
+  const unique = [...new Set(labels)];
+  const blocks: PastedBlockInfo[] = [];
+  for (const label of unique) {
+    const beginMarker = `--- Begin ${label} ---`;
+    const endMarker = `--- End ${label} ---`;
+    const beginIdx = submitText.indexOf(beginMarker);
+    const endIdx = submitText.indexOf(endMarker);
+    if (beginIdx < 0 || endIdx <= beginIdx) continue;
+    const contentStart = beginIdx + beginMarker.length;
+    const content = submitText.slice(contentStart, endIdx).replace(/^\r?\n/, "");
+    blocks.push({ label, content });
+  }
+  return blocks;
+}
+
 function MemoryCitations({ citations }: { citations?: MemoryCitation[] }) {
   const t = useT();
   const bodyRef = useRef<HTMLDivElement>(null);
@@ -181,6 +207,65 @@ export function UserMessage({
   const [editSubmitting, setEditSubmitting] = useState(false);
   const editRef = useRef<HTMLTextAreaElement>(null);
   const [imagePreviews, setImagePreviews] = useState<Record<string, string>>({});
+  const [imageViewer, setImageViewer] = useState<{ open: boolean; url: string; name: string }>({ open: false, url: "", name: "" });
+  const openImageViewer = useCallback(async (path: string, name: string) => {
+    let url = imagePreviews[path];
+    if (!url) {
+      try {
+        url = await app.AttachmentDataURL(path);
+        setImagePreviews((prev) => (prev[path] ? prev : { ...prev, [path]: url }));
+      } catch {
+        return;
+      }
+    }
+    setImageViewer({ open: true, url, name });
+  }, [imagePreviews]);
+
+  const closeImageViewer = useCallback(() => {
+    setImageViewer((prev) => (prev.open ? { ...prev, open: false } : prev));
+  }, []);
+
+  const pasteBlocks = useMemo(() => parsePastedBlocks(actionText, submitText), [actionText, submitText]);
+  const [expandedPasteLabels, setExpandedPasteLabels] = useState<Record<string, boolean>>({});
+
+  type DisplaySegment =
+    | { type: "text"; content: string }
+    | { type: "paste"; block: PastedBlockInfo };
+
+  const displaySegments = useMemo((): DisplaySegment[] => {
+    if (pasteBlocks.length === 0) return [{ type: "text", content: displayText }];
+    const segments: DisplaySegment[] = [];
+    // Order blocks by their position in the text so cards appear inline.
+    const ordered = pasteBlocks
+      .map((b) => ({ block: b, pos: displayText.indexOf(b.label) }))
+      .filter((x) => x.pos >= 0)
+      .sort((a, b) => a.pos - b.pos);
+    let remaining = displayText;
+    for (const { block } of ordered) {
+      const idx = remaining.indexOf(block.label);
+      if (idx < 0) continue;
+      // Text before the label: strip the trailing newline that separated the
+      // label from the preceding line so the card sits tight against the text.
+      if (idx > 0) {
+        let before = remaining.slice(0, idx);
+        before = before.replace(/\n$/, "");
+        if (before) segments.push({ type: "text", content: before });
+      }
+      segments.push({ type: "paste", block });
+      remaining = remaining.slice(idx + block.label.length);
+    }
+    // Strip the leading newline that followed the label.
+    remaining = remaining.replace(/^\n/, "");
+    if (remaining.trim()) segments.push({ type: "text", content: remaining });
+    return segments.length > 0 ? segments : [{ type: "text", content: displayText }];
+  }, [displayText, pasteBlocks]);
+
+  const togglePasteExpand = (label: string) => {
+    setExpandedPasteLabels((prev) => ({
+      ...prev,
+      [label]: !prev[label],
+    }));
+  };
   const orderedDraftAttachments = sortDisplayAttachments(draftAttachments);
   const imagePreviewKey = orderedAttachments
     .concat(orderedDraftAttachments)
@@ -305,11 +390,12 @@ export function UserMessage({
                     <ComposerContextCard
                       key={attachment.path}
                       variant={attachment.source === "workspace" ? "workspace" : "attachment"}
-                      tooltipLabel={attachment.source === "workspace" ? formatAttachmentRefForSubmit(attachment) : attachment.path}
+                      tooltipLabel={imagePreview ? `${t("imageViewer.clickToPreview")} — ${attachment.path}` : attachment.source === "workspace" ? formatAttachmentRefForSubmit(attachment) : attachment.path}
                       removeLabel={attachment.source === "workspace" ? t("composer.removeReference") : t("composer.removeImage")}
                       removeDisabled={editSubmitting}
                       onRemove={() => removeDraftAttachment(attachment.path)}
                       previewUrl={imagePreview}
+                      onImageClick={imagePreview ? () => openImageViewer(attachment.path, attachment.name) : undefined}
                       imageOnly={imageOnly}
                       folder={attachment.kind === "folder"}
                       label={attachment.kind === "folder" ? `${attachment.name}/` : attachment.name}
@@ -355,26 +441,78 @@ export function UserMessage({
             )}
           </div>
         ) : (
-          displayText && <div className="msg__text">{displayText}</div>
+          <>
+            {displaySegments.map((seg, i) => {
+              if (seg.type === "text") {
+                return seg.content ? <div className="msg__text" key={`s${i}`}>{seg.content}</div> : null;
+              }
+              const expanded = Boolean(expandedPasteLabels[seg.block.label]);
+              return (
+                <div className="msg-pasted" key={seg.block.label}>
+                  <div className="msg-pasted-block">
+                    <div className="msg-pasted-head">
+                      <FileText size={15} />
+                      <span className="msg-pasted-label">{seg.block.label}</span>
+                      <div className="msg-pasted-actions">
+                        <Tooltip label={t(expanded ? "msg.pastedCollapseTooltip" : "msg.pastedExpandTooltip")}>
+                          <button type="button" onClick={() => togglePasteExpand(seg.block.label)}>
+                            {expanded ? t("common.collapse") : t("composer.pastedExpand")}
+                          </button>
+                        </Tooltip>
+                      </div>
+                    </div>
+                    {expanded && (
+                      <div className="msg-pasted-expanded">{seg.block.content}</div>
+                    )}
+                  </div>
+                </div>
+              );
+            })}
+          </>
         )}
         {failed && <div className="msg__send-failed">{t("msg.sendFailed")}</div>}
         {orderedAttachments.length > 0 && (
           <div className="msg-attachments" aria-label={t("msg.attachments")}>
-            {orderedAttachments.map((attachment, index) => (
-              <div className={`msg-attachment msg-attachment--${attachment.kind}`} key={`${attachment.path}:${index}`} title={attachment.path}>
-                <span className={`msg-attachment__icon msg-attachment__icon--${attachment.kind}`} aria-hidden="true">
-                  {attachment.kind === "image" && imagePreviews[attachment.path] ? <img src={imagePreviews[attachment.path]} alt="" draggable={false} /> : attachmentIcon(attachment.kind)}
-                </span>
-                <span className="msg-attachment__main">
-                  <span className="msg-attachment__name">{attachment.name}</span>
-                  <span className="msg-attachment__meta">
-                    {attachment.kind === "folder"
-                      ? t("msg.folderReference")
-                      : `${attachment.ext || t("msg.fileAttachment")} · ${attachment.source === "workspace" ? t("msg.workspaceReference") : attachment.kind === "image" ? t("msg.imageAttachment") : t("msg.fileAttachment")}`}
+            {orderedAttachments.map((attachment, index) => {
+              const isImage = attachment.kind === "image";
+              const el = (
+                <div
+                  className={`msg-attachment msg-attachment--${attachment.kind}`}
+                  key={isImage ? undefined : `${attachment.path}:${index}`}
+                  title={isImage ? undefined : attachment.path}
+                  onClick={isImage ? () => openImageViewer(attachment.path, attachment.name) : undefined}
+                  role={isImage ? "button" : undefined}
+                  tabIndex={isImage ? 0 : undefined}
+                  onKeyDown={isImage ? (e) => { if (e.key === "Enter" || e.key === " ") { e.preventDefault(); openImageViewer(attachment.path, attachment.name); } } : undefined}
+                >
+                  <span className={`msg-attachment__icon msg-attachment__icon--${attachment.kind}`} aria-hidden="true">
+                    {isImage && imagePreviews[attachment.path] ? <img src={imagePreviews[attachment.path]} alt="" draggable={false} /> : attachmentIcon(attachment.kind)}
                   </span>
-                </span>
-              </div>
-            ))}
+                  <span className="msg-attachment__main">
+                    <span className="msg-attachment__name">{attachment.name}</span>
+                    <span className="msg-attachment__meta">
+                      {attachment.kind === "folder"
+                        ? t("msg.folderReference")
+                        : `${attachment.ext || t("msg.fileAttachment")} · ${attachment.source === "workspace" ? t("msg.workspaceReference") : attachment.kind === "image" ? t("msg.imageAttachment") : t("msg.fileAttachment")}`}
+                    </span>
+                  </span>
+                </div>
+              );
+              if (isImage) {
+                return (
+                  <Tooltip key={`${attachment.path}:${index}`} label={t("imageViewer.clickToPreview")} block>
+                    {el}
+                  </Tooltip>
+                );
+              }
+              return el;
+            })}
+            <ImageViewer
+              open={imageViewer.open}
+              imageUrl={imageViewer.url}
+              imageName={imageViewer.name}
+              onClose={closeImageViewer}
+            />
           </div>
         )}
       </div>

desktop/frontend/src/components/ProjectTree.tsx

@@ -1140,7 +1140,15 @@ export function ProjectTree({
       const imSourceLabel = imSource?.label || "";
       const imSourceTitle = imSourceLabel ? t("msg.fromIm", { source: imSourceLabel }) : "";
       const imSourcePlatform = (imSource?.platform || "im").replace(/[^a-z0-9_-]/gi, "").toLowerCase() || "im";
-      const title = [label, imSourceTitle, statusLabel, meta, exactTimeLabel].filter(Boolean).join(" · ");
+      const recovered = Boolean(node.recovered);
+      const recoveryTitle = recovered
+        ? [
+            t("recovery.branch"),
+            node.recoveryReason ? t("recovery.reason", { reason: node.recoveryReason }) : "",
+            node.recoveryDigest ? t("recovery.digest", { digest: node.recoveryDigest.slice(0, 12) }) : "",
+          ].filter(Boolean).join(" · ")
+        : "";
+      const title = [label, recoveryTitle, imSourceTitle, statusLabel, meta, exactTimeLabel].filter(Boolean).join(" · ");
       const topicMenuOpen = !isSessionNode && menuTopic === topicId;
       const pinned = Boolean(node.pinned);
       const pinLabel = t(pinned ? "projectTree.unpinTopic" : "projectTree.pinTopic");
@@ -1217,7 +1225,7 @@ export function ProjectTree({
       }
       const row = (
         <div
-          className={`project-tree__topic${scopeClass}${isSessionNode ? " project-tree__topic--session" : ""}${active ? " project-tree__topic--active" : ""}${node.running ? " project-tree__topic--running" : ""}${status ? ` project-tree__topic--status-${status}` : ""}${unread ? " project-tree__topic--unread" : ""}${!isSessionNode && pinned ? " project-tree__topic--pinned" : ""}${topicMenuOpen ? " project-tree__topic--menu-open" : ""}${sideTimeVisible && (timeLabel || showStatusInSide) ? " project-tree__topic--with-side" : meta ? " project-tree__topic--has-meta" : ""}${imSource ? " project-tree__topic--im-source" : ""}${shortcutIndex > 0 ? " project-tree__topic--show-shortcut" : ""}`}
+          className={`project-tree__topic${scopeClass}${isSessionNode ? " project-tree__topic--session" : ""}${active ? " project-tree__topic--active" : ""}${node.running ? " project-tree__topic--running" : ""}${status ? ` project-tree__topic--status-${status}` : ""}${unread ? " project-tree__topic--unread" : ""}${!isSessionNode && pinned ? " project-tree__topic--pinned" : ""}${recovered ? " project-tree__topic--recovered" : ""}${topicMenuOpen ? " project-tree__topic--menu-open" : ""}${sideTimeVisible && (timeLabel || showStatusInSide) ? " project-tree__topic--with-side" : meta ? " project-tree__topic--has-meta" : ""}${imSource ? " project-tree__topic--im-source" : ""}${shortcutIndex > 0 ? " project-tree__topic--show-shortcut" : ""}`}
           style={accentStyle}
           onContextMenu={isSessionNode ? undefined : openTopicMenu}
         >
@@ -1271,6 +1279,11 @@ export function ProjectTree({
                   </span>
                 )}
                 {!compactTopics && statusLabel && <span className={`project-tree__topic-status project-tree__topic-status--${status}`}>{statusLabel}</span>}
+                {recovered && (
+                  <span className="project-tree__topic-recovery" title={recoveryTitle}>
+                    {t("recovery.badge")}
+                  </span>
+                )}
               </span>
               {!compactTopics && !creationTopics && meta && (
                 <span className="project-tree__topic-meta">
@@ -1294,6 +1307,11 @@ export function ProjectTree({
                 {meta}
               </span>
             )}
+            {compactTopics && recovered && (
+              <span className="sr-only">
+                {t("recovery.branch")}
+              </span>
+            )}
           </button>
           {unread && <span className="project-tree__topic-unread-dot" aria-hidden="true" />}
           {projectTreeShouldRenderTopicActions(isSessionNode, compactTopics, unread) && (

desktop/frontend/src/components/WorkspacePanel.tsx

@@ -24,6 +24,7 @@ import {
   Search,
   X,
 } from "lucide-react";
+import { asArray } from "../lib/array";
 import { app } from "../lib/bridge";
 import { useT } from "../lib/i18n";
 import {
@@ -206,6 +207,7 @@ export function WorkspacePanel({
   onPreviewModeChange,
   onAddToChat,
   onRequestPanelWidth,
+  onFileTreeRefresh,
   refreshKey,
   initialViewMode = "files",
   revealPathRequest,
@@ -224,6 +226,7 @@ export function WorkspacePanel({
   onPreviewModeChange?: (active: boolean) => void;
   onAddToChat?: (text: string) => void;
   onRequestPanelWidth?: (width: number) => void;
+  onFileTreeRefresh?: () => void;
   refreshKey?: number;
   initialViewMode?: "files" | "changed";
   revealPathRequest?: WorkspaceRevealRequest | null;
@@ -275,6 +278,8 @@ export function WorkspacePanel({
   const workspaceChangesRequestIdRef = useRef(0);
   const gitHistoryRequestIdRef = useRef(0);
   const commitDetailRequestIdRef = useRef(0);
+  const dirLoadGenerationRef = useRef(0);
+  const dirLoadRequestIdsRef = useRef<Record<string, number>>({});
   const recentAnchorRef = useRef<HTMLButtonElement>(null);
   const openDirsRef = useRef(openDirs);
   const pendingTreeRevealPathRef = useRef<string | null>(null);
@@ -284,8 +289,12 @@ export function WorkspacePanel({
   }, [openDirs]);
 
   const loadDir = useCallback(async (dir: string) => {
-    const entries = await app.ListDir(dir).catch(() => []);
-    setEntriesByDir((prev) => ({ ...prev, [dir]: entries ?? [] }));
+    const generation = dirLoadGenerationRef.current;
+    const requestId = (dirLoadRequestIdsRef.current[dir] ?? 0) + 1;
+    dirLoadRequestIdsRef.current[dir] = requestId;
+    const entries = await app.ListDir(dir).catch((): DirEntry[] => []);
+    if (dirLoadGenerationRef.current !== generation || dirLoadRequestIdsRef.current[dir] !== requestId) return;
+    setEntriesByDir((prev) => ({ ...prev, [dir]: asArray(entries) }));
   }, []);
 
   const loadGitHistory = useCallback(async () => {
@@ -399,15 +408,15 @@ export function WorkspacePanel({
       setOpenTabs((tabs) => [...tabs.filter((tab) => tab !== path), path].slice(-WORKSPACE_MAX_PREVIEW_TABS));
       const dirs = parentDirs(path);
       setOpenDirs((prev) => new Set([...Array.from(prev), ...dirs]));
-      dirs.forEach((dir) => {
-        if (!entriesByDir[dir]) void loadDir(dir);
-      });
+      dirs.forEach((dir) => void loadDir(dir));
     },
-    [entriesByDir, loadDir, openTabs.length, panelWidth, selectedPath, treeVisible],
+    [loadDir, openTabs.length, panelWidth, selectedPath, treeVisible],
   );
 
   useEffect(() => {
     if (!open) return;
+    dirLoadGenerationRef.current += 1;
+    dirLoadRequestIdsRef.current = {};
     setEntriesByDir({});
     setOpenDirs(new Set([""]));
     setSelectedPath(null);
@@ -503,10 +512,8 @@ export function WorkspacePanel({
     setTreeMenu(null);
     const dirs = Array.from(new Set(paths.flatMap(parentDirs)));
     setOpenDirs((prev) => new Set([...Array.from(prev), ...dirs]));
-    dirs.forEach((dir) => {
-      if (!entriesByDir[dir]) void loadDir(dir);
-    });
-  }, [entriesByDir, fileListRequest, loadDir, open, scopedFilePaths, viewMode]);
+    dirs.forEach((dir) => void loadDir(dir));
+  }, [fileListRequest, loadDir, open, scopedFilePaths, viewMode]);
 
   useEffect(() => {
     if (!open || changeListRequest) return;
@@ -649,10 +656,10 @@ export function WorkspacePanel({
       void loadWorkspaceChanges();
       return;
     }
+    onFileTreeRefresh?.();
     const dirs = Array.from(openDirsRef.current);
-    setEntriesByDir({});
     dirs.forEach((dir) => void loadDir(dir));
-  }, [loadGitHistory, loadWorkspaceChanges, loadDir, viewMode]);
+  }, [loadGitHistory, loadWorkspaceChanges, loadDir, onFileTreeRefresh, viewMode]);
 
   const refreshSelected = useCallback(() => {
     if (!selectedPath) return;
@@ -698,12 +705,12 @@ export function WorkspacePanel({
           next.delete(dir);
         } else {
           next.add(dir);
-          if (!entriesByDir[dir]) void loadDir(dir);
+          void loadDir(dir);
         }
         return next;
       });
     },
-    [entriesByDir, loadDir],
+    [loadDir],
   );
 
   const breadcrumbDirs = selectedPath ? parentDirs(selectedPath) : [""];
@@ -1262,6 +1269,7 @@ export function WorkspacePanel({
                 setFilter("");
                 showTreeEvenSplit();
                 setOpenDirs((prev) => new Set([...Array.from(prev), ""]));
+                void loadDir("");
               }}
             >
               {shortCwd(cwd) || t("workspace.title")}

desktop/frontend/src/lib/bridge.ts

@@ -48,13 +48,18 @@ import type {
   Mode,
   ModelInfo,
   NetworkView,
+  PluginInstallOptions,
+  PluginView,
   ProjectNode,
   PromptHistoryEntry,
   PromptHistoryResult,
+  ProviderPresetView,
   ProviderView,
   QuestionAnswer,
   ServerView,
   SessionMeta,
+  SessionRecoveryFailedEvent,
+  SessionRecoveryEvent,
   SettingsView,
   SkillsSettingsView,
   SkillRootView,
@@ -207,6 +212,13 @@ export interface AppBindings {
   Capabilities(): Promise<CapabilitiesView>;
   MCPServers(): Promise<ServerView[]>;
   SkillsSettings(): Promise<SkillsSettingsView>;
+  Plugins(): Promise<PluginView[]>;
+  PlanPluginInstall(source: string, options: PluginInstallOptions): Promise<string>;
+  InstallPlugin(source: string, options: PluginInstallOptions): Promise<string>;
+  RemovePlugin(name: string): Promise<void>;
+  SetPluginEnabled(name: string, enabled: boolean): Promise<void>;
+  UpdatePlugin(name: string): Promise<string>;
+  PluginDoctor(name: string): Promise<PluginView>;
   AddMCPServer(input: MCPServerInput): Promise<number>;
   UpdateMCPServer(name: string, input: MCPServerInput): Promise<void>;
   RemoveMCPServer(name: string): Promise<void>;
@@ -216,6 +228,7 @@ export interface AppBindings {
   TrustMCPServerTools(name: string, toolNames: string[]): Promise<void>;
   UntrustMCPServerTool(name: string, toolName: string): Promise<void>;
   PickSkillFolder(): Promise<string>;
+  PickPluginFolder(): Promise<string>;
   AddSkillPath(path: string): Promise<void>;
   RemoveSkillPath(path: string): Promise<void>;
   RefreshSkills(): Promise<void>;
@@ -277,10 +290,13 @@ export interface AppBindings {
   SetPlannerModel(ref: string): Promise<void>;
   SetSubagentModel(ref: string): Promise<void>;
   SetSubagentEffort(level: string): Promise<void>;
+  SetMaxSubagentDepth(depth: number): Promise<void>;
   SetAutoPlan(mode: string): Promise<void>;
   SetDefaultToolApprovalMode(mode: string): Promise<void>;
   SaveProvider(p: ProviderView): Promise<void>;
   AddOfficialProviderAccess(kind: string, key: string): Promise<string>;
+  AddProviderPresetAccess(id: string, key: string): Promise<string>;
+  ResetProviderPresetAccess(id: string): Promise<void>;
   FetchProviderModels(p: ProviderView): Promise<string[]>;
   DeleteProvider(name: string): Promise<void>;
   RemoveProviderAccess(name: string): Promise<void>;
@@ -289,6 +305,7 @@ export interface AppBindings {
   SetPermissionMode(mode: string): Promise<void>;
   AddPermissionRule(list: string, rule: string): Promise<void>;
   RemovePermissionRule(list: string, rule: string): Promise<void>;
+  ReloadSettings(): Promise<void>;
   SetSandbox(bash: string, network: boolean, workspaceRoot: string, allowWrite: string[], shell: string): Promise<void>;
   SetNetwork(n: NetworkView): Promise<void>;
   SetBotSettings(b: BotSettingsView): Promise<void>;
@@ -571,17 +588,31 @@ export function onProjectTreeChanged(cb: () => void): () => void {
   return () => {};
 }
 
+export function onSessionRecovered(cb: (payload: SessionRecoveryEvent) => void): () => void {
+  if (realApp() && typeof window !== "undefined" && window.runtime) {
+    return window.runtime.EventsOn("session:recovered", (payload?: unknown) => cb((payload ?? {}) as SessionRecoveryEvent));
+  }
+  return () => {};
+}
+
+export function onSessionRecoveryFailed(cb: (payload: SessionRecoveryFailedEvent) => void): () => void {
+  if (realApp() && typeof window !== "undefined" && window.runtime) {
+    return window.runtime.EventsOn("session:recovery-failed", (payload?: unknown) => cb((payload ?? {}) as SessionRecoveryFailedEvent));
+  }
+  return () => {};
+}
+
 // app proxies each call to the live binding (or the dev mock only when truly
 // outside the shell), so a late-injected window.go is picked up transparently.
 function bridgeBreadcrumb(method: string): string {
   if (method === "ReportCrash") return "";
   if (/^(Submit|SubmitDisplay|RunShell|Steer|Cancel|Approve|AnswerQuestion|ReplayPendingPrompts)/.test(method))
     return `turn ${method}`;
-  if (/^(SetModel|SetEffort|SetTokenMode|SetDefaultModel|SetPlannerModel|SetSubagentModel|SetSubagentEffort)/.test(method))
+  if (/^(SetModel|SetEffort|SetTokenMode|SetDefaultModel|SetPlannerModel|SetSubagentModel|SetSubagentEffort|SetMaxSubagentDepth)/.test(method))
     return `model ${method}`;
   if (/^(SetDesktop|SetCloseBehavior|SetDisplayMode|SetStatusBar|SetExpandThinking|SetAutoPlan|SetDefaultToolApprovalMode|SetMemoryCompilerEnabled|SetReasoningLanguage)/.test(method))
     return `settings ${method}`;
-  if (/^(SaveProvider|AddOfficialProviderAccess|RemoveProviderAccess|DeleteProvider|SetProviderKey|ClearProviderKey|FetchProviderModels|ConnectKey)/.test(method))
+  if (/^(SaveProvider|AddOfficialProviderAccess|AddProviderPresetAccess|ResetProviderPresetAccess|RemoveProviderAccess|DeleteProvider|SetProviderKey|ClearProviderKey|FetchProviderModels|ConnectKey)/.test(method))
     return `provider ${method}`;
   if (/^(CheckUpdate|DownloadUpdate|InstallUpdate|ApplyUpdate|OpenDownloadPage)/.test(method)) return `update ${method}`;
   if (/^(AddMCPServer|UpdateMCPServer|RemoveMCPServer|ReconnectMCPServer|ClearMCPServerAuthentication|TrustMCPServerTool|TrustMCPServerTools|UntrustMCPServerTool|SetMCPServer)/.test(method))
@@ -695,11 +726,140 @@ function mockScenario(): "demo" | "fresh" | "running" | "guidance" {
   return "demo";
 }
 
+type MockProviderPresetTemplate = {
+  id: string;
+  label: string;
+  description: string;
+  keyEnv: string;
+  provider: ProviderView;
+};
+
+function mockProviderTemplate(p: Pick<ProviderView, "name" | "kind" | "baseUrl" | "models" | "default" | "apiKeyEnv"> & Partial<ProviderView>): ProviderView {
+  return {
+    name: p.name,
+    builtIn: false,
+    added: true,
+    kind: p.kind,
+    baseUrl: p.baseUrl,
+    modelsUrl: p.modelsUrl ?? "",
+    models: p.models,
+    visionModels: p.visionModels ?? [],
+    visionModelsConfigured: Boolean(p.visionModelsConfigured ?? ((p.visionModels ?? []).length > 0)),
+    default: p.default,
+    apiKeyEnv: p.apiKeyEnv,
+    headers: p.headers,
+    extraBody: p.extraBody,
+    authHeader: p.authHeader,
+    keySet: Boolean(p.keySet),
+    balanceUrl: p.balanceUrl ?? "",
+    contextWindow: p.contextWindow ?? 0,
+    reasoningProtocol: p.reasoningProtocol ?? "",
+    thinking: p.thinking ?? "",
+    supportedEfforts: p.supportedEfforts ?? [],
+    defaultEffort: p.defaultEffort ?? "",
+    modelOverrides: p.modelOverrides,
+  };
+}
+
+function mockPreset(id: string, label: string, description: string, keyEnv: string, provider: ProviderView): MockProviderPresetTemplate {
+  return { id, label, description, keyEnv, provider };
+}
+
+const mockKimiAPIModels = ["kimi-k2.7-code", "kimi-k2.7-code-highspeed", "kimi-k2.6", "kimi-k2.5"];
+const mockMiMoV25Models = ["mimo-v2.5-pro", "mimo-v2.5"];
+const mockMiniMaxModels = ["MiniMax-M3", "MiniMax-M2.7", "MiniMax-M2.7-highspeed"];
+const mockGLMAPIModels = ["glm-5.2", "glm-5.1", "glm-5", "glm-5-turbo", "glm-5v-turbo", "glm-4.7", "glm-4.7-flash", "glm-4.7-flashx", "glm-4.6", "glm-4.5", "glm-4.5-air", "glm-4.5-flash"];
+const mockGLMCodingModels = ["glm-5.2", "glm-5.1", "glm-5", "glm-4.7"];
+const mockGLMAnthropicModels = ["glm-5.2[1m]", "glm-5.2", "glm-5.1", "glm-5", "glm-4.7", "glm-4.5-air"];
+const mockQwenAPIModels = ["qwen3.7-plus", "qwen3.7-max", "qwen3.6-plus", "qwen3.5-plus", "qwen3-max-2026-01-23", "qwen3-coder-next", "qwen3-coder-plus", "MiniMax-M2.5", "glm-5", "glm-4.7", "kimi-k2.5"];
+const mockQwenPlanModels = ["qwen3.7-plus", "qwen3.6-plus", "kimi-k2.5", "glm-5", "MiniMax-M2.5", "qwen3.5-plus", "qwen3-max-2026-01-23", "qwen3-coder-next", "qwen3-coder-plus", "glm-4.7"];
+const mockQwenPlanVisionModels = ["qwen3.7-plus", "qwen3.6-plus", "qwen3.5-plus", "kimi-k2.5"];
+const mockStepFunModels = ["step-3.7-flash", "step-3.5-flash", "step-3.5-flash-2603"];
+const mockOpenCodeGoModels = ["glm-5.2", "glm-5.1", "kimi-k2.7-code", "kimi-k2.6", "deepseek-v4-pro", "deepseek-v4-flash", "mimo-v2.5-pro", "mimo-v2.5"];
+const mockOpenCodeGoAnthropicModels = ["qwen3.7-plus", "qwen3.7-max", "qwen3.6-plus", "minimax-m3", "minimax-m2.7", "minimax-m2.5"];
+const mockOpenCodeZenAnthropicModels = ["claude-sonnet-4-6", "claude-opus-4-8", "claude-haiku-4-5", "qwen3.6-plus", "qwen3.5-plus", "qwen3.6-plus-free"];
+const mockNovitaModels = ["zai-org/glm-5.2", "moonshotai/kimi-k2.7-code", "minimax/minimax-m3", "deepseek/deepseek-v4-pro", "deepseek/deepseek-v4-flash", "qwen/qwen3.7-max", "qwen/qwen3.6-plus", "zai-org/glm-5v-turbo"];
+const mockGMIModels = ["zai-org/GLM-5.2-FP8", "deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash", "moonshotai/Kimi-K2.7-Code", "anthropic/claude-sonnet-4.6", "openai/gpt-5.5"];
+const mockVercelModels = ["anthropic/claude-sonnet-4.6", "anthropic/claude-opus-4.8", "openai/gpt-5.4", "openai/gpt-5.4-pro", "moonshotai/kimi-k2.7-code", "zai/glm-5.2", "deepseek/deepseek-v4-pro"];
+const mockOllamaCloudModels = ["glm-5.2", "kimi-k2.7-code", "deepseek-v4-pro", "deepseek-v4-flash", "minimax-m3", "nemotron-3-nano:30b", "qwen3-coder-next"];
+
+const mockProviderPresetTemplates: MockProviderPresetTemplate[] = [
+  mockPreset("kimi-cn", "Kimi CN API", "Moonshot Kimi China OpenAI-compatible API.", "KIMI_API_KEY", mockProviderTemplate({ name: "kimi-cn", kind: "openai", baseUrl: "https://api.moonshot.cn/v1", models: mockKimiAPIModels, visionModels: mockKimiAPIModels, default: "kimi-k2.7-code", apiKeyEnv: "KIMI_API_KEY", balanceUrl: "https://api.moonshot.cn/v1/users/me/balance", contextWindow: 262144, reasoningProtocol: "none" })),
+  mockPreset("kimi-global", "Kimi Global API", "Moonshot Kimi international OpenAI-compatible API.", "MOONSHOT_API_KEY", mockProviderTemplate({ name: "kimi-global", kind: "openai", baseUrl: "https://api.moonshot.ai/v1", models: mockKimiAPIModels, visionModels: mockKimiAPIModels, default: "kimi-k2.7-code", apiKeyEnv: "MOONSHOT_API_KEY", balanceUrl: "https://api.moonshot.ai/v1/users/me/balance", contextWindow: 262144, reasoningProtocol: "none" })),
+  mockPreset("kimi-coding-plan", "Kimi Coding Plan", "Kimi Coding Plan via its dedicated Anthropic-compatible endpoint.", "KIMI_CODING_API_KEY", mockProviderTemplate({ name: "kimi-coding-plan", kind: "anthropic", baseUrl: "https://api.kimi.com/coding/", models: ["kimi-for-coding"], visionModels: ["kimi-for-coding"], default: "kimi-for-coding", apiKeyEnv: "KIMI_CODING_API_KEY", headers: { "User-Agent": "claude-code/0.1.0" }, thinking: "adaptive", contextWindow: 262144 })),
+  mockPreset("mimo-api", "MiMo API", "Xiaomi MiMo direct API with text and vision-capable models.", "MIMO_API_KEY", mockProviderTemplate({ name: "mimo-api", kind: "openai", baseUrl: "https://api.xiaomimimo.com/v1", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_API_KEY", contextWindow: 1048576 })),
+  mockPreset("mimo-anthropic", "MiMo Anthropic", "Xiaomi MiMo direct Anthropic-compatible endpoint.", "MIMO_API_KEY", mockProviderTemplate({ name: "mimo-anthropic", kind: "anthropic", baseUrl: "https://api.xiaomimimo.com/anthropic", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_API_KEY", thinking: "adaptive", contextWindow: 1048576 })),
+  mockPreset("mimo-token-plan-cn", "MiMo Token Plan CN", "Xiaomi MiMo token-plan China endpoint.", "MIMO_TOKEN_PLAN_API_KEY", mockProviderTemplate({ name: "mimo-token-plan-cn", kind: "openai", baseUrl: "https://token-plan-cn.xiaomimimo.com/v1", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_TOKEN_PLAN_API_KEY", contextWindow: 1048576 })),
+  mockPreset("mimo-token-plan-cn-anthropic", "MiMo Token Plan CN Anthropic", "Xiaomi MiMo token-plan China Anthropic-compatible endpoint.", "MIMO_TOKEN_PLAN_API_KEY", mockProviderTemplate({ name: "mimo-token-plan-cn-anthropic", kind: "anthropic", baseUrl: "https://token-plan-cn.xiaomimimo.com/anthropic", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_TOKEN_PLAN_API_KEY", thinking: "adaptive", contextWindow: 1048576 })),
+  mockPreset("mimo-token-plan-sgp", "MiMo Token Plan SGP", "Xiaomi MiMo token-plan Singapore endpoint.", "MIMO_TOKEN_PLAN_API_KEY", mockProviderTemplate({ name: "mimo-token-plan-sgp", kind: "openai", baseUrl: "https://token-plan-sgp.xiaomimimo.com/v1", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_TOKEN_PLAN_API_KEY", contextWindow: 1048576 })),
+  mockPreset("mimo-token-plan-sgp-anthropic", "MiMo Token Plan SGP Anthropic", "Xiaomi MiMo token-plan Singapore Anthropic-compatible endpoint.", "MIMO_TOKEN_PLAN_API_KEY", mockProviderTemplate({ name: "mimo-token-plan-sgp-anthropic", kind: "anthropic", baseUrl: "https://token-plan-sgp.xiaomimimo.com/anthropic", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_TOKEN_PLAN_API_KEY", thinking: "adaptive", contextWindow: 1048576 })),
+  mockPreset("mimo-token-plan-ams", "MiMo Token Plan AMS", "Xiaomi MiMo token-plan Amsterdam endpoint.", "MIMO_TOKEN_PLAN_API_KEY", mockProviderTemplate({ name: "mimo-token-plan-ams", kind: "openai", baseUrl: "https://token-plan-ams.xiaomimimo.com/v1", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_TOKEN_PLAN_API_KEY", contextWindow: 1048576 })),
+  mockPreset("mimo-token-plan-ams-anthropic", "MiMo Token Plan AMS Anthropic", "Xiaomi MiMo token-plan Amsterdam Anthropic-compatible endpoint.", "MIMO_TOKEN_PLAN_API_KEY", mockProviderTemplate({ name: "mimo-token-plan-ams-anthropic", kind: "anthropic", baseUrl: "https://token-plan-ams.xiaomimimo.com/anthropic", models: mockMiMoV25Models, visionModels: ["mimo-v2.5"], default: "mimo-v2.5-pro", apiKeyEnv: "MIMO_TOKEN_PLAN_API_KEY", thinking: "adaptive", contextWindow: 1048576 })),
+  mockPreset("minimax-cn-api", "MiniMax CN API", "MiniMax China OpenAI-compatible M-series API endpoint.", "MINIMAX_API_KEY", mockProviderTemplate({ name: "minimax-cn-api", kind: "openai", baseUrl: "https://api.minimaxi.com/v1", models: mockMiniMaxModels, visionModels: ["MiniMax-M3"], default: "MiniMax-M3", apiKeyEnv: "MINIMAX_API_KEY", extraBody: { reasoning_split: true }, contextWindow: 1048576, thinking: "adaptive", supportedEfforts: ["disabled", "adaptive"], defaultEffort: "adaptive" })),
+  mockPreset("minimax-global-api", "MiniMax Global API", "MiniMax international OpenAI-compatible M-series API endpoint.", "MINIMAX_API_KEY", mockProviderTemplate({ name: "minimax-global-api", kind: "openai", baseUrl: "https://api.minimax.io/v1", models: mockMiniMaxModels, visionModels: ["MiniMax-M3"], default: "MiniMax-M3", apiKeyEnv: "MINIMAX_API_KEY", extraBody: { reasoning_split: true }, contextWindow: 1048576, thinking: "adaptive", supportedEfforts: ["disabled", "adaptive"], defaultEffort: "adaptive" })),
+  mockPreset("minimax-cn-anthropic", "MiniMax CN Anthropic", "MiniMax China Anthropic-compatible M-series endpoint.", "MINIMAX_PLAN_API_KEY", mockProviderTemplate({ name: "minimax-cn-anthropic", kind: "anthropic", baseUrl: "https://api.minimaxi.com/anthropic", models: mockMiniMaxModels, visionModels: ["MiniMax-M3"], default: "MiniMax-M3", apiKeyEnv: "MINIMAX_PLAN_API_KEY", authHeader: true, contextWindow: 1048576, thinking: "adaptive", supportedEfforts: ["disabled", "adaptive"], defaultEffort: "adaptive" })),
+  mockPreset("minimax-global-anthropic", "MiniMax Global Anthropic", "MiniMax international Anthropic-compatible endpoint with Bearer auth.", "MINIMAX_API_KEY", mockProviderTemplate({ name: "minimax-global-anthropic", kind: "anthropic", baseUrl: "https://api.minimax.io/anthropic", models: mockMiniMaxModels, visionModels: ["MiniMax-M3"], default: "MiniMax-M3", apiKeyEnv: "MINIMAX_API_KEY", authHeader: true, contextWindow: 1048576, thinking: "adaptive", supportedEfforts: ["disabled", "adaptive"], defaultEffort: "adaptive" })),
+  mockPreset("glm-cn", "GLM CN API", "Zhipu GLM China OpenAI-compatible API with thinking controls.", "GLM_API_KEY", mockProviderTemplate({ name: "glm-cn", kind: "openai", baseUrl: "https://open.bigmodel.cn/api/paas/v4", models: mockGLMAPIModels, visionModels: ["glm-5v-turbo"], default: "glm-5.2", apiKeyEnv: "GLM_API_KEY", contextWindow: 1000000, thinking: "enabled", supportedEfforts: ["enabled", "disabled"], defaultEffort: "enabled" })),
+  mockPreset("zai-global", "Z.AI Global API", "Z.AI international OpenAI-compatible GLM API.", "ZAI_API_KEY", mockProviderTemplate({ name: "zai-global", kind: "openai", baseUrl: "https://api.z.ai/api/paas/v4", models: mockGLMAPIModels, visionModels: ["glm-5v-turbo"], default: "glm-5.2", apiKeyEnv: "ZAI_API_KEY", contextWindow: 1000000, thinking: "enabled", supportedEfforts: ["enabled", "disabled"], defaultEffort: "enabled" })),
+  mockPreset("glm-coding-plan-cn", "GLM Coding Plan CN", "Zhipu GLM China coding-plan endpoint.", "GLM_PLAN_API_KEY", mockProviderTemplate({ name: "glm-coding-plan-cn", kind: "openai", baseUrl: "https://open.bigmodel.cn/api/coding/paas/v4", models: mockGLMCodingModels, default: "glm-5.2", apiKeyEnv: "GLM_PLAN_API_KEY", contextWindow: 1000000, thinking: "enabled", supportedEfforts: ["enabled", "disabled"], defaultEffort: "enabled" })),
+  mockPreset("glm-coding-plan-cn-anthropic", "GLM Coding Plan CN Anthropic", "Zhipu GLM China coding-plan Anthropic-compatible endpoint.", "GLM_PLAN_API_KEY", mockProviderTemplate({ name: "glm-coding-plan-cn-anthropic", kind: "anthropic", baseUrl: "https://open.bigmodel.cn/api/anthropic", models: mockGLMAnthropicModels, default: "glm-5.2[1m]", apiKeyEnv: "GLM_PLAN_API_KEY", authHeader: true, thinking: "adaptive", contextWindow: 1000000 })),
+  mockPreset("zai-coding-plan-global", "Z.AI Coding Plan Global", "Z.AI international coding-plan endpoint.", "ZAI_CODING_API_KEY", mockProviderTemplate({ name: "zai-coding-plan-global", kind: "openai", baseUrl: "https://api.z.ai/api/coding/paas/v4", models: mockGLMCodingModels, default: "glm-5.2", apiKeyEnv: "ZAI_CODING_API_KEY", contextWindow: 1000000, thinking: "enabled", supportedEfforts: ["enabled", "disabled"], defaultEffort: "enabled" })),
+  mockPreset("zai-coding-plan-global-anthropic", "Z.AI Coding Plan Global Anthropic", "Z.AI international coding-plan Anthropic-compatible endpoint.", "ZAI_CODING_API_KEY", mockProviderTemplate({ name: "zai-coding-plan-global-anthropic", kind: "anthropic", baseUrl: "https://api.z.ai/api/anthropic", models: mockGLMAnthropicModels, default: "glm-5.2[1m]", apiKeyEnv: "ZAI_CODING_API_KEY", authHeader: true, thinking: "adaptive", contextWindow: 1000000 })),
+  mockPreset("opencode-go", "OpenCode Go", "OpenCode Go relay with per-model capability overrides.", "OPENCODE_GO_API_KEY", mockProviderTemplate({ name: "opencode-go", kind: "openai", baseUrl: "https://opencode.ai/zen/go/v1", models: mockOpenCodeGoModels, default: "glm-5.2", apiKeyEnv: "OPENCODE_GO_API_KEY", contextWindow: 128000 })),
+  mockPreset("opencode-go-anthropic", "OpenCode Go Anthropic", "OpenCode Go subscription Anthropic-compatible route for Qwen and MiniMax models.", "OPENCODE_GO_API_KEY", mockProviderTemplate({ name: "opencode-go-anthropic", kind: "anthropic", baseUrl: "https://opencode.ai/zen/go", models: mockOpenCodeGoAnthropicModels, visionModels: ["qwen3.7-plus", "qwen3.6-plus"], default: "qwen3.7-plus", apiKeyEnv: "OPENCODE_GO_API_KEY", thinking: "adaptive", contextWindow: 262144 })),
+  mockPreset("opencode-zen-anthropic", "OpenCode Zen Anthropic", "OpenCode Zen Anthropic-compatible route for Claude and Qwen models.", "OPENCODE_API_KEY", mockProviderTemplate({ name: "opencode-zen-anthropic", kind: "anthropic", baseUrl: "https://opencode.ai/zen", models: mockOpenCodeZenAnthropicModels, visionModels: ["claude-sonnet-4-6", "claude-opus-4-8", "claude-haiku-4-5"], default: "claude-sonnet-4-6", apiKeyEnv: "OPENCODE_API_KEY", contextWindow: 262144 })),
+  mockPreset("qwen-cn", "Qwen CN API", "Alibaba DashScope China standard OpenAI-compatible endpoint.", "QWEN_API_KEY", mockProviderTemplate({ name: "qwen-cn", kind: "openai", baseUrl: "https://dashscope.aliyuncs.com/compatible-mode/v1", models: mockQwenAPIModels, visionModels: ["qwen3.7-plus", "qwen3.6-plus", "qwen3.5-plus", "kimi-k2.5"], default: "qwen3.7-plus", apiKeyEnv: "QWEN_API_KEY" })),
+  mockPreset("qwen-global", "Qwen Global API", "Alibaba DashScope international standard OpenAI-compatible endpoint.", "QWEN_API_KEY", mockProviderTemplate({ name: "qwen-global", kind: "openai", baseUrl: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1", models: mockQwenAPIModels, visionModels: ["qwen3.7-plus", "qwen3.6-plus", "qwen3.5-plus", "kimi-k2.5"], default: "qwen3.7-plus", apiKeyEnv: "QWEN_API_KEY" })),
+  mockPreset("qwen-coding-plan-cn", "Qwen Coding Plan CN", "Alibaba Cloud Qwen Coding Plan China endpoint.", "QWEN_CODING_API_KEY", mockProviderTemplate({ name: "qwen-coding-plan-cn", kind: "openai", baseUrl: "https://coding.dashscope.aliyuncs.com/v1", models: mockQwenPlanModels, visionModels: mockQwenPlanVisionModels, default: "qwen3.7-plus", apiKeyEnv: "QWEN_CODING_API_KEY" })),
+  mockPreset("qwen-coding-plan-cn-anthropic", "Qwen Coding Plan CN Anthropic", "Alibaba Cloud Qwen Coding Plan China Anthropic-compatible endpoint.", "QWEN_CODING_API_KEY", mockProviderTemplate({ name: "qwen-coding-plan-cn-anthropic", kind: "anthropic", baseUrl: "https://coding.dashscope.aliyuncs.com/apps/anthropic", models: mockQwenPlanModels, visionModels: mockQwenPlanVisionModels, default: "qwen3.7-plus", apiKeyEnv: "QWEN_CODING_API_KEY", thinking: "adaptive" })),
+  mockPreset("qwen-coding-plan-global", "Qwen Coding Plan Global", "Alibaba Cloud Qwen Coding Plan international endpoint.", "QWEN_CODING_API_KEY", mockProviderTemplate({ name: "qwen-coding-plan-global", kind: "openai", baseUrl: "https://coding-intl.dashscope.aliyuncs.com/v1", models: mockQwenPlanModels, visionModels: mockQwenPlanVisionModels, default: "qwen3.7-plus", apiKeyEnv: "QWEN_CODING_API_KEY" })),
+  mockPreset("qwen-coding-plan-global-anthropic", "Qwen Coding Plan Global Anthropic", "Alibaba Cloud Qwen Coding Plan international Anthropic-compatible endpoint.", "QWEN_CODING_API_KEY", mockProviderTemplate({ name: "qwen-coding-plan-global-anthropic", kind: "anthropic", baseUrl: "https://coding-intl.dashscope.aliyuncs.com/apps/anthropic", models: mockQwenPlanModels, visionModels: mockQwenPlanVisionModels, default: "qwen3.7-plus", apiKeyEnv: "QWEN_CODING_API_KEY", thinking: "adaptive" })),
+  mockPreset("stepfun", "StepFun", "StepFun coding-plan OpenAI-compatible endpoint.", "STEPFUN_API_KEY", mockProviderTemplate({ name: "stepfun", kind: "openai", baseUrl: "https://api.stepfun.ai/step_plan/v1", models: mockStepFunModels, default: "step-3.7-flash", apiKeyEnv: "STEPFUN_API_KEY", supportedEfforts: ["low", "medium", "high"], defaultEffort: "medium" })),
+  mockPreset("stepfun-anthropic", "StepFun Anthropic", "StepFun coding-plan Anthropic-compatible endpoint.", "STEPFUN_API_KEY", mockProviderTemplate({ name: "stepfun-anthropic", kind: "anthropic", baseUrl: "https://api.stepfun.ai/step_plan", models: mockStepFunModels, default: "step-3.7-flash", apiKeyEnv: "STEPFUN_API_KEY", thinking: "adaptive", supportedEfforts: ["low", "medium", "high"], defaultEffort: "medium" })),
+  mockPreset("novita", "NovitaAI", "NovitaAI OpenAI-compatible multi-model gateway.", "NOVITA_API_KEY", mockProviderTemplate({ name: "novita", kind: "openai", baseUrl: "https://api.novita.ai/openai/v1", models: mockNovitaModels, default: "zai-org/glm-5.2", apiKeyEnv: "NOVITA_API_KEY" })),
+  mockPreset("gmi", "GMI Cloud", "GMI Cloud direct multi-model OpenAI-compatible gateway.", "GMI_API_KEY", mockProviderTemplate({ name: "gmi", kind: "openai", baseUrl: "https://api.gmi-serving.com/v1", models: mockGMIModels, default: "zai-org/GLM-5.2-FP8", apiKeyEnv: "GMI_API_KEY", headers: { "User-Agent": "Reasonix" } })),
+  mockPreset("vercel-ai-gateway", "Vercel AI Gateway", "Vercel AI Gateway via Anthropic-compatible Messages API.", "AI_GATEWAY_API_KEY", mockProviderTemplate({ name: "vercel-ai-gateway", kind: "anthropic", baseUrl: "https://ai-gateway.vercel.sh", models: mockVercelModels, visionModels: ["anthropic/claude-sonnet-4.6", "anthropic/claude-opus-4.8", "openai/gpt-5.4", "openai/gpt-5.4-pro", "moonshotai/kimi-k2.7-code"], default: "anthropic/claude-sonnet-4.6", apiKeyEnv: "AI_GATEWAY_API_KEY", authHeader: true, contextWindow: 1000000 })),
+  mockPreset("huggingface", "HuggingFace Router", "HuggingFace Inference Router OpenAI-compatible endpoint.", "HF_TOKEN", mockProviderTemplate({ name: "huggingface", kind: "openai", baseUrl: "https://router.huggingface.co/v1", models: ["zai-org/GLM-5.2", "deepseek-ai/DeepSeek-V3.2", "Qwen/Qwen3.5-72B-Instruct"], default: "zai-org/GLM-5.2", apiKeyEnv: "HF_TOKEN" })),
+  mockPreset("nvidia", "NVIDIA NIM", "NVIDIA NIM OpenAI-compatible accelerated inference endpoint.", "NVIDIA_API_KEY", mockProviderTemplate({ name: "nvidia", kind: "openai", baseUrl: "https://integrate.api.nvidia.com/v1", models: ["nvidia/nemotron-3-nano-30b-a3b", "nvidia/nemotron-3-super-120b-a12b", "nvidia/nemotron-3-ultra-550b-a55b", "deepseek-ai/deepseek-v4-pro", "qwen/qwen3.5-397b-a17b"], default: "nvidia/nemotron-3-nano-30b-a3b", apiKeyEnv: "NVIDIA_API_KEY" })),
+  mockPreset("kilocode", "KiloCode", "Kilo Code gateway OpenAI-compatible endpoint.", "KILOCODE_API_KEY", mockProviderTemplate({ name: "kilocode", kind: "openai", baseUrl: "https://api.kilo.ai/api/gateway", models: ["kilo/auto"], default: "kilo/auto", apiKeyEnv: "KILOCODE_API_KEY" })),
+  mockPreset("ollama-cloud", "Ollama Cloud", "Hosted Ollama Cloud OpenAI-compatible endpoint with max reasoning effort.", "OLLAMA_API_KEY", mockProviderTemplate({ name: "ollama-cloud", kind: "openai", baseUrl: "https://ollama.com/v1", models: mockOllamaCloudModels, default: "glm-5.2", apiKeyEnv: "OLLAMA_API_KEY" })),
+];
+
+function mockProviderPresetViews(): ProviderPresetView[] {
+  return mockProviderPresetTemplates.map((template) => ({
+    id: template.id,
+    label: template.label,
+    description: template.description,
+    keyEnv: template.keyEnv,
+    providerNames: [template.provider.name],
+    models: [...template.provider.models],
+    added: false,
+    status: "available",
+    statusProviderNames: [],
+    keySet: false,
+    requiresKey: true,
+    configured: false,
+  }));
+}
+
+function cloneMockProviderTemplate(id: string, key: string): ProviderView | undefined {
+  const template = mockProviderPresetTemplates.find((candidate) => candidate.id === id);
+  if (!template) return undefined;
+  return {
+    ...JSON.parse(JSON.stringify(template.provider)) as ProviderView,
+    keySet: Boolean(key.trim()),
+  };
+}
+
+const mockPreviewImageDataURL =
+  "data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' width='160' height='120' viewBox='0 0 160 120'%3E%3Cdefs%3E%3ClinearGradient id='g' x1='0' y1='0' x2='1' y2='1'%3E%3Cstop offset='0' stop-color='%23f97316'/%3E%3Cstop offset='1' stop-color='%232563eb'/%3E%3C/linearGradient%3E%3C/defs%3E%3Crect width='160' height='120' rx='14' fill='url(%23g)'/%3E%3Ccircle cx='44' cy='38' r='16' fill='%23fff7ed' opacity='.9'/%3E%3Cpath d='M18 96 62 58l24 22 18-16 38 32z' fill='%23ffffff' opacity='.9'/%3E%3C/svg%3E";
+
 function makeMockApp(): AppBindings {
   const scenario = mockScenario();
   const freshMock = scenario === "fresh";
   const guidanceMock = scenario === "guidance";
   const runningMock = scenario === "running" || guidanceMock;
+  const mockAttachmentDataURLs = new Map<string, string>();
   let cancelled = false;
   let pendingAskPreview = false;
   let pendingApprovalPreview = false;
@@ -788,6 +948,7 @@ function makeMockApp(): AppBindings {
       ],
     },
   ];
+  let capPlugins: PluginView[] = [];
   const mockSwitchWorkspace = async (path: string) => {
     cwd = path || "~";
     workspaces = [cwd, ...workspaces.filter((p) => p !== cwd)].slice(0, 12);
@@ -870,37 +1031,64 @@ function makeMockApp(): AppBindings {
     subagentEffort: "",
     autoPlan: "off",
     providers: [
-      { name: "deepseek", builtIn: true, added: false, kind: "openai", baseUrl: "https://api.deepseek.com", modelsUrl: "", models: ["deepseek-v4-flash"], visionModels: [], visionModelsConfigured: false, default: "deepseek-v4-flash", apiKeyEnv: "DEEPSEEK_API_KEY", keySet: true, balanceUrl: "https://api.deepseek.com/user/balance", contextWindow: 1_000_000, reasoningProtocol: "", supportedEfforts: [], defaultEffort: "" },
+      { name: "deepseek", builtIn: true, added: false, kind: "openai", baseUrl: "https://api.deepseek.com", modelsUrl: "", models: ["deepseek-v4-flash"], visionModels: [], visionModelsConfigured: false, default: "deepseek-v4-flash", apiKeyEnv: "DEEPSEEK_API_KEY", keySet: true, balanceUrl: "https://api.deepseek.com/user/balance", contextWindow: 1_000_000, reasoningProtocol: "", thinking: "", supportedEfforts: [], defaultEffort: "" },
     ],
     officialProviders: [
-      { name: "deepseek", builtIn: true, added: false, kind: "openai", baseUrl: "https://api.deepseek.com", modelsUrl: "", models: ["deepseek-v4-flash", "deepseek-v4-pro"], visionModels: [], visionModelsConfigured: false, default: "deepseek-v4-flash", apiKeyEnv: "DEEPSEEK_API_KEY", keySet: true, balanceUrl: "https://api.deepseek.com/user/balance", contextWindow: 1_000_000, reasoningProtocol: "", supportedEfforts: [], defaultEffort: "" },
+      { name: "deepseek", builtIn: true, added: false, kind: "openai", baseUrl: "https://api.deepseek.com", modelsUrl: "", models: ["deepseek-v4-flash", "deepseek-v4-pro"], visionModels: [], visionModelsConfigured: false, default: "deepseek-v4-flash", apiKeyEnv: "DEEPSEEK_API_KEY", keySet: true, balanceUrl: "https://api.deepseek.com/user/balance", contextWindow: 1_000_000, reasoningProtocol: "", thinking: "", supportedEfforts: [], defaultEffort: "" },
     ],
+    providerPresets: mockProviderPresetViews(),
     permissions: { mode: "ask", allow: ["ls", "read_file"], ask: [], deny: ["Bash(rm:*)"] },
-    sandbox: { bash: "enforce", network: true, workspaceRoot: "", allowWrite: [], shell: "auto" },
+    sandbox: { bash: "enforce", network: true, workspaceRoot: "", allowWrite: [], effectiveWorkspaceRoot: cwd, effectiveWriteRoots: [cwd], shell: "auto" },
     network: {
       proxyMode: "auto",
       proxyUrl: "",
       noProxy: "",
       proxy: { type: "socks5", server: "127.0.0.1", port: 7890, username: "", password: "" },
     },
-    agent: { temperature: 0.2, maxSteps: 0, plannerMaxSteps: 0, systemPrompt: "You are Reasonix, a coding agent.", coldResumePrune: true, reasoningLanguage: "auto" },
+    agent: { temperature: 0.2, maxSteps: 0, plannerMaxSteps: 0, maxSubagentDepth: 2, systemPrompt: "You are Reasonix, a coding agent.", coldResumePrune: true, reasoningLanguage: "auto" },
     bot: {
       enabled: !freshMock,
       model: "",
       toolApprovalMode: "ask",
       maxSteps: 25,
       debounceMs: 1500,
+      queueMode: "steer",
+      queueCap: 20,
+      queueDrop: "summarize",
+      ignoreSelfMessages: true,
+      selfUserIds: {
+        qq: [],
+        feishu: [],
+        weixin: [],
+      },
+      control: {
+        enabled: false,
+        addr: "127.0.0.1:37913",
+        tokenEnv: "REASONIX_BOT_CONTROL_TOKEN",
+      },
+      pairing: {
+        enabled: true,
+        requestTtlMinutes: 60,
+        maxPendingPerPlatform: 3,
+      },
+      routes: [],
       allowlist: {
         enabled: true,
         allowAll: false,
         qqUsers: [],
         feishuUsers: freshMock ? [] : ["ou_mock_user_001"],
         weixinUsers: freshMock ? [] : ["wxid_mock_user_001"],
+        qqApprovers: [],
+        feishuApprovers: [],
+        weixinApprovers: [],
+        qqAdmins: [],
+        feishuAdmins: [],
+        weixinAdmins: [],
         qqGroups: [],
         feishuGroups: [],
         weixinGroups: [],
       },
-      qq: { enabled: false, appId: "", appSecretEnv: "QQ_BOT_APP_SECRET", secretSet: false, sandbox: false },
+      qq: { enabled: false, appId: "", appSecretEnv: "QQ_BOT_APP_SECRET", secretSet: false, sandbox: false, model: "", toolApprovalMode: "ask", workspaceRoot: "", access: { enabled: true, allowAll: false, pairingEnabled: true, users: [], groups: [], approvers: [], admins: [] } },
       feishu: {
         enabled: false,
         domain: "feishu",
@@ -927,10 +1115,11 @@ function makeMockApp(): AppBindings {
           label: "kun",
           enabled: true,
           status: "connected",
-          model: "",
-          toolApprovalMode: "",
-          workspaceRoot: "",
-          credential: {
+	          model: "",
+	          toolApprovalMode: "",
+	          workspaceRoot: "",
+	          access: { enabled: true, allowAll: false, pairingEnabled: true, users: ["ou_mock_user_001"], groups: [], approvers: [], admins: [] },
+	          credential: {
             appId: "cli_mock_lark",
             appSecretEnv: "FEISHU_BOT_APP_SECRET",
             accountId: "",
@@ -961,10 +1150,11 @@ function makeMockApp(): AppBindings {
           label: "kun",
           enabled: true,
           status: "connected",
-          model: "",
-          toolApprovalMode: "",
-          workspaceRoot: "",
-          credential: {
+	          model: "",
+	          toolApprovalMode: "",
+	          workspaceRoot: "",
+	          access: { enabled: true, allowAll: false, pairingEnabled: true, users: ["wxid_mock_user_001"], groups: [], approvers: [], admins: [] },
+	          credential: {
             appId: "",
             appSecretEnv: "",
             accountId: "default",
@@ -1004,7 +1194,7 @@ function makeMockApp(): AppBindings {
     metrics: true,
     memoryCompilerEnabled: true,
     configPath: "~/projects/reasonix/reasonix.toml",
-    providerKinds: ["openai"],
+    providerKinds: ["openai", "anthropic"],
     autoApproveTools: false,
     bypass: false,
   };
@@ -2155,6 +2345,7 @@ function makeMockApp(): AppBindings {
         { name: "model", description: "Switch model", kind: "builtin" as const },
         { name: "effort", description: "Set reasoning effort", kind: "builtin" as const },
         { name: "skill", description: "List skills", kind: "builtin" as const },
+        { name: "plugins", description: "Manage plugin packages", kind: "builtin" as const },
         { name: "explore", description: "Investigate the codebase in an isolated subagent", kind: "skill" as const },
         { name: "review", description: "Review the staged diff", hint: "[focus]", kind: "custom" as const },
       ];
@@ -2164,6 +2355,7 @@ function makeMockApp(): AppBindings {
         servers: capServers.map((s) => ({ ...s })),
         skills: capSkills.map((s) => ({ ...s })),
         skillRoots: capSkillRoots.map((s) => ({ ...s })),
+        plugins: capPlugins.map((p) => ({ ...p })),
       };
     },
     async MCPServers() {
@@ -2175,6 +2367,59 @@ function makeMockApp(): AppBindings {
         skillRoots: capSkillRoots.map((s) => ({ ...s })),
       };
     },
+    async Plugins() {
+      return capPlugins.map((p) => ({ ...p }));
+    },
+    async PlanPluginInstall(source: string, options: PluginInstallOptions) {
+      const name = options.name || source.split("/").filter(Boolean).pop()?.replace(/\.git$/, "") || "plugin";
+      return JSON.stringify({
+        ok: true,
+        status: "planned",
+        kind: "plugin",
+        actions: [{ kind: "plugin", action: "install_plugin_package", name, source, status: "planned" }],
+      });
+    },
+    async InstallPlugin(source: string, options: PluginInstallOptions) {
+      const name = options.name || source.split("/").filter(Boolean).pop()?.replace(/\.git$/, "") || "plugin";
+      const existing = capPlugins.findIndex((p) => p.name === name);
+      const view: PluginView = {
+        name,
+        version: "dev",
+        description: "Mock plugin",
+        source,
+        root: `~/.reasonix/plugins/${name}`,
+        manifestKind: "reasonix",
+        enabled: true,
+        skills: 1,
+        hooks: 0,
+        mcpServers: 0,
+        skillDetails: [{ name: "plan", description: "Plan work before implementation", invocation: "/plan", runAs: "inline" }],
+      };
+      if (existing >= 0) capPlugins[existing] = view;
+      else capPlugins.push(view);
+      return JSON.stringify({ ok: true, status: "done", kind: "plugin", actions: [{ kind: "plugin", name }] });
+    },
+    async RemovePlugin(name: string) {
+      capPlugins = capPlugins.filter((p) => p.name !== name);
+    },
+    async SetPluginEnabled(name: string, enabled: boolean) {
+      capPlugins = capPlugins.map((p) => p.name === name ? { ...p, enabled } : p);
+    },
+    async UpdatePlugin(name: string) {
+      capPlugins = capPlugins.map((p) => p.name === name ? { ...p, version: p.version || "dev" } : p);
+      return JSON.stringify({ ok: true, status: "done", kind: "plugin", name });
+    },
+    async PluginDoctor(name: string) {
+      return capPlugins.find((p) => p.name === name) || {
+        name,
+        root: "",
+        enabled: false,
+        skills: 0,
+        hooks: 0,
+        mcpServers: 0,
+        error: "plugin is not installed",
+      };
+    },
     async AddMCPServer(input: MCPServerInput) {
       const tools = input.transport === "stdio" ? 3 : 5;
       capServers.push({
@@ -2281,6 +2526,9 @@ function makeMockApp(): AppBindings {
     async PickSkillFolder() {
       return "~/my-skills";
     },
+    async PickPluginFolder() {
+      return "~/plugins/superpowers";
+    },
     async AddSkillPath(path: string) {
       const dir = path.trim() || "~/my-skills";
       if (!capSkillRoots.some((r) => r.scope === "custom" && r.dir === dir)) {
@@ -2451,14 +2699,20 @@ function makeMockApp(): AppBindings {
     async RevealPath(path: string) {
       console.info("mock RevealPath", path);
     },
-    async SavePastedImage(_dataUrl: string) {
-      return ".reasonix/attachments/mock.png";
+    async SavePastedImage(dataUrl: string) {
+      const path = `.reasonix/attachments/mock-${mockAttachmentDataURLs.size + 1}.png`;
+      mockAttachmentDataURLs.set(path, dataUrl);
+      return path;
     },
     async SaveClipboardImage() {
-      return ".reasonix/attachments/mock-clipboard.png";
+      const path = `.reasonix/attachments/mock-clipboard-${mockAttachmentDataURLs.size + 1}.png`;
+      mockAttachmentDataURLs.set(path, mockPreviewImageDataURL);
+      return path;
     },
-    async SavePastedFile(name: string, _dataUrl: string) {
-      return `.reasonix/attachments/mock-${name}`;
+    async SavePastedFile(name: string, dataUrl: string) {
+      const path = `.reasonix/attachments/mock-${name}`;
+      mockAttachmentDataURLs.set(path, dataUrl);
+      return path;
     },
     async PickExportFile(defaultFilename: string, _mimeType: string) {
       return defaultFilename;
@@ -2485,10 +2739,12 @@ function makeMockApp(): AppBindings {
         const tokenName = name.replace(/[^\w.-]+/g, "-") || "folder";
         return { kind: "workspace" as const, path: `__reasonix_external_folder/mock/${tokenName}`, isDir: true, displayPath: path };
       }
-      return { kind: "attachment" as const, path: `.reasonix/attachments/mock-${name}` };
+      const attachmentPath = `.reasonix/attachments/mock-${name}`;
+      mockAttachmentDataURLs.set(attachmentPath, mockPreviewImageDataURL);
+      return { kind: "attachment" as const, path: attachmentPath };
     },
-    async AttachmentDataURL(_path: string) {
-      return "data:image/png;base64,iVBORw0KGgo=";
+    async AttachmentDataURL(path: string) {
+      return mockAttachmentDataURLs.get(path) ?? mockPreviewImageDataURL;
     },
         async Models() {
           const active = mockTabs.find((tab) => tab.active) ?? mockTabs[0];
@@ -2687,6 +2943,9 @@ function makeMockApp(): AppBindings {
     async SetSubagentEffort(level: string) {
       settings.subagentEffort = level;
     },
+    async SetMaxSubagentDepth(depth: number) {
+      settings.agent = { ...settings.agent, maxSubagentDepth: depth <= 1 ? 1 : 2 };
+    },
     async SetAutoPlan(mode: string) {
       settings.autoPlan = mode;
     },
@@ -2701,7 +2960,7 @@ function makeMockApp(): AppBindings {
     },
     async AddOfficialProviderAccess(kind: string, key: string) {
       const templates: Record<string, ProviderView> = {
-        deepseek: { name: "deepseek", builtIn: true, added: true, kind: "openai", baseUrl: "https://api.deepseek.com", modelsUrl: "", models: ["deepseek-v4-flash", "deepseek-v4-pro"], visionModels: [], visionModelsConfigured: false, default: "deepseek-v4-flash", apiKeyEnv: "DEEPSEEK_API_KEY", keySet: !!key.trim(), balanceUrl: "https://api.deepseek.com/user/balance", contextWindow: 1_000_000, reasoningProtocol: "", supportedEfforts: [], defaultEffort: "" },
+        deepseek: { name: "deepseek", builtIn: true, added: true, kind: "openai", baseUrl: "https://api.deepseek.com", modelsUrl: "", models: ["deepseek-v4-flash", "deepseek-v4-pro"], visionModels: [], visionModelsConfigured: false, default: "deepseek-v4-flash", apiKeyEnv: "DEEPSEEK_API_KEY", keySet: !!key.trim(), balanceUrl: "https://api.deepseek.com/user/balance", contextWindow: 1_000_000, reasoningProtocol: "", thinking: "", supportedEfforts: [], defaultEffort: "" },
       };
       const next = templates[kind];
       if (!next) throw new Error(`unknown official provider template ${kind}`);
@@ -2710,12 +2969,47 @@ function makeMockApp(): AppBindings {
       else settings.providers.push(next);
       return "";
     },
+    async AddProviderPresetAccess(id: string, key: string) {
+      const preset = settings.providerPresets.find((p) => p.id === id);
+      if (!preset) throw new Error(`unknown provider preset ${id}`);
+      const next = cloneMockProviderTemplate(id, key);
+      if (!next) throw new Error(`unknown provider preset ${id}`);
+      const i = settings.providers.findIndex((x) => x.name === next.name);
+      if (i >= 0) settings.providers[i] = { ...settings.providers[i], ...next, keySet: next.keySet || settings.providers[i].keySet };
+      else settings.providers.push(next);
+      preset.added = true;
+      preset.status = "installed";
+      preset.statusProviderNames = [...preset.providerNames];
+      preset.keySet = preset.keySet || !!key.trim();
+      preset.configured = !preset.requiresKey || preset.keySet;
+      return "";
+    },
+    async ResetProviderPresetAccess(id: string) {
+      const preset = settings.providerPresets.find((p) => p.id === id);
+      if (!preset) throw new Error(`unknown provider preset ${id}`);
+      const next = cloneMockProviderTemplate(id, "");
+      if (!next) throw new Error(`unknown provider preset ${id}`);
+      const i = settings.providers.findIndex((x) => x.name === next.name);
+      if (i < 0) throw new Error(`provider preset ${id} cannot be reset because no same-name provider exists`);
+      const existing = settings.providers[i];
+      settings.providers[i] = {
+        ...next,
+        added: true,
+        keySet: existing.apiKeyEnv === next.apiKeyEnv ? existing.keySet : next.keySet,
+      };
+      preset.added = true;
+      preset.status = "installed";
+      preset.statusProviderNames = [...preset.providerNames];
+      preset.keySet = preset.keySet || settings.providers[i].keySet;
+      preset.configured = !preset.requiresKey || preset.keySet;
+    },
     async FetchProviderModels(p: ProviderView) {
       if (!p.baseUrl.trim()) throw new Error(t("settings.fetchModelsMissingBaseUrl"));
       if (providerRequiresKey(p) && !p.apiKeyEnv.trim()) throw new Error(t("settings.fetchModelsMissingKeyEnv"));
       await delay(350);
       if (p.baseUrl.includes("deepseek")) return ["deepseek-v4-flash", "deepseek-v4-pro"];
-      if (p.baseUrl.includes("mimo") || p.baseUrl.includes("xiaomimimo")) return ["mimo-v2.5", "mimo-v2.5-pro"];
+      if (p.baseUrl.includes("token-plan")) return ["mimo-v2.5", "mimo-v2.5-pro"];
+      if (p.baseUrl.includes("xiaomimimo")) return ["mimo-v2.5-pro", "mimo-v2.5"];
       return ["gpt-5", "gpt-5-mini", "qwen3-coder"];
     },
     async DeleteProvider(name: string) {
@@ -2748,8 +3042,10 @@ function makeMockApp(): AppBindings {
       const k = list as "allow" | "ask" | "deny";
       settings.permissions[k] = settings.permissions[k].filter((r) => r !== rule);
     },
+        async ReloadSettings() {},
         async SetSandbox(bash: string, network: boolean, workspaceRoot: string, allowWrite: string[], shell: string) {
-          settings.sandbox = { bash, network, workspaceRoot, allowWrite, shell };
+          const effectiveWorkspaceRoot = workspaceRoot.trim() || cwd;
+          settings.sandbox = { bash, network, workspaceRoot, allowWrite, effectiveWorkspaceRoot, effectiveWriteRoots: [effectiveWorkspaceRoot, ...allowWrite], shell };
         },
         async SetNetwork(n: NetworkView) {
           settings.network = n;
@@ -2823,10 +3119,11 @@ function makeMockApp(): AppBindings {
             label: domain === "lark" ? "Lark" : domain === "weixin" ? "微信" : "飞书",
             enabled: true,
             status: "connected",
-            model: "",
-            toolApprovalMode: "",
-            workspaceRoot: "",
-            credential: {
+	            model: "",
+	            toolApprovalMode: "",
+	            workspaceRoot: "",
+	            access: { enabled: true, allowAll: false, pairingEnabled: true, users: [provider === "weixin" ? "wxid_mock_user_001" : "ou_mock_user_001"], groups: [], approvers: [], admins: [] },
+	            credential: {
               appId: provider === "feishu" ? "cli_mock" : "",
               appSecretEnv: provider === "feishu" ? (domain === "lark" ? "LARK_BOT_APP_SECRET" : "FEISHU_BOT_APP_SECRET") : "",
               accountId: provider === "weixin" ? "mock-account" : "",
@@ -2971,7 +3268,7 @@ function makeMockApp(): AppBindings {
     },
     async OpenDownloadPage() {
       if (typeof window !== "undefined") {
-        window.open("https://github.com/esengine/reasonix/releases/latest", "_blank", "noopener");
+        window.open("https://reasonix.io/#start", "_blank", "noopener");
       }
     },
     // Dev seam: drives the overlay flow in the browser until ConnectKey sets the

desktop/frontend/src/lib/toast.tsx

@@ -4,11 +4,19 @@ export interface Toast {
   id: number;
   text: string;
   level: "info" | "warn" | "error";
+  actionLabel?: string;
+  onAction?: () => void;
+}
+
+export interface ToastOptions {
+  actionLabel?: string;
+  onAction?: () => void;
+  durationMs?: number;
 }
 
 export interface ToastContextValue {
   toasts: Toast[];
-  showToast: (text: string, level?: Toast["level"]) => void;
+  showToast: (text: string, level?: Toast["level"], options?: ToastOptions) => void;
 }
 
 const ToastContext = createContext<ToastContextValue>({ toasts: [], showToast: () => {} });
@@ -23,13 +31,13 @@ export function ToastProvider({ children }: { children: ReactNode }) {
   const [toasts, setToasts] = useState<Toast[]>([]);
   const timers = useRef(new Map<number, ReturnType<typeof setTimeout>>());
 
-  const showToast = useCallback((text: string, level: Toast["level"] = "info") => {
+  const showToast = useCallback((text: string, level: Toast["level"] = "info", options: ToastOptions = {}) => {
     const id = nextId++;
-    setToasts((prev) => [...prev, { id, text, level }]);
+    setToasts((prev) => [...prev, { id, text, level, actionLabel: options.actionLabel, onAction: options.onAction }]);
     const timer = setTimeout(() => {
       setToasts((prev) => prev.filter((t) => t.id !== id));
       timers.current.delete(id);
-    }, 2500);
+    }, options.durationMs ?? (options.actionLabel ? 8000 : 2500));
     timers.current.set(id, timer);
   }, []);
 
@@ -49,6 +57,19 @@ export function ToastProvider({ children }: { children: ReactNode }) {
             {t.level === "warn" && <span className="toast__icon">⚠️</span>}
             {t.level === "error" && <span className="toast__icon">❌</span>}
             <span className="toast__text">{t.text}</span>
+            {t.actionLabel && t.onAction && (
+              <button
+                type="button"
+                className="toast__action"
+                onClick={(event) => {
+                  event.stopPropagation();
+                  dismissToast(t.id);
+                  t.onAction?.();
+                }}
+              >
+                {t.actionLabel}
+              </button>
+            )}
           </div>
         ))}
       </div>

desktop/frontend/src/lib/types.ts

@@ -208,6 +208,10 @@ export interface TabMeta {
   goal?: string;
   goalStatus?: GoalStatus;
   autoResearch?: AutoResearchCompactView;
+  recovered?: boolean;
+  recoveryReason?: string;
+  recoveryDigest?: string;
+  recoveryParentId?: string;
   startupErr?: string;
   active: boolean;
   cwd: string;
@@ -228,6 +232,10 @@ export interface ProjectNode {
   running?: boolean;
   status?: ProjectTopicStatus;
   pinned?: boolean;
+  recovered?: boolean;
+  recoveryReason?: string;
+  recoveryDigest?: string;
+  recoveryParentId?: string;
   children?: ProjectNode[];
 }
 
@@ -239,6 +247,23 @@ export interface TopicMeta {
   createdAt: number;
 }
 
+export interface SessionRecoveryEvent {
+  originalPath?: string;
+  recoveryPath: string;
+  scope?: string;
+  workspaceRoot?: string;
+  topicId?: string;
+  topicTitle?: string;
+  recoveryReason?: string;
+  recoveryDigest?: string;
+  recoveryParentId?: string;
+  existing?: boolean;
+}
+
+export interface SessionRecoveryFailedEvent {
+  reason?: "lease_held" | "lease_unavailable" | string;
+}
+
 export interface ContextPanelInfo {
   usedTokens: number;
   windowTokens: number;
@@ -683,11 +708,55 @@ export interface CapabilitiesView {
   servers: ServerView[];
   skills: SkillView[];
   skillRoots: SkillRootView[];
+  plugins: PluginView[];
 }
 export interface SkillsSettingsView {
   skills: SkillView[];
   skillRoots: SkillRootView[];
 }
+export interface PluginView {
+  name: string;
+  version?: string;
+  description?: string;
+  source?: string;
+  root: string;
+  manifestKind?: string;
+  enabled: boolean;
+  skills: number;
+  hooks: number;
+  mcpServers: number;
+  skillDetails?: PluginSkillView[];
+  hookDetails?: PluginHookView[];
+  mcpServerDetails?: PluginMCPServerView[];
+  warnings?: string[];
+  error?: string;
+}
+export interface PluginSkillView {
+  name: string;
+  description?: string;
+  path?: string;
+  invocation?: string;
+  runAs?: string;
+}
+export interface PluginHookView {
+  event: string;
+  match?: string;
+  command?: string;
+  contextFile?: string;
+  description?: string;
+}
+export interface PluginMCPServerView {
+  name: string;
+  transport?: string;
+  command?: string;
+  url?: string;
+}
+export interface PluginInstallOptions {
+  dryRun?: boolean;
+  link?: boolean;
+  replace?: boolean;
+  name?: string;
+}
 export interface MCPServerInput {
   name: string;
   transport: string; // stdio | http | sse
@@ -791,7 +860,7 @@ export interface MemoryView {
 }
 
 // SettingsTab is the top-level navigation item in the Settings Centre modal.
-export type SettingsTab = "general" | "models" | "providers" | "bots" | "mcp" | "skills" | "memory" | "hooks" | "shortcuts" | "permissions" | "sandbox" | "network" | "appearance" | "updates";
+export type SettingsTab = "general" | "models" | "providers" | "bots" | "mcp" | "skills" | "plugins" | "memory" | "hooks" | "shortcuts" | "permissions" | "sandbox" | "network" | "appearance" | "updates";
 
 // Settings panel payloads (desktop/settings_app.go).
 export interface ProviderView {
@@ -808,6 +877,8 @@ export interface ProviderView {
   default: string;
   apiKeyEnv: string;
   headers?: Record<string, string> | null; // optional extra request headers for compatible gateways
+  extraBody?: Record<string, unknown> | null; // optional extra top-level request body fields for compatible gateways
+  authHeader?: boolean; // Anthropic-compatible: send Authorization: Bearer instead of x-api-key
   keySet: boolean; // the env var currently resolves to a value
   requiresKey?: boolean; // false for explicit no-auth providers
   configured?: boolean; // selectable: key is set or no key is required
@@ -816,11 +887,29 @@ export interface ProviderView {
   balanceUrl: string; // optional wallet-balance endpoint; "" disables the readout
   contextWindow: number;
   reasoningProtocol: string; // auto|deepseek|openai|none; empty = auto/model registry
+  thinking: string; // provider-specific thinking override: ""|enabled|disabled|adaptive
   supportedEfforts: string[]; // custom /effort levels; empty = use built-in Kind/BaseURL default
   defaultEffort: string; // /effort level when user picks "auto" or unset; "" = supportedEfforts[0]
   modelOverrides?: ProviderModelOverrideView[] | null;
 }
 
+export interface ProviderPresetView {
+  id: string;
+  label: string;
+  description: string;
+  keyEnv: string;
+  providerNames: string[];
+  models: string[];
+  added: boolean;
+  status?: "available" | "installed" | "installed_modified" | "name_conflict" | "similar_existing";
+  statusProviderNames?: string[];
+  keySet: boolean;
+  requiresKey?: boolean;
+  configured?: boolean;
+  keySource?: string;
+  keySourcePath?: string;
+}
+
 export interface ProviderModelOverrideView {
   model: string;
   reasoningProtocol: string;
@@ -859,6 +948,8 @@ export interface SandboxView {
   network: boolean;
   workspaceRoot: string;
   allowWrite: string[];
+  effectiveWorkspaceRoot: string;
+  effectiveWriteRoots: string[];
   shell: string; // "auto" | "bash" | "powershell" | "pwsh"
 }
 
@@ -881,6 +972,7 @@ export interface AgentView {
   temperature: number;
   maxSteps: number;
   plannerMaxSteps: number;
+  maxSubagentDepth: number;
   systemPrompt: string;
   coldResumePrune: boolean;
   reasoningLanguage: string; // "auto" | "zh" | "en"
@@ -892,17 +984,67 @@ export interface BotAllowlistView {
   qqUsers: string[];
   feishuUsers: string[];
   weixinUsers: string[];
+  qqApprovers: string[];
+  feishuApprovers: string[];
+  weixinApprovers: string[];
+  qqAdmins: string[];
+  feishuAdmins: string[];
+  weixinAdmins: string[];
   qqGroups: string[];
   feishuGroups: string[];
   weixinGroups: string[];
 }
 
+export interface BotAccessView {
+  enabled: boolean;
+  allowAll: boolean;
+  pairingEnabled: boolean;
+  users: string[];
+  groups: string[];
+  approvers: string[];
+  admins: string[];
+}
+
+export interface BotSelfUserIDsView {
+  qq: string[];
+  feishu: string[];
+  weixin: string[];
+}
+
+export interface BotPairingView {
+  enabled: boolean;
+  requestTtlMinutes: number;
+  maxPendingPerPlatform: number;
+}
+
+export interface BotControlView {
+  enabled: boolean;
+  addr: string;
+  tokenEnv: string;
+}
+
+export interface BotRouteView {
+  connectionId: string;
+  platform: string;
+  chatType: string;
+  chatId: string;
+  userId: string;
+  threadId: string;
+  model: string;
+  toolApprovalMode: ToolApprovalMode | "" | string;
+  workspaceRoot: string;
+}
+
 export interface QQBotView {
   enabled: boolean;
   appId: string;
   appSecretEnv: string;
   secretSet: boolean;
   sandbox: boolean;
+  model: string;
+  toolApprovalMode: ToolApprovalMode | "" | string;
+  workspaceRoot: string;
+  access: BotAccessView;
 }
 
 export interface FeishuBotView {
@@ -955,6 +1097,7 @@ export interface BotConnectionView {
   model: string;
   toolApprovalMode: ToolApprovalMode | "" | string;
   workspaceRoot: string;
+  access: BotAccessView;
   credential: BotConnectionCredentialView;
   sessionMappings: BotConnectionSessionMappingView[];
   lastError: string;
@@ -968,6 +1111,14 @@ export interface BotSettingsView {
   toolApprovalMode: ToolApprovalMode | "" | string;
   maxSteps: number;
   debounceMs: number;
+  queueMode: string;
+  queueCap: number;
+  queueDrop: string;
+  ignoreSelfMessages: boolean;
+  selfUserIds: BotSelfUserIDsView;
+  control: BotControlView;
+  pairing: BotPairingView;
+  routes: BotRouteView[];
   allowlist: BotAllowlistView;
   qq: QQBotView;
   feishu: FeishuBotView;
@@ -1043,6 +1194,7 @@ export interface SettingsView {
   autoPlan: string;
   providers: ProviderView[];
   officialProviders: ProviderView[];
+  providerPresets: ProviderPresetView[];
   permissions: PermissionsView;
   sandbox: SandboxView;
   network: NetworkView;

desktop/frontend/src/lib/useController.ts

@@ -168,6 +168,7 @@ type RuntimeMetaSnapshot = {
   running: boolean;
   pendingPrompt?: boolean;
   backgroundJobs?: number;
+  cancelRequested?: boolean;
   cancellable?: boolean;
 };
 
@@ -354,6 +355,18 @@ type Action =
   | { type: "clearAsk" }
   | { type: "reset" };
 
+function backendStatusFromRuntimeMeta(meta: RuntimeMetaSnapshot): Extract<Action, { type: "backend_status" }> {
+  const foregroundRunning = foregroundRunningFromRuntimeMeta(meta);
+  return {
+    type: "backend_status",
+    running: foregroundRunning,
+    pendingPrompt: Boolean(meta.pendingPrompt),
+    backgroundJobs: meta.backgroundJobs ?? 0,
+    cancelRequested: Boolean(meta.cancelRequested),
+    cancellable: foregroundRunning,
+  };
+}
+
 // ---- reducer helpers (unchanged logic) ----
 
 export function historyMessagesToItems(messages: HistoryMessage[], idPrefix: string, startSeq = 0): { items: Item[]; seq: number } {
@@ -1192,6 +1205,22 @@ export function useController() {
     return tabs.find((tab) => tab.active) ?? tabs[0];
   }, []);
 
+  const dispatchRuntimeStatusForTab = useCallback((tabId: string, tab: RuntimeMetaSnapshot) => {
+    const foregroundRunning = foregroundRunningFromRuntimeMeta(tab);
+    dispatchTo(tabId, {
+      type: "backend_status",
+      running: foregroundRunning,
+      pendingPrompt: Boolean(tab.pendingPrompt),
+      backgroundJobs: tab.backgroundJobs ?? 0,
+      cancelRequested: Boolean(tab.cancelRequested),
+      cancellable: foregroundRunning,
+    });
+    // backend_status reconciliation can clear a live prompt from frontend state.
+    // If the backend is still blocked, ask it to replay the approval/ask event.
+    if (tab.pendingPrompt) replayPendingPromptsForActiveTab(tabId);
+    return foregroundRunning;
+  }, [dispatchTo]);
+
   const waitForTabReady = useCallback(async (tabId: string): Promise<void> => {
     for (let attempt = 0; attempt < 60; attempt += 1) {
       const tabs = asArray(await app.ListTabs().catch(() => [] as TabMeta[]));
@@ -1216,12 +1245,14 @@ export function useController() {
     confirmBackendActiveTab(active.id);
     dispatchTo(active.id, { type: "optimistic_meta", meta: metaFromTab(active, statesRef.current.get(active.id)?.meta) });
     const preserveCachedHistory = options.preserveCachedHistory ?? !reset;
+    if (!reset) dispatchRuntimeStatusForTab(active.id, active);
     await loadSessionDataForTab(active.id, reset, "startup", {
       preserveCachedHistory,
       sessionPath: active.sessionPath,
     });
+    if (reset) dispatchRuntimeStatusForTab(active.id, active);
     return active.id;
-  }, [activeTabFromBackend, confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [activeTabFromBackend, confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab]);
 
   const reconcileTabRuntime = useCallback(async (
     tabId: string,
@@ -1233,19 +1264,8 @@ export function useController() {
     if (!tab) return undefined;
     const local = statesRef.current.get(tabId);
     const needsInitialLoad = !local?.meta;
-    const foregroundRunning = foregroundRunningFromRuntimeMeta(tab);
+    const foregroundRunning = dispatchRuntimeStatusForTab(tabId, tab);
     const missedTurnDone = Boolean(local?.running && !foregroundRunning);
-    dispatchTo(tabId, {
-      type: "backend_status",
-      running: foregroundRunning,
-      pendingPrompt: Boolean(tab.pendingPrompt),
-      backgroundJobs: tab.backgroundJobs ?? 0,
-      cancelRequested: Boolean(tab.cancelRequested),
-      cancellable: foregroundRunning,
-    });
-    // backend_status reconciliation can clear a live prompt from frontend state.
-    // If the backend is still blocked, ask it to replay the approval/ask event.
-    if (tab.pendingPrompt) replayPendingPromptsForActiveTab(tabId);
     if (hydrateSessionData && (needsInitialLoad || missedTurnDone)) {
       await loadSessionDataForTab(tabId, missedTurnDone, "startup");
       return tabs;
@@ -1259,7 +1279,7 @@ export function useController() {
     if (effort) dispatchTo(tabId, { type: "effort", effort });
     if (balance) dispatchTo(tabId, { type: "balance", balance });
     return tabs;
-  }, [dispatchTo, loadSessionDataForTab]);
+  }, [dispatchRuntimeStatusForTab, loadSessionDataForTab]);
 
   const clearCancelReconcileTimer = useCallback((tabId: string) => {
     const timer = cancelReconcileTimers.current.get(tabId);
@@ -1422,9 +1442,10 @@ export function useController() {
       setActiveTabId(active.id);
       activeTabIdRef.current = active.id;
       confirmBackendActiveTab(active.id);
+      dispatchRuntimeStatusForTab(active.id, active);
       return sendToTab(active.id, displayText, submitText);
     });
-  }, [activeTabFromBackend, activeTabId, confirmBackendActiveTab, sendToTab]);
+  }, [activeTabFromBackend, activeTabId, confirmBackendActiveTab, dispatchRuntimeStatusForTab, sendToTab]);
 
   const runShell = useCallback(async (command: string) => {
     if (!activeTabId) throw new Error("workspace is still starting");
@@ -1731,11 +1752,13 @@ export function useController() {
           setActiveTabId(tab.id);
           activeTabIdRef.current = tab.id;
           confirmBackendActiveTab(tab.id);
+          dispatchRuntimeStatusForTab(tab.id, tab);
           // The fork's controller builds in a background goroutine: an immediate
           // load reads empty history, and the ready-event fallback can still
           // target the source tab, leaving the fork blank (#3742).
           await waitForTabReady(tab.id);
           await loadSessionDataForTab(tab.id, true);
+          await reconcileTabRuntime(tab.id, { hydrateSessionData: false });
         } else {
           await syncActiveTabFromBackend(true);
         }
@@ -1758,7 +1781,7 @@ export function useController() {
     } finally {
       dispatchTo(sourceTabId, { type: "message_action_done" });
     }
-  }, [activeTabId, confirmBackendActiveTab, dispatchTo, loadSessionDataForTab, syncActiveTabFromBackend, waitForBackendActiveTab, waitForTabReady]);
+  }, [activeTabId, confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime, syncActiveTabFromBackend, waitForBackendActiveTab, waitForTabReady]);
 
   // Tab management: switch preserves per-tab state; open creates it.
   const switchTab = useCallback(async (tabId: string, optimisticTab?: TabMeta): Promise<TabMeta[] | undefined> => {
@@ -1772,11 +1795,25 @@ export function useController() {
     dispatchTo(tabId, { type: "backend_activation_start" });
     if (optimisticTab) {
       dispatchTo(tabId, { type: "optimistic_meta", meta: metaFromTab(optimisticTab, statesRef.current.get(tabId)?.meta) });
+      const optimisticStatus = backendStatusFromRuntimeMeta(optimisticTab);
+      if (optimisticStatus.running) dispatchTo(tabId, optimisticStatus);
     }
     dispatchTo(tabId, { type: "hydrate_start", reason: "switch-tab" });
     addBreadcrumb("tab.switch", `active-rendered ${tabId} ms=${Date.now() - startedAt}`);
     const backendActivation = app.SetActiveTab(tabId)
-      .then(() => {
+      .then(async () => {
+        if (activeTabIdRef.current !== tabId) {
+          const currentTabId = activeTabIdRef.current;
+          if (currentTabId) {
+            await app.SetActiveTab(currentTabId).then(() => {
+              if (activeTabIdRef.current === currentTabId) confirmBackendActiveTab(currentTabId);
+            }).catch((err) => {
+              addBreadcrumb("tab.switch", `set-active-stale-reassert-failed ${currentTabId}: ${errorMessage(err)}`);
+            });
+          }
+          addBreadcrumb("tab.switch", `set-active-stale ${tabId} current=${currentTabId ?? ""} ms=${Date.now() - startedAt}`);
+          return false;
+        }
         confirmBackendActiveTab(tabId);
         addBreadcrumb("tab.switch", `set-active-done ${tabId} ms=${Date.now() - startedAt}`);
         return true;
@@ -1820,13 +1857,16 @@ export function useController() {
     activeTabIdRef.current = meta.id;
     confirmBackendActiveTab(meta.id);
     dispatchTo(meta.id, { type: "optimistic_meta", meta: metaFromTab(meta, statesRef.current.get(meta.id)?.meta) });
-    void loadSessionDataForTab(meta.id, isNewTab, "open-topic", {
+    dispatchRuntimeStatusForTab(meta.id, meta);
+    const load = loadSessionDataForTab(meta.id, isNewTab, "open-topic", {
       placeholderItems: isNewTab ? prevItems : undefined,
       preserveCachedHistory,
       sessionPath: meta.sessionPath,
     });
+    if (isNewTab) void load.then(() => reconcileTabRuntime(meta.id, { hydrateSessionData: false })).catch(() => {});
+    else void load;
     return meta;
-  }, [confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime]);
 
   const openGlobalTab = useCallback(async (topicId: string): Promise<TabMeta> => {
     const meta = await app.OpenGlobalTab(topicId);
@@ -1838,13 +1878,16 @@ export function useController() {
     activeTabIdRef.current = meta.id;
     confirmBackendActiveTab(meta.id);
     dispatchTo(meta.id, { type: "optimistic_meta", meta: metaFromTab(meta, statesRef.current.get(meta.id)?.meta) });
-    void loadSessionDataForTab(meta.id, isNewTab, "open-topic", {
+    dispatchRuntimeStatusForTab(meta.id, meta);
+    const load = loadSessionDataForTab(meta.id, isNewTab, "open-topic", {
       placeholderItems: isNewTab ? prevItems : undefined,
       preserveCachedHistory,
       sessionPath: meta.sessionPath,
     });
+    if (isNewTab) void load.then(() => reconcileTabRuntime(meta.id, { hydrateSessionData: false })).catch(() => {});
+    else void load;
     return meta;
-  }, [confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime]);
 
   const openTopicSession = useCallback(async (scope: string, workspaceRoot: string, topicId: string, sessionPath: string): Promise<TabMeta> => {
     const meta = await app.OpenTopicSession(scope, workspaceRoot, topicId, sessionPath);
@@ -1856,13 +1899,16 @@ export function useController() {
     activeTabIdRef.current = meta.id;
     confirmBackendActiveTab(meta.id);
     dispatchTo(meta.id, { type: "optimistic_meta", meta: metaFromTab(meta, statesRef.current.get(meta.id)?.meta) });
-    void loadSessionDataForTab(meta.id, isNewTab, "open-topic", {
+    dispatchRuntimeStatusForTab(meta.id, meta);
+    const load = loadSessionDataForTab(meta.id, isNewTab, "open-topic", {
       placeholderItems: isNewTab ? prevItems : undefined,
       preserveCachedHistory,
       sessionPath: meta.sessionPath,
     });
+    if (isNewTab) void load.then(() => reconcileTabRuntime(meta.id, { hydrateSessionData: false })).catch(() => {});
+    else void load;
     return meta;
-  }, [confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime]);
 
   const activateTopic = useCallback(async (scope: string, workspaceRoot: string, topicId: string, sessionPath = ""): Promise<TabMeta> => {
     const meta = await app.ActivateTopic(scope, workspaceRoot, topicId, sessionPath);
@@ -1876,21 +1922,28 @@ export function useController() {
     activeTabIdRef.current = meta.id;
     confirmBackendActiveTab(meta.id);
     dispatchTo(meta.id, { type: "optimistic_meta", meta: metaFromTab(meta, statesRef.current.get(meta.id)?.meta) });
-    void loadSessionDataForTab(meta.id, true, "open-topic", { placeholderItems: prevItems });
+    dispatchRuntimeStatusForTab(meta.id, meta);
+    void loadSessionDataForTab(meta.id, true, "open-topic", { placeholderItems: prevItems })
+      .then(() => reconcileTabRuntime(meta.id, { hydrateSessionData: false }))
+      .catch(() => {});
     return meta;
-  }, [confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime]);
 
   // Ensure a blank tab exists for the given scope — reuses an existing one
   // or creates a new tab, then loads its session data.
   const ensureBlankTab = useCallback(async (scope: string, workspaceRoot: string): Promise<TabMeta> => {
     const meta = await app.EnsureBlankTab(scope, workspaceRoot);
+    const isNewTab = !statesRef.current.has(meta.id);
     setActiveTabId(meta.id);
     activeTabIdRef.current = meta.id;
     confirmBackendActiveTab(meta.id);
     dispatchTo(meta.id, { type: "optimistic_meta", meta: metaFromTab(meta, statesRef.current.get(meta.id)?.meta) });
-    void loadSessionDataForTab(meta.id, !statesRef.current.has(meta.id), "open-topic");
+    dispatchRuntimeStatusForTab(meta.id, meta);
+    const load = loadSessionDataForTab(meta.id, isNewTab, "open-topic");
+    if (isNewTab) void load.then(() => reconcileTabRuntime(meta.id, { hydrateSessionData: false })).catch(() => {});
+    else void load;
     return meta;
-  }, [confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime]);
 
   const ensureBlankSurface = useCallback(async (scope: string, workspaceRoot: string): Promise<TabMeta> => {
     const meta = await app.EnsureBlankSurface(scope, workspaceRoot);
@@ -1901,9 +1954,12 @@ export function useController() {
     activeTabIdRef.current = meta.id;
     confirmBackendActiveTab(meta.id);
     dispatchTo(meta.id, { type: "optimistic_meta", meta: metaFromTab(meta, statesRef.current.get(meta.id)?.meta) });
-    void loadSessionDataForTab(meta.id, true, "open-topic");
+    dispatchRuntimeStatusForTab(meta.id, meta);
+    void loadSessionDataForTab(meta.id, true, "open-topic")
+      .then(() => reconcileTabRuntime(meta.id, { hydrateSessionData: false }))
+      .catch(() => {});
     return meta;
-  }, [confirmBackendActiveTab, dispatchTo, loadSessionDataForTab]);
+  }, [confirmBackendActiveTab, dispatchRuntimeStatusForTab, dispatchTo, loadSessionDataForTab, reconcileTabRuntime]);
 
   const closeTab = useCallback(async (tabId: string) => {
     try {

desktop/frontend/src/locales/en.ts

@@ -205,6 +205,7 @@ export const en = {
   "caps.mcpSummary": "{connected} connected · {failed} issues · {tools} tools",
   "caps.connectorsTab": "MCP servers",
   "caps.skillsTab": "Agent skills",
+  "caps.pluginsTab": "Plugins",
   "caps.servers": "MCP servers",
   "caps.noServers": "No MCP servers configured. Add one with /mcp add.",
   "caps.availableServers": "Available MCPs",
@@ -311,6 +312,62 @@ export const en = {
   "caps.trustedReadOnlyTitle": "Remembered as read-only for future planning and research",
   "caps.searchSkills": "Search skills…",
   "caps.noSkillMatches": "No matching skills.",
+  "caps.pluginRefresh": "Refresh plugins",
+  "caps.pluginInstallTitle": "Install plugin",
+  "caps.pluginInstallHint": "Add a local plugin folder or install from a Git repository.",
+  "caps.pluginInstallMethod": "Plugin install method",
+  "caps.pluginInstallLocal": "Local folder",
+  "caps.pluginInstallGit": "Git repository",
+  "caps.pluginSource": "Plugin source",
+  "caps.pluginGitSource": "Git repository URL",
+  "caps.pluginSourcePlaceholder": "git:github.com/obra/superpowers",
+  "caps.pluginChooseLocalFolder": "Choose plugin folder",
+  "caps.pluginLocalFolder": "Local plugin folder",
+  "caps.pluginNoLocalFolder": "No folder selected",
+  "caps.pluginInstallName": "Install name (optional)",
+  "caps.pluginInstallNamePlaceholder": "install name (optional)",
+  "caps.pluginLink": "Developer mode: link source folder",
+  "caps.pluginLinkHint": "Do not copy the plugin; use the selected folder directly. Useful for plugin debugging. Moving or deleting the folder breaks the plugin.",
+  "caps.pluginReplace": "Overwrite same-name plugin",
+  "caps.pluginReplaceHint": "When a plugin with the same name already exists, allow this source to replace it.",
+  "caps.pluginPreview": "Preview",
+  "caps.pluginInstall": "Install plugin",
+  "caps.noPluginsTitle": "No plugins installed",
+  "caps.noPluginsHint": "Install a plugin package from Git or a local path to add skills, hooks, and MCP servers through one managed unit.",
+  "caps.installedPlugins": "Installed plugins",
+  "caps.pluginsSummary": "{enabled} enabled · {total} installed · {issues} issues",
+  "caps.pluginCounts": "{skills} skills · {hooks} hooks · {mcps} MCPs",
+  "caps.pluginNoCapabilities": "No exported capabilities",
+  "caps.pluginWarnings": "{count} warnings",
+  "caps.pluginEnable": "Enable plugin",
+  "caps.pluginDisable": "Disable plugin",
+  "caps.pluginEnabled": "Enabled",
+  "caps.pluginDisabled": "Disabled",
+  "caps.pluginVersion": "Version",
+  "caps.pluginRoot": "Install path",
+  "caps.pluginUpdate": "Update",
+  "caps.pluginDoctor": "Doctor",
+  "caps.pluginRemove": "Remove plugin",
+  "caps.pluginConfirmRemove": "Confirm remove",
+  "caps.pluginPlanError": "Install check failed",
+  "caps.pluginPlanReady": "Install plan",
+  "caps.pluginAction": "Action",
+  "caps.pluginPlanStatus": "Plugin operation: {status}",
+  "caps.pluginPlanComplete": "Plugin operation complete",
+  "caps.pluginPlanInstalled": "Plugin installed. Expand it below to see how to use the exported skills, hooks, and MCP servers.",
+  "caps.pluginUsageTitle": "How to use",
+  "caps.pluginUsageEnabledHint": "Enabled plugins load into new sessions. Use /skills to browse skills, invoke /<skill>, or ask naturally; hooks and MCP servers run through the normal session lifecycle.",
+  "caps.pluginUsageDisabledHint": "Enable this plugin before its skills, hooks, or MCP servers can participate in sessions.",
+  "caps.pluginNoCapabilityDetails": "No detailed capability inventory is available for this plugin.",
+  "caps.pluginSkillsTitle": "Skills",
+  "caps.pluginSkillsHint": "Browse with /skills, invoke directly, or ask for the workflow by name.",
+  "caps.pluginHooksTitle": "Hooks",
+  "caps.pluginHooksHint": "Hooks trigger automatically on matching lifecycle events.",
+  "caps.pluginMCPTitle": "MCP servers",
+  "caps.pluginMCPHint": "Tools from these servers are available through the normal MCP/tool flow.",
+  "caps.pluginNoDescription": "No description",
+  "caps.pluginHookNoTarget": "No hook target",
+  "caps.pluginMCPNoTarget": "No launch target",
   "caps.sources": "Sources",
   "caps.sourcesSummaryActive": "{active} active",
   "caps.sourcesSummaryMissing": "{missing} missing",
@@ -432,6 +489,9 @@ export const en = {
   "composer.pastedExpand": "Expand",
   "composer.pastedRemove": "Remove pasted text",
   "composer.removeImage": "Remove image",
+  "imageViewer.clickToPreview": "Click to preview",
+  "imageViewer.closePreview": "Close preview",
+  "imageViewer.title": "Image preview",
   "composer.attachImageFailed": "Image paste failed",
   "composer.attachFileFailed": "File attach failed",
   "composer.attachDropFailed": "Dropped file attach failed",
@@ -671,6 +731,18 @@ export const en = {
   "history.historySessionActions": "History session actions",
   "history.trashActions": "Trash actions",
 
+  // session recovery
+  "recovery.branch": "Recovery branch",
+  "recovery.badge": "Recovery",
+  "recovery.open": "Open recovery branch",
+  "recovery.toast": "Unsaved local changes were saved as {title}.",
+  "recovery.failedLease": "The recovery branch is in use by another runtime, so this tab did not switch to it.",
+  "recovery.failedUnavailable": "The recovery branch could not be opened, so this tab stayed on the original session.",
+  "recovery.banner": "This session was recovered from a save conflict. Compare it with the original before continuing.",
+  "recovery.reason": "Reason: {reason}",
+  "recovery.digest": "Digest: {digest}",
+  "recovery.parent": "Parent: {parent}",
+
   // project tree
   "projectTree.workspaceTitle": "Projects",
   "projectTree.pinnedTitle": "Pinned",
@@ -844,6 +916,7 @@ export const en = {
   "settings.tab.capabilities": "MCP & Skills",
   "settings.tab.mcp": "MCP & Tools",
   "settings.tab.skills": "Skills",
+  "settings.tab.plugins": "Plugins",
   "settings.tab.memory": "Memory",
   "settings.tab.hooks": "Hooks",
   "settings.tab.shortcuts": "Shortcuts",
@@ -859,6 +932,7 @@ export const en = {
   "settings.tabSub.capabilities": "Servers, tools & skills",
   "settings.tabSub.mcp": "Servers & tools",
   "settings.tabSub.skills": "Sources & enablement",
+  "settings.tabSub.plugins": "Packages & diagnostics",
   "settings.tabSub.memory": "Project & global",
   "settings.tabSub.hooks": "Shell automation",
   "settings.tabSub.shortcuts": "Keys & help",
@@ -940,6 +1014,10 @@ export const en = {
   "settings.subagentEffort": "Subagent effort",
   "settings.subagentEffortDefault": "auto (provider default)",
   "settings.subagentHint": "Applies to task and runAs=subagent skills unless a tool call or skill-specific config overrides it.",
+  "settings.subagentDepth": "Subagent nesting depth",
+  "settings.subagentDepthHint": "Controls whether a first-layer subagent may dispatch one more reviewer or worker subagent.",
+  "settings.subagentDepthOne": "1 layer",
+  "settings.subagentDepthTwo": "2 layers",
   "settings.defaultToolApprovalMode": "New session approval",
   "settings.defaultToolApprovalModeHint": "Saved to your global settings file and used only for new sessions; switching or restoring an existing session will not overwrite its Ask / Auto / Yolo state.",
   "settings.defaultToolApprovalMode.ask": "Ask",
@@ -962,7 +1040,7 @@ export const en = {
   "settings.coldResumePrune.on": "On",
   "settings.coldResumePrune.off": "Off",
   "settings.reasoningLanguage": "Thinking language",
-  "settings.reasoningLanguageHint": "Affects visible reasoning only. Auto follows the conversation language without injecting a language instruction.",
+  "settings.reasoningLanguageHint": "Affects visible reasoning only. Auto anchors clear Chinese prompts; referenced context is ignored.",
   "settings.reasoningLanguage.auto": "Auto",
   "settings.reasoningLanguage.zh": "中文",
   "settings.reasoningLanguage.en": "English",
@@ -1069,11 +1147,76 @@ export const en = {
   "settings.toggleOff": "Off",
   "settings.botSetup": "Bot entry",
   "settings.botSetupHint": "Enable the entry, then connect the chat channels you need.",
-  "settings.botConnectPhoneTitle": "Add IM bot",
-  "settings.botConnectPhoneSubtitle": "Manual setup for QQ, QR connection for Feishu, Lark, and WeChat.",
   "settings.botGateway": "Bot gateway",
-  "settings.botGatewayHint": "Manage the IM bot gateway, model, and runtime parameters from the desktop app.",
+  "settings.botGatewayHint": "Manage the IM bot gateway master switch and connection status from the desktop app.",
   "settings.botBrowserPreviewWarning": "This is a browser preview. It saves mock settings only and does not write local config or start the QQ/IM bot gateway. Configure it in the desktop app or CLI before testing from QQ.",
+  "settings.botSimpleEyebrow": "Guided setup",
+  "settings.botSimpleTitle": "Mobile access and chat bots",
+  "settings.botSimpleHint": "Scan from a phone to control the current workspace. For longer-lived access from chat apps, pick a Bot Channel and configure it.",
+  "settings.botMobileRemoteEyebrow": "Separate entry",
+  "settings.botMobileRemoteTitle": "Mobile remote control",
+  "settings.botMobileRemoteHint": "Scan or copy a link to open the current workspace control entry from a mobile device. This is configured separately from IM bot channels.",
+  "settings.botMobileRemoteControlApi": "Control link",
+  "settings.botMobileRemoteDisabled": "Enable it to generate the QR code and link",
+  "settings.botMobileRemoteEnableFirst": "Enable to generate QR",
+  "settings.botMobileRemoteReady": "Waiting for phone connection",
+  "settings.botMobileRemoteWaiting": "Connect by phone scan",
+  "settings.botMobileRemoteLoopbackNotice": "The control API is still bound to local loopback, so this is for local preview for now. Cross-device relay can reuse this entry later.",
+  "settings.botMobileRemoteCopyLink": "Copy link",
+  "settings.botMobileRemoteCopied": "Copied",
+  "settings.botMobileRemoteCopyFailed": "Copy failed",
+  "settings.botChannelEntryTitle": "Use Bot Channel",
+  "settings.botChannelEntryHint": "Connect chat bots for longer mobile sessions.",
+  "settings.botManageBots": "Bot management",
+  "settings.botManageBotsHint": "Choose a channel on the left. If it is not configured, scan or add keys; once configured, manage that bot's runtime settings on the right.",
+  "settings.botBackToOverview": "Back to entry",
+  "settings.botChannelOpenConfigured": "Configured. You can continue editing settings.",
+  "settings.botChannelNeedsSetup": "Not configured",
+  "settings.botChannelTabsLabel": "Bot channels",
+  "settings.botChannelSetupTitle": "Configure {provider}",
+  "settings.botChannelSetupHint": "Before setup, only credentials or scan binding are needed. After it succeeds, this becomes the bot's settings page.",
+  "settings.botSetupStatus": "Bot setup status",
+  "settings.botSetupStatusConnected": "{online}/{total} online",
+  "settings.botSetupStatusAccessTrusted": "Trusted list {count}",
+  "settings.botSetupStatusAccessEveryone": "Everyone allowed",
+  "settings.botSetupStatusAccessPerBot": "{configured}/{total} bots configured",
+  "settings.botMode": "Bot settings mode",
+  "settings.botModeSimple": "Simple",
+  "settings.botModeAdvanced": "Advanced",
+  "settings.botShowAdvancedSettings": "Advanced settings",
+  "settings.botHideAdvancedSettings": "Hide advanced settings",
+  "settings.botAdvancedSettingsHint": "Lower-frequency controls: routes, queue details, and bot echo filtering.",
+  "settings.botShowAddBot": "Add channel",
+  "settings.botStepConnectTitle": "Connect bots",
+  "settings.botStepConnectHint": "Manual keys for QQ; scan a QR code for Feishu, Lark, and WeChat.",
+  "settings.botStepConnectEmpty": "Not connected yet",
+  "settings.botSharedAllowlistHint": "Every channel shares the trusted list and pairing settings from step 2. Edit once and it applies everywhere.",
+  "settings.botSharedAllowlistGo": "Edit in step 2",
+  "settings.botGlobalAllowlist": "Legacy global allowlist",
+  "settings.botGlobalAllowlistHint": "Compatibility for older configs; new bots should set access in their own detail panel.",
+  "settings.botAccessShowAllPlatforms": "Show all platforms",
+  "settings.botAccessShowConnectedOnly": "Show connected platforms only",
+  "settings.botListUsers": "Users",
+  "settings.botListGroups": "Groups",
+  "settings.botListApprovers": "Approvers",
+  "settings.botListAdmins": "Admins",
+  "settings.botPlatformFeishuLark": "Feishu / Lark",
+  "settings.botQueueDropLabel": "When the queue is full",
+  "settings.botQueueDropHint": "How overflow messages are handled once the queue hits its cap.",
+  "settings.botPairingDetailHint": "The pairing switch lives in step 2 (Who can use it). Tune only the code lifetime and pending cap here.",
+  "settings.botSimpleAccessTitle": "Who can use it",
+  "settings.botSimpleAccessTrustedSummary": "Trusted list is active with {count} member or group entries.",
+  "settings.botSimpleAccessEveryoneSummary": "Everyone can use it right now. Best for short tests only.",
+  "settings.botAccessTrusted": "Trusted list",
+  "settings.botAccessTrustedHint": "Only listed users or groups can call the bot.",
+  "settings.botAccessEveryone": "Everyone",
+  "settings.botAccessEveryoneHint": "Convenient for quick testing, but not recommended long term.",
+  "settings.botAccessPairing": "Pair private chats",
+  "settings.botAccessPairingHint": "Unknown direct messages get a code and wait for local approval.",
+  "settings.botRoleAccess": "Admins and approvers",
+  "settings.botRoleAccessHint": "Optional. Leave this empty unless remote approval or management is needed.",
+  "settings.botQueueModeSimple": "When a new message arrives",
+  "settings.botQueueModeSimpleHint": "Choose what happens when the bot is already handling a previous message.",
   "settings.saveBotSettings": "Save bot settings",
   "settings.botEnableBot": "Enable bot",
   "settings.botGatewayMasterSwitch": "Master switch",
@@ -1084,8 +1227,48 @@ export const en = {
   "settings.botModelHint": "Leave empty to use the default model.",
   "settings.botModelDefault": "Use default model",
   "settings.botRuntime": "Runtime limits",
+  "settings.botRuntimeHint": "Gateway-level limits that control how far each bot task can run.",
   "settings.botMaxSteps": "Max steps",
   "settings.botDebounceMs": "Debounce ms",
+  "settings.botQueueCap": "Queue cap",
+  "settings.botQueueMode.steer": "Steer running turn",
+  "settings.botQueueMode.followup": "Follow up",
+  "settings.botQueueMode.collect": "Collect",
+  "settings.botQueueMode.interrupt": "Interrupt",
+  "settings.botQueueDrop.summarize": "Summarize overflow",
+  "settings.botQueueDrop.old": "Drop oldest",
+  "settings.botQueueDrop.new": "Drop newest",
+  "settings.botIgnoreSelfMessages": "Ignore bot echoes",
+  "settings.botIgnoreSelfMessagesHint": "Prevents the gateway from reacting to messages sent by configured bot identities.",
+  "settings.botSelfUserIds": "Bot self IDs",
+  "settings.botSelfUserIdsHint": "Optional user/open IDs for each platform, used to filter bot echo messages.",
+  "settings.botPairing": "Private chat pairing",
+  "settings.botPairingTTL": "TTL minutes",
+  "settings.botPairingMaxPending": "Max pending",
+  "settings.botControlApi": "Local control API",
+  "settings.botControlApiHint": "Loopback HTTP API for status and controlled outbound sends; requires a Bearer token from the named env var.",
+  "settings.botControlAddr": "Control address",
+  "settings.botControlTokenEnv": "Token env",
+  "settings.botGatewayDefaults": "Gateway runtime settings",
+  "settings.botGatewayDefaultsHint": "Tune queueing, pairing, and self-message filtering without editing TOML.",
+  "settings.botRoutes": "Bind chats to projects",
+  "settings.botRoutesHint": "Pin a group, contact, or thread to a workspace, model, or approval mode.",
+  "settings.botRoutesEmpty": "No bindings yet. Add one to bind a chat, user, or thread to a project or model.",
+  "settings.botRouteTitle": "Route {n}",
+  "settings.botAddRoute": "Add route",
+  "settings.botRouteConnection": "Bot source",
+  "settings.botRouteAny": "Any",
+  "settings.botRoutePlatform": "Platform",
+  "settings.botRouteChatType": "Chat type",
+  "settings.botRouteChatType.any": "Any",
+  "settings.botRouteChatType.dm": "DM",
+  "settings.botRouteChatType.group": "Group",
+  "settings.botRouteChatType.guild": "Guild",
+  "settings.botRouteChatType.direct": "Direct",
+  "settings.botRouteChatType.thread": "Thread",
+  "settings.botRouteChatId": "Group/chat ID",
+  "settings.botRouteUserId": "Contact ID",
+  "settings.botRouteThreadId": "Thread ID",
   "settings.botChannels": "Connect channels",
   "settings.botChannelsHint": "Pick a channel; QQ uses manual setup, while Feishu, Lark, and WeChat connect by QR.",
   "settings.botQQHint": "QQ official Bot API v2.",
@@ -1140,8 +1323,7 @@ export const en = {
   "settings.botInstallLarkHint": "Scan to create a Lark PersonalAgent.",
   "settings.botInstallWeixinHint": "Scan WeChat and save a local token.",
   "settings.botConnectedBots": "Connected bots",
-  "settings.botConnectedBotsSummary": "{online}/{total} online",
-  "settings.botConnectionsEmpty": "No bot channel is connected yet.",
+  "settings.botSelectBotHint": "Select a bot on the left to configure it.",
   "settings.botConnectionColumnChannel": "Channel",
   "settings.botConnectionColumnName": "Bot",
   "settings.botConnectionColumnRemote": "Remote ID",
@@ -1159,16 +1341,17 @@ export const en = {
   "settings.botConnectionConnected": "Connected",
   "settings.botConnectionConfigured": "Configured",
   "settings.botConnectionDisconnected": "Disconnected",
-  "settings.botManage": "Advanced settings",
+  "settings.botManage": "Edit config",
   "settings.deleteBot": "Delete bot",
+  "settings.deleteBotShort": "Delete",
   "settings.confirmDeleteBot": "Confirm delete",
   "settings.deleteBotHint": "Removes this bot from connected bots; saved secrets are not cleared automatically.",
   "settings.botChannelModel": "Model",
-  "settings.botChannelModelHint": "Leave empty to use the bot default model.",
-  "settings.botChannelModelAuto": "Auto",
+  "settings.botChannelModelHint": "The model this bot uses when handling messages.",
+  "settings.botChannelModelAuto": "Use current desktop model",
   "settings.botToolApprovalMode": "Tool approval",
   "settings.botToolApprovalModeHint": "Controls whether this channel asks before running tools; plan confirmation and ask questions still wait for a reply.",
-  "settings.botToolApprovalMode.inherit": "Inherit",
+  "settings.botToolApprovalMode.inherit": "Follow bot",
   "settings.botToolApprovalMode.ask": "Ask",
   "settings.botToolApprovalMode.auto": "Auto",
   "settings.botToolApprovalMode.yolo": "YOLO",
@@ -1191,26 +1374,17 @@ export const en = {
   "settings.botDiagnosticReportFailed": "Failed to send diagnostic report.",
   "settings.botDiagnosticPrivacy": "Only sanitized connection metadata is included.",
   "settings.botAccessControl": "Access control",
-  "settings.botAccessControlHint": "Whitelist is the default; switch to everyone for quick testing.",
-  "settings.botAccessMode": "Who can use it",
-  "settings.botAccessAll": "Everyone",
-  "settings.botAccessWhitelist": "Whitelist only",
   "settings.botEditAllowlist": "Edit allowlist",
   "settings.botAllowlist": "Allowlist",
-  "settings.botAllowlistHint": "Restrict which users or groups can invoke the bot.",
   "settings.botAllowlistEnabled": "Enable allowlist",
   "settings.botAllowAll": "Allow everyone",
   "settings.botAllowAllWarn": "When enabled, every incoming channel message can invoke the bot.",
-  "settings.botQQAccessRequired": "Add a QQ user/group allowlist, or temporarily allow everyone in Advanced settings, before enabling QQ Bot.",
-  "settings.botAllowlistEntries": "Users and groups",
+  "settings.botQQAccessRequired": "Add QQ users/groups in step 2 (Who can use it), or temporarily allow everyone, before enabling QQ Bot.",
   "settings.botQQUsers": "QQ users",
-  "settings.botQQGroups": "QQ groups",
   "settings.botFeishuUsers": "Feishu users",
   "settings.botFeishuGroups": "Feishu groups",
   "settings.botFeishuLarkUsers": "Feishu / Lark users",
-  "settings.botFeishuLarkGroups": "Feishu / Lark groups",
   "settings.botWeixinUsers": "WeChat users",
-  "settings.botWeixinGroups": "WeChat groups",
   "settings.botListPlaceholder": "One ID per line, or comma-separated",
   "settings.permissions": "Permissions",
   "settings.writerMode": "Writer mode",
@@ -1222,8 +1396,11 @@ export const en = {
   "settings.yoloHint": "YOLO-approves approval-gated tool calls (writers, bash). Deny rules still block, and ask questions plus plan approvals still wait for you. Runtime-only — resets on relaunch.",
   "settings.addRule": "add {list} rule…",
   "settings.sandboxTitle": "Sandbox & workspace",
+  "settings.sandboxBoundaryHint": "Sandbox write roots limit where approved Bash and file tools can write; they do not auto-approve commands. Use Permissions or approval mode for deploy-style commands.",
+  "settings.reloadSessionConfig": "Apply manual config changes",
+  "settings.reloadSessionConfigHint": "Changes made in Settings apply automatically; use this only after manually editing config.toml or reasonix.toml.",
   "settings.shellInterpreter": "Shell interpreter",
-  "settings.shellAuto": "auto (bash, else PowerShell on Windows)",
+  "settings.shellAuto": "auto (bash first)",
   "settings.shellBash": "bash",
   "settings.shellPowershell": "PowerShell (Windows PowerShell)",
   "settings.shellPwsh": "pwsh (PowerShell 7+)",
@@ -1233,6 +1410,9 @@ export const en = {
   "settings.allowNetwork": "Allow network egress from sandboxed bash",
   "settings.workspaceRoot": "Workspace root",
   "settings.workspaceDefault": "(default: cwd)",
+  "settings.effectiveWriteRoots": "Effective write roots",
+  "settings.effectiveWriteRootsHint": "Expanded for the active workspace. Settings changes apply automatically; use the button above after manually editing config files.",
+  "settings.noEffectiveWriteRoots": "No active write roots",
   "settings.appearance": "Appearance",
   "settings.theme": "Theme",
   "settings.themeStyle": "Visual style",
@@ -1340,6 +1520,7 @@ export const en = {
   "settings.pageDesc.bots": "Configure QQ, Feishu, Lark, and WeChat bot channels, including each bot's model and runtime parameters.",
   "settings.pageDesc.mcp": "Manage MCP server connections and tool discovery.",
   "settings.pageDesc.skills": "Browse, enable, and manage agent skills and skill sources.",
+  "settings.pageDesc.plugins": "Install and manage plugin packages that can provide skills, hooks, and MCP servers.",
   "settings.pageDesc.memory": "Manage preferences, project conventions, and instruction files Reasonix keeps referencing.",
   "settings.pageDesc.hooks": "Configure shell hooks around turns, tool calls, compaction, and session lifecycle.",
   "settings.pageDesc.shortcuts": "Customize keyboard shortcuts and keep the help sheet in sync.",
@@ -1354,7 +1535,7 @@ export const en = {
   "settings.customProviders": "Custom providers",
   "settings.customProvidersHint": "OpenAI-compatible or proxy providers you add yourself. These can be edited, deleted, and updated with fetched model lists.",
   "settings.customProvidersEmpty": "No custom providers yet. Prefer the built-in access above for common services.",
-  "settings.providerAccessEmptyHint": "Click “Add provider” to choose DeepSeek or a custom OpenAI-compatible access.",
+  "settings.providerAccessEmptyHint": "Use Recommended preset to quickly add common providers, usually with only an API key. Use Custom provider for OpenAI-compatible or Anthropic-compatible endpoints.",
   "settings.builtinProviderBadge": "built-in",
   "settings.customProviderBadge": "custom",
   "settings.configureProvider": "Configure",
@@ -1366,24 +1547,75 @@ export const en = {
   "settings.searchModels": "Search models...",
   "settings.noMatchingModels": "No matching models",
   "settings.addProvider.chooseTitle": "Add provider",
-  "settings.addProvider.chooseHint": "Choose an official preset, or add a custom OpenAI-compatible access.",
-  "settings.addProvider.officialChoice": "Official provider",
-  "settings.addProvider.officialChoiceHint": "Choose from Reasonix presets for DeepSeek.",
+  "settings.addProvider.chooseHint": "Choose a recommended preset, or add a custom provider manually.",
+  "settings.addProvider.officialChoice": "Recommended preset",
+  "settings.addProvider.officialChoiceHint": "Choose DeepSeek or curated custom-provider presets.",
   "settings.addProvider.customChoice": "Custom provider",
   "settings.addProvider.customChoiceHint": "Configure base_url, models, key env, and related parameters manually.",
   "settings.addProvider.officialTitle": "Add official provider",
-  "settings.addProvider.officialHint": "Official presets use fixed endpoints and defaults; choose the provider and set a key.",
+  "settings.addProvider.officialHint": "Recommended presets fill endpoints, models, vision, and reasoning defaults; usually only the key is needed.",
   "settings.addProvider.customTitle": "Add custom provider",
   "settings.addProvider.customHint": "Use this for proxies, aggregators, or other OpenAI-compatible services.",
   "settings.addProvider.official.deepseek": "DeepSeek Official",
   "settings.addProvider.official.deepseekDesc": "Official API key access with DeepSeek V4 Flash / Pro.",
+  "settings.addProvider.preset.kimiCnDesc": "Kimi China API via Moonshot CN endpoint with K2 vision models.",
+  "settings.addProvider.preset.kimiGlobalDesc": "Kimi international API via Moonshot global endpoint with K2 vision models.",
+  "settings.addProvider.preset.kimiCodingPlanDesc": "Kimi Coding Plan via the dedicated Anthropic-compatible endpoint; not a CN/Global variant.",
+  "settings.addProvider.preset.mimoApiDesc": "Xiaomi MiMo direct API for v2.5 pay-as-you-go models.",
+  "settings.addProvider.preset.mimoAnthropicDesc": "Xiaomi MiMo direct Anthropic-compatible endpoint with MiMo v2.5 models.",
+  "settings.addProvider.preset.mimoTokenPlanCnDesc": "Xiaomi MiMo token-plan China endpoint.",
+  "settings.addProvider.preset.mimoTokenPlanCnAnthropicDesc": "Xiaomi MiMo token-plan China Anthropic-compatible endpoint.",
+  "settings.addProvider.preset.mimoTokenPlanSgpDesc": "Xiaomi MiMo token-plan Singapore endpoint.",
+  "settings.addProvider.preset.mimoTokenPlanSgpAnthropicDesc": "Xiaomi MiMo token-plan Singapore Anthropic-compatible endpoint.",
+  "settings.addProvider.preset.mimoTokenPlanAmsDesc": "Xiaomi MiMo token-plan Amsterdam endpoint.",
+  "settings.addProvider.preset.mimoTokenPlanAmsAnthropicDesc": "Xiaomi MiMo token-plan Amsterdam Anthropic-compatible endpoint.",
+  "settings.addProvider.preset.minimaxCnApiDesc": "MiniMax China OpenAI-compatible M-series API endpoint.",
+  "settings.addProvider.preset.minimaxGlobalApiDesc": "MiniMax international OpenAI-compatible M-series API endpoint.",
+  "settings.addProvider.preset.minimaxCnAnthropicDesc": "MiniMax China Anthropic-compatible M-series endpoint.",
+  "settings.addProvider.preset.minimaxGlobalAnthropicDesc": "MiniMax international Anthropic-compatible endpoint using Bearer auth.",
+  "settings.addProvider.preset.glmCnDesc": "Zhipu GLM China API with GLM thinking controls.",
+  "settings.addProvider.preset.zaiGlobalDesc": "Z.AI international GLM API.",
+  "settings.addProvider.preset.glmCodingPlanCnDesc": "Zhipu GLM China coding-plan endpoint with GLM-5.2 default.",
+  "settings.addProvider.preset.glmCodingPlanCnAnthropicDesc": "Zhipu GLM China Anthropic-compatible coding-plan endpoint with GLM-5.2[1m] default.",
+  "settings.addProvider.preset.zaiCodingPlanGlobalDesc": "Z.AI international coding-plan endpoint with GLM-5.2 default.",
+  "settings.addProvider.preset.zaiCodingPlanGlobalAnthropicDesc": "Z.AI international Anthropic-compatible coding-plan endpoint with GLM-5.2[1m] default.",
+  "settings.addProvider.preset.opencodeGoDesc": "OpenCode Go relay with mixed-model capability overrides.",
+  "settings.addProvider.preset.opencodeGoAnthropicDesc": "OpenCode Go Anthropic route for Qwen and MiniMax models.",
+  "settings.addProvider.preset.opencodeZenAnthropicDesc": "OpenCode Zen Anthropic route for Claude and Qwen models.",
+  "settings.addProvider.preset.qwenCnDesc": "Qwen/DashScope China standard API endpoint.",
+  "settings.addProvider.preset.qwenGlobalDesc": "Qwen/DashScope international standard API endpoint.",
+  "settings.addProvider.preset.qwenCodingPlanCnDesc": "Qwen Coding Plan China OpenAI-compatible endpoint with qwen3.7-plus default.",
+  "settings.addProvider.preset.qwenCodingPlanCnAnthropicDesc": "Qwen Coding Plan China Anthropic-compatible endpoint with qwen3.7-plus default.",
+  "settings.addProvider.preset.qwenCodingPlanGlobalDesc": "Qwen Coding Plan international OpenAI-compatible endpoint with qwen3.7-plus default.",
+  "settings.addProvider.preset.qwenCodingPlanGlobalAnthropicDesc": "Qwen Coding Plan international Anthropic-compatible endpoint with qwen3.7-plus default.",
+  "settings.addProvider.preset.stepfunDesc": "StepFun coding-plan OpenAI-compatible endpoint with step-3.7-flash default.",
+  "settings.addProvider.preset.stepfunAnthropicDesc": "StepFun coding-plan Anthropic-compatible endpoint with step-3.7-flash default.",
+  "settings.addProvider.preset.novitaDesc": "NovitaAI multi-model gateway with Kimi, MiniMax, GLM, DeepSeek, and Qwen.",
+  "settings.addProvider.preset.gmiDesc": "GMI Cloud direct multi-model gateway with GLM, DeepSeek, Kimi, Gemini, Claude, and GPT.",
+  "settings.addProvider.preset.vercelAiGatewayDesc": "Vercel AI Gateway through Anthropic-compatible Messages API.",
+  "settings.addProvider.preset.huggingfaceDesc": "HuggingFace Inference Router with Qwen and DeepSeek defaults.",
+  "settings.addProvider.preset.nvidiaDesc": "NVIDIA NIM accelerated inference endpoint with Nemotron defaults.",
+  "settings.addProvider.preset.kilocodeDesc": "Kilo Code gateway with kilo/auto routing defaults.",
+  "settings.addProvider.preset.ollamaCloudDesc": "Hosted Ollama Cloud endpoint with none-to-max reasoning effort.",
+  "settings.addProvider.addedBadge": "Added",
+  "settings.addProvider.modifiedBadge": "Modified",
+  "settings.addProvider.nameConflictBadge": "Name taken",
+  "settings.addProvider.similarExistingBadge": "Similar setup",
   "settings.addProvider.confirm": "Add provider",
+  "settings.addProvider.alreadyAddedAction": "Already added",
+  "settings.addProvider.nameConflictAction": "Resolve name first",
+  "settings.addProvider.viewPresetProvider": "View config",
+  "settings.addProvider.viewConflictProvider": "View same-name config",
+  "settings.addProvider.resetPreset": "Reset to preset",
+  "settings.addProvider.confirmResetPreset": "Confirm reset",
   "settings.customProviderName": "Custom provider name",
   "settings.customProviderNamePlaceholder": "",
   "settings.providerNamePlaceholder": "Provider name (e.g. deepseek-proxy)",
   "settings.providerProtocol": "Connection protocol",
   "settings.providerProtocolOpenAI": "OpenAI-compatible",
   "settings.providerProtocolOpenAIHint": "For proxies, aggregators, and self-hosted services compatible with the OpenAI API.",
+  "settings.providerProtocolAnthropic": "Anthropic-compatible",
+  "settings.providerProtocolAnthropicHint": "For Claude-style Messages API services such as some coding-plan endpoints.",
   "settings.providerBaseUrlLabel": "API address",
   "settings.providerChatUrlLabel": "Chat request URL",
   "settings.providerChatUrlPlaceholder": "e.g. https://api.openai.com/v1/chat/completions",
@@ -1406,6 +1638,12 @@ export const en = {
   "settings.providerHeaders": "Extra request headers (optional)",
   "settings.providerHeadersPlaceholder": "HTTP-Referer: https://example.com\nX-Title: Reasonix",
   "settings.providerHeadersHint": "For gateways such as OpenRouter that require site headers. Use one Header: value per line; the API key still belongs in the key field above.",
+  "settings.providerExtraBody": "Extra request body (optional)",
+  "settings.providerExtraBodyPlaceholder": "{\n  \"enable_thinking\": true\n}",
+  "settings.providerExtraBodyHint": "Top-level JSON fields merged into the chat request body. Core fields such as model, messages, tools, stream, and thinking stay controlled by Reasonix.",
+  "settings.providerExtraBodyError": "Enter a JSON object without null values, for example { \"enable_thinking\": true }.",
+  "settings.providerAuthHeader": "Use Authorization: Bearer for this key",
+  "settings.providerAuthHeaderHint": "For Anthropic-compatible gateways such as MiniMax Global or Vercel AI Gateway that do not use x-api-key.",
   "settings.fetchingModels": "Refreshing…",
   "settings.fetchModelsSuccess": "Fetched {n} models",
   "settings.fetchModelsManualFallbackAuth": "The model discovery endpoint rejected the key or permission. Check the API key; if this service does not expose model discovery, enter model IDs manually below.",
@@ -1425,10 +1663,16 @@ export const en = {
   "settings.newProviderDraftName": "new provider",
 	  "settings.reasoningProtocol": "Model capability mode",
 	  "settings.reasoningProtocol.auto": "Auto-detect (recommended)",
-	  "settings.reasoningProtocol.deepseek": "DeepSeek thinking",
-	  "settings.reasoningProtocol.openai": "OpenAI reasoning",
+	  "settings.reasoningProtocol.deepseek": "DeepSeek thinking (thinking + depth)",
+	  "settings.reasoningProtocol.openai": "OpenAI reasoning (reasoning_effort)",
 	  "settings.reasoningProtocol.none": "Plain chat (no reasoning fields)",
 	  "settings.reasoningProtocolHint": "Auto-detect is best for most gateways. Switch only when a proxy is incompatible or the model documentation requires a specific reasoning shape.",
+	  "settings.thinkingMode": "Thinking override",
+	  "settings.thinkingMode.auto": "Auto (provider default)",
+	  "settings.thinkingMode.enabled": "Enabled",
+	  "settings.thinkingMode.disabled": "Disabled",
+	  "settings.thinkingMode.adaptive": "Adaptive (self-adjusting)",
+	  "settings.thinkingModeHint": "Provider-specific override. Use enabled/disabled for compatible gateways, adaptive only for providers that document it.",
   "settings.updateKey": "update {env} (saved globally)",
   "settings.updateKeyAction": "Update key",
   "settings.keyStatus": "Key",
@@ -1543,6 +1787,8 @@ export const en = {
   "msg.imageAttachment": "Image",
   "msg.workspaceReference": "Workspace reference",
   "msg.folderReference": "Folder · Workspace reference",
+  "msg.pastedExpandTooltip": "Expand pasted text",
+  "msg.pastedCollapseTooltip": "Collapse pasted text",
   "turnActions.summary": "Summary",
   "turnActions.rewind": "Rewind",
   "transcript.jumpToBottom": "Jump to bottom",
@@ -1624,7 +1870,7 @@ export const en = {
   "settings.telemetryLabel": "Anonymous usage ping",
   "settings.telemetryHint": "On launch, send a random install id plus version and OS to count active installs, and — if the app crashed on a previous run — resend that crash report. Never includes conversations, keys, or file contents.",
   "settings.metricsLabel": "Share aggregate quality metrics",
-  "settings.metricsHint": "On by default. Sends anonymous turn-end counts, settings preference snapshots, and Memory v5 aggregate buckets such as injection on/off, size buckets, IR count buckets, and memory graph size buckets. Includes the random install id used to de-duplicate DAU. Buckets may include normalized custom provider and model names — never conversations, prompts, keys, paths, base URLs, memory text, tool outputs, or file contents.",
+  "settings.metricsHint": "On by default. Sends anonymous turn-end counts, updater error categories, settings preference snapshots, and Memory v5 aggregate buckets such as injection on/off, size buckets, IR count buckets, and memory graph size buckets. Includes the random install id used to de-duplicate DAU. Buckets may include normalized custom provider and model names — never conversations, prompts, keys, paths, base URLs, memory text, tool outputs, or file contents.",
   "updater.currentVersion": "Current version: {v}",
   "updater.channelLabel": "Update channel: {channel}",
   "updater.checkButton": "Check for updates",

desktop/frontend/src/locales/zh-TW.ts

@@ -134,6 +134,7 @@ export const zhTW: Record<DictKey, string> = {
   "caps.summary": "{connected} 個已連線 · {failed} 個異常 · {skills} 個技能",
   "caps.connectorsTab": "MCP 伺服器",
   "caps.skillsTab": "Agent 技能",
+  "caps.pluginsTab": "插件",
   "caps.servers": "MCP 伺服器",
   "caps.noServers": "未設定 MCP 伺服器。用 /mcp add 新增。",
   "caps.failureTitle": "{failed} 個 MCP 啟動異常",
@@ -227,6 +228,62 @@ export const zhTW: Record<DictKey, string> = {
   "caps.trustedReadOnlyTitle": "已記住為之後規劃和唯讀 research 可用的唯讀工具",
   "caps.searchSkills": "搜尋技能…",
   "caps.noSkillMatches": "沒有匹配的技能。",
+  "caps.pluginRefresh": "重新整理插件",
+  "caps.pluginInstallTitle": "安裝插件",
+  "caps.pluginInstallHint": "選擇本機插件目錄，或從 Git 倉庫安裝。",
+  "caps.pluginInstallMethod": "插件安裝方式",
+  "caps.pluginInstallLocal": "本機目錄",
+  "caps.pluginInstallGit": "Git 倉庫",
+  "caps.pluginSource": "插件來源",
+  "caps.pluginGitSource": "Git 倉庫地址",
+  "caps.pluginSourcePlaceholder": "git:github.com/obra/superpowers",
+  "caps.pluginChooseLocalFolder": "選擇插件目錄",
+  "caps.pluginLocalFolder": "本機插件目錄",
+  "caps.pluginNoLocalFolder": "尚未選擇目錄",
+  "caps.pluginInstallName": "安裝名稱（可選）",
+  "caps.pluginInstallNamePlaceholder": "安裝名稱（可選）",
+  "caps.pluginLink": "開發模式：連結來源目錄",
+  "caps.pluginLinkHint": "不複製插件，直接使用所選目錄；適合調試插件。移動或刪除該目錄會導致插件失效。",
+  "caps.pluginReplace": "覆蓋同名插件",
+  "caps.pluginReplaceHint": "插件名已存在時，允許用目前來源替換它。",
+  "caps.pluginPreview": "預檢",
+  "caps.pluginInstall": "安裝插件",
+  "caps.noPluginsTitle": "尚未安裝插件",
+  "caps.noPluginsHint": "從 Git 或本機路徑安裝插件包，把技能、hooks 和 MCP 伺服器作為一個統一單元管理。",
+  "caps.installedPlugins": "已安裝插件",
+  "caps.pluginsSummary": "{enabled} 個啟用 · 共 {total} 個 · {issues} 個異常",
+  "caps.pluginCounts": "{skills} 技能 · {hooks} hooks · {mcps} MCP",
+  "caps.pluginNoCapabilities": "未匯出能力",
+  "caps.pluginWarnings": "{count} 個警告",
+  "caps.pluginEnable": "啟用插件",
+  "caps.pluginDisable": "停用插件",
+  "caps.pluginEnabled": "已啟用",
+  "caps.pluginDisabled": "已停用",
+  "caps.pluginVersion": "版本",
+  "caps.pluginRoot": "安裝路徑",
+  "caps.pluginUpdate": "更新",
+  "caps.pluginDoctor": "診斷",
+  "caps.pluginRemove": "移除插件",
+  "caps.pluginConfirmRemove": "確認移除",
+  "caps.pluginPlanError": "安裝檢查失敗",
+  "caps.pluginPlanReady": "安裝計畫",
+  "caps.pluginAction": "操作",
+  "caps.pluginPlanStatus": "插件操作：{status}",
+  "caps.pluginPlanComplete": "插件操作完成",
+  "caps.pluginPlanInstalled": "插件已安裝。展開下方插件，可查看它匯出的 skills、hooks 和 MCP 伺服器的使用方式。",
+  "caps.pluginUsageTitle": "使用方法",
+  "caps.pluginUsageEnabledHint": "啟用的插件會載入到新會話中。可用 /skills 查看技能、用 /<skill> 直接呼叫，或自然描述任務；hooks 和 MCP 伺服器會按正常會話流程自動參與。",
+  "caps.pluginUsageDisabledHint": "啟用此插件後，它的 skills、hooks 或 MCP 伺服器才會參與會話。",
+  "caps.pluginNoCapabilityDetails": "此插件暫無可展示的能力明細。",
+  "caps.pluginSkillsTitle": "Skills",
+  "caps.pluginSkillsHint": "用 /skills 瀏覽，直接呼叫，或在對話中說出工作流名稱。",
+  "caps.pluginHooksTitle": "Hooks",
+  "caps.pluginHooksHint": "Hooks 會在匹配的生命週期事件中自動觸發。",
+  "caps.pluginMCPTitle": "MCP 伺服器",
+  "caps.pluginMCPHint": "這些伺服器提供的工具會透過正常 MCP/工具流程可用。",
+  "caps.pluginNoDescription": "無描述",
+  "caps.pluginHookNoTarget": "未指定 hook 目標",
+  "caps.pluginMCPNoTarget": "未指定啟動目標",
   "caps.sources": "來源",
   "caps.sourcesSummaryActive": "{active} 個有效",
   "caps.sourcesSummaryMissing": "{missing} 個缺失",
@@ -309,6 +366,9 @@ export const zhTW: Record<DictKey, string> = {
   "composer.pastedExpand": "展開",
   "composer.pastedRemove": "刪除貼上文字",
   "composer.removeImage": "移除圖片",
+  "imageViewer.clickToPreview": "點擊查看大圖",
+  "imageViewer.closePreview": "關閉預覽",
+  "imageViewer.title": "圖片預覽",
   "composer.imageInputUnsupported": "目前模型不會直接接收圖片。可切換到支援圖片的模型，或讓 OCR/識圖 MCP 工具讀取附件路徑。",
   "composer.contextItems": "上下文項目",
   "composer.workspaceReferences": "工作區引用",
@@ -514,6 +574,18 @@ export const zhTW: Record<DictKey, string> = {
   "history.historySessionActions": "歷史會話操作",
   "history.trashActions": "回收站操作",
 
+  // 會話恢復
+  "recovery.branch": "恢復分支",
+  "recovery.badge": "恢復",
+  "recovery.open": "開啟恢復分支",
+  "recovery.toast": "未儲存的本機內容已儲存為 {title}。",
+  "recovery.failedLease": "恢復分支被另一個執行階段占用，未切換過去。",
+  "recovery.failedUnavailable": "無法開啟恢復分支，目前標籤仍停留在原會話。",
+  "recovery.banner": "目前會話來自儲存衝突恢復。繼續前建議與原會話對比。",
+  "recovery.reason": "原因：{reason}",
+  "recovery.digest": "摘要：{digest}",
+  "recovery.parent": "父會話：{parent}",
+
   // 專案樹
   "projectTree.workspaceTitle": "專案",
   "projectTree.pinnedTitle": "置頂",
@@ -629,6 +701,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.tab.capabilities": "MCP 與技能",
   "settings.tab.mcp": "MCP 與工具",
   "settings.tab.skills": "技能",
+  "settings.tab.plugins": "插件",
   "settings.tab.memory": "記憶",
   "settings.tab.shortcuts": "快捷鍵",
 	  "settings.tab.network": "網路",
@@ -642,6 +715,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.tabSub.capabilities": "伺服器、工具與技能",
   "settings.tabSub.mcp": "伺服器與工具",
   "settings.tabSub.skills": "來源與啟用",
+  "settings.tabSub.plugins": "包與診斷",
   "settings.tabSub.memory": "專案與全域",
   "settings.tabSub.shortcuts": "按鍵與幫助",
   "settings.tabSub.network": "代理與連線",
@@ -698,6 +772,10 @@ export const zhTW: Record<DictKey, string> = {
   "settings.subagentEffort": "子代理 effort",
   "settings.subagentEffortDefault": "auto（模型服務預設）",
   "settings.subagentHint": "作用於 task 和 runAs=subagent skills；工具呼叫參數或單個 skill 設定會覆蓋它。",
+  "settings.subagentDepth": "子代理巢狀深度",
+  "settings.subagentDepthHint": "控制第一層子代理能否再派發一層 reviewer 或 worker 子代理。",
+  "settings.subagentDepthOne": "1 層",
+  "settings.subagentDepthTwo": "2 層",
   "settings.defaultToolApprovalMode": "新會話預設審批",
   "settings.defaultToolApprovalModeHint": "保存在全域設定中，僅用於新建會話；切換或恢復已有會話時不會覆蓋其 Ask / Auto / Yolo 狀態。",
   "settings.defaultToolApprovalMode.ask": "詢問",
@@ -720,7 +798,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.coldResumePrune.on": "開",
   "settings.coldResumePrune.off": "關",
   "settings.reasoningLanguage": "思考語言",
-  "settings.reasoningLanguageHint": "只影響可見思考過程。自動跟隨對話語言，不額外注入語言指令。",
+  "settings.reasoningLanguageHint": "只影響可見思考過程。自動在原始提問明顯為中文時錨定中文，引用內容不參與判斷。",
   "settings.reasoningLanguage.auto": "自動",
   "settings.reasoningLanguage.zh": "中文",
   "settings.reasoningLanguage.en": "English",
@@ -807,12 +885,18 @@ export const zhTW: Record<DictKey, string> = {
   "settings.yoloHint": "所有工具呼叫（寫檔案、bash）自動執行、不再詢問。deny 規則仍會攔截。僅執行時生效，重啟後失效。",
   "settings.addRule": "新增 {list} 規則…",
   "settings.sandboxTitle": "沙箱與工作區",
+  "settings.sandboxBoundaryHint": "沙箱可寫根只限制已核准的 Bash 和檔案工具能寫到哪裡；它不會自動核准命令。mvn deploy 這類命令還需要權限規則或核准模式放行。",
+  "settings.reloadSessionConfig": "套用手動設定修改",
+  "settings.reloadSessionConfigHint": "設定頁內修改會自動生效；此按鈕用於套用手動編輯 config.toml / reasonix.toml 後的變更。",
   "settings.bashSandbox": "Bash 沙箱",
   "settings.bashEnforce": "enforce（隔離 bash）",
   "settings.bashOff": "off（不受限執行）",
   "settings.allowNetwork": "允許沙箱內 bash 存取網路",
   "settings.workspaceRoot": "工作區根目錄",
   "settings.workspaceDefault": "（預設：當前目錄）",
+  "settings.effectiveWriteRoots": "實際可寫根",
+  "settings.effectiveWriteRootsHint": "按目前工作區展開。設定頁內修改會自動生效；手動編輯設定檔後可用上方按鈕刷新目前會話。",
+  "settings.noEffectiveWriteRoots": "暫無可寫根",
   "settings.appearance": "外觀",
   "settings.theme": "主題",
   "settings.themeStyle": "強調色",
@@ -847,6 +931,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.pageDesc.providers": "管理模型供應商、API 位址、金鑰與可用模型。",
   "settings.pageDesc.mcp": "管理 MCP 伺服器連線和工具發現。",
   "settings.pageDesc.skills": "瀏覽、啟用和管理智慧體技能與來源。",
+  "settings.pageDesc.plugins": "安裝和管理插件包，用統一單元提供技能、hooks 與 MCP 伺服器。",
   "settings.pageDesc.memory": "管理 Reasonix 會長期參考的偏好、專案約定和指令檔案。",
   "settings.pageDesc.network": "代理與網路設定。",
   "settings.pageDesc.permissions": "寫操作模式與細粒度工具權限規則。",
@@ -859,7 +944,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.customProviders": "自訂供應商",
   "settings.customProvidersHint": "你自己新增的 OpenAI-compatible 或代理接入，可以編輯、刪除並持久化模型列表。",
   "settings.customProvidersEmpty": "還沒有自訂供應商。常用服務優先使用上方內建接入。",
-  "settings.providerAccessEmptyHint": "點擊「新增供應商」，選擇 DeepSeek 或自訂 OpenAI-compatible 接入。",
+  "settings.providerAccessEmptyHint": "使用「推薦預設」可快速接入常用供應商，通常只需填寫 API Key；使用「自訂供應商」可手動設定 OpenAI-compatible 或 Anthropic-compatible 端點。",
   "settings.builtinProviderBadge": "內建",
   "settings.customProviderBadge": "自訂",
   "settings.configureProvider": "設定",
@@ -871,24 +956,75 @@ export const zhTW: Record<DictKey, string> = {
   "settings.searchModels": "搜尋模型...",
   "settings.noMatchingModels": "沒有匹配的模型",
   "settings.addProvider.chooseTitle": "新增供應商",
-  "settings.addProvider.chooseHint": "選擇官方預設，或新增一個自訂 OpenAI-compatible 接入。",
-  "settings.addProvider.officialChoice": "官方供應商",
-  "settings.addProvider.officialChoiceHint": "從 Reasonix 預設的 DeepSeek 接入中選擇。",
+  "settings.addProvider.chooseHint": "選擇推薦預設，或手動新增一個自訂供應商。",
+  "settings.addProvider.officialChoice": "推薦預設",
+  "settings.addProvider.officialChoiceHint": "選擇 DeepSeek 或已適配的自訂供應商預設。",
   "settings.addProvider.customChoice": "自訂供應商",
   "settings.addProvider.customChoiceHint": "按 base_url、模型、金鑰環境變數等參數手動設定。",
   "settings.addProvider.officialTitle": "新增官方供應商",
-  "settings.addProvider.officialHint": "官方預設會寫入固定端點和預設參數，只需要選擇供應商並設定金鑰。",
+  "settings.addProvider.officialHint": "推薦預設會填好端點、模型、視覺和推理預設值，通常只需要填 API Key。",
   "settings.addProvider.customTitle": "新增自訂供應商",
   "settings.addProvider.customHint": "適合代理、聚合平台或其它 OpenAI-compatible 服務。",
   "settings.addProvider.official.deepseek": "DeepSeek 官方",
   "settings.addProvider.official.deepseekDesc": "官方 API Key 接入，可用 DeepSeek V4 Flash / Pro。",
+  "settings.addProvider.preset.kimiCnDesc": "Kimi 中國 API，走 Moonshot CN 端點，預置 K2 視覺模型。",
+  "settings.addProvider.preset.kimiGlobalDesc": "Kimi 國際 API，走 Moonshot global 端點，預置 K2 視覺模型。",
+  "settings.addProvider.preset.kimiCodingPlanDesc": "Kimi Coding Plan，走獨立 Anthropic-compatible 端點，不歸屬 CN/Global。",
+  "settings.addProvider.preset.mimoApiDesc": "小米 MiMo 直連 API，適合 v2.5 按量付費模型。",
+  "settings.addProvider.preset.mimoAnthropicDesc": "小米 MiMo 直連 Anthropic-compatible 端點，自動使用 MiMo v2.5 模型。",
+  "settings.addProvider.preset.mimoTokenPlanCnDesc": "小米 MiMo token plan 中國區端點。",
+  "settings.addProvider.preset.mimoTokenPlanCnAnthropicDesc": "小米 MiMo token plan 中國區 Anthropic-compatible 端點。",
+  "settings.addProvider.preset.mimoTokenPlanSgpDesc": "小米 MiMo token plan 新加坡端點。",
+  "settings.addProvider.preset.mimoTokenPlanSgpAnthropicDesc": "小米 MiMo token plan 新加坡 Anthropic-compatible 端點。",
+  "settings.addProvider.preset.mimoTokenPlanAmsDesc": "小米 MiMo token plan 阿姆斯特丹端點。",
+  "settings.addProvider.preset.mimoTokenPlanAmsAnthropicDesc": "小米 MiMo token plan 阿姆斯特丹 Anthropic-compatible 端點。",
+  "settings.addProvider.preset.minimaxCnApiDesc": "MiniMax 中國 OpenAI-compatible M 系列 API 端點。",
+  "settings.addProvider.preset.minimaxGlobalApiDesc": "MiniMax 國際 OpenAI-compatible M 系列 API 端點。",
+  "settings.addProvider.preset.minimaxCnAnthropicDesc": "MiniMax 中國 Anthropic-compatible M 系列端點。",
+  "settings.addProvider.preset.minimaxGlobalAnthropicDesc": "MiniMax 國際 Anthropic-compatible 端點，自動使用 Bearer 認證。",
+  "settings.addProvider.preset.glmCnDesc": "智譜 GLM 中國 API，預置 GLM thinking 控制。",
+  "settings.addProvider.preset.zaiGlobalDesc": "Z.AI 國際 GLM API。",
+  "settings.addProvider.preset.glmCodingPlanCnDesc": "智譜 GLM 中國 coding plan 端點，預設 GLM-5.2。",
+  "settings.addProvider.preset.glmCodingPlanCnAnthropicDesc": "智譜 GLM 中國 Anthropic-compatible coding plan 端點，預設 GLM-5.2[1m]。",
+  "settings.addProvider.preset.zaiCodingPlanGlobalDesc": "Z.AI 國際 coding plan 端點，預設 GLM-5.2。",
+  "settings.addProvider.preset.zaiCodingPlanGlobalAnthropicDesc": "Z.AI 國際 Anthropic-compatible coding plan 端點，預設 GLM-5.2[1m]。",
+  "settings.addProvider.preset.opencodeGoDesc": "OpenCode Go 混合模型閘道，內建每模型能力覆蓋。",
+  "settings.addProvider.preset.opencodeGoAnthropicDesc": "OpenCode Go Anthropic 線路，預置 Qwen 與 MiniMax 模型。",
+  "settings.addProvider.preset.opencodeZenAnthropicDesc": "OpenCode Zen Anthropic 線路，預置 Claude 與 Qwen 模型。",
+  "settings.addProvider.preset.qwenCnDesc": "Qwen/DashScope 中國標準 API 端點。",
+  "settings.addProvider.preset.qwenGlobalDesc": "Qwen/DashScope 國際標準 API 端點。",
+  "settings.addProvider.preset.qwenCodingPlanCnDesc": "Qwen Coding Plan 中國 OpenAI-compatible 端點，預設 qwen3.7-plus。",
+  "settings.addProvider.preset.qwenCodingPlanCnAnthropicDesc": "Qwen Coding Plan 中國 Anthropic-compatible 端點，預設 qwen3.7-plus。",
+  "settings.addProvider.preset.qwenCodingPlanGlobalDesc": "Qwen Coding Plan 國際 OpenAI-compatible 端點，預設 qwen3.7-plus。",
+  "settings.addProvider.preset.qwenCodingPlanGlobalAnthropicDesc": "Qwen Coding Plan 國際 Anthropic-compatible 端點，預設 qwen3.7-plus。",
+  "settings.addProvider.preset.stepfunDesc": "階躍星辰 StepFun coding plan OpenAI-compatible 端點，預設 step-3.7-flash。",
+  "settings.addProvider.preset.stepfunAnthropicDesc": "階躍星辰 StepFun coding plan Anthropic-compatible 端點，預設 step-3.7-flash。",
+  "settings.addProvider.preset.novitaDesc": "NovitaAI 多模型閘道，預置 Kimi、MiniMax、GLM、DeepSeek、Qwen。",
+  "settings.addProvider.preset.gmiDesc": "GMI Cloud 多模型直連閘道，預置 GLM、DeepSeek、Kimi、Gemini、Claude、GPT。",
+  "settings.addProvider.preset.vercelAiGatewayDesc": "Vercel AI Gateway，走 Anthropic-compatible Messages API。",
+  "settings.addProvider.preset.huggingfaceDesc": "HuggingFace Inference Router，預置 Qwen 與 DeepSeek 模型。",
+  "settings.addProvider.preset.nvidiaDesc": "NVIDIA NIM 加速推理端點，預置 Nemotron 模型。",
+  "settings.addProvider.preset.kilocodeDesc": "Kilo Code 閘道，預置 kilo/auto 自動路由模型。",
+  "settings.addProvider.preset.ollamaCloudDesc": "Ollama Cloud 託管端點，支援 none 到 max 的 reasoning effort。",
+  "settings.addProvider.addedBadge": "已新增",
+  "settings.addProvider.modifiedBadge": "已修改",
+  "settings.addProvider.nameConflictBadge": "同名占用",
+  "settings.addProvider.similarExistingBadge": "有相似設定",
   "settings.addProvider.confirm": "新增供應商",
+  "settings.addProvider.alreadyAddedAction": "已新增",
+  "settings.addProvider.nameConflictAction": "先處理同名設定",
+  "settings.addProvider.viewPresetProvider": "查看設定",
+  "settings.addProvider.viewConflictProvider": "查看同名設定",
+  "settings.addProvider.resetPreset": "重設為預設",
+  "settings.addProvider.confirmResetPreset": "確認重設",
   "settings.customProviderName": "自訂供應商名稱",
   "settings.customProviderNamePlaceholder": "",
   "settings.providerNamePlaceholder": "供應商名稱（如 deepseek-proxy）",
   "settings.providerProtocol": "接入協定",
   "settings.providerProtocolOpenAI": "OpenAI-compatible",
   "settings.providerProtocolOpenAIHint": "適用於相容 OpenAI API 的代理、聚合平台和自建服務。",
+  "settings.providerProtocolAnthropic": "Anthropic-compatible",
+  "settings.providerProtocolAnthropicHint": "適用於 Claude Messages API 風格的服務，例如部分 coding plan 端點。",
   "settings.providerBaseUrlLabel": "API 位址",
   "settings.providerChatUrlLabel": "完整請求 URL",
   "settings.providerChatUrlPlaceholder": "例如 https://api.openai.com/v1/chat/completions",
@@ -911,6 +1047,12 @@ export const zhTW: Record<DictKey, string> = {
   "settings.providerHeaders": "額外請求標頭（可選）",
   "settings.providerHeadersPlaceholder": "HTTP-Referer: https://example.com\nX-Title: Reasonix",
   "settings.providerHeadersHint": "用於 OpenRouter 等要求站點來源的中轉站。一行一個 Header: value；API Key 仍透過上方金鑰保存。",
+  "settings.providerExtraBody": "額外請求體（可選）",
+  "settings.providerExtraBodyPlaceholder": "{\n  \"enable_thinking\": true\n}",
+  "settings.providerExtraBodyHint": "會合併到聊天請求體頂層 JSON。model、messages、tools、stream、thinking 等核心欄位仍由 Reasonix 控制。",
+  "settings.providerExtraBodyError": "請輸入不含 null 值的 JSON 物件，例如 { \"enable_thinking\": true }。",
+  "settings.providerAuthHeader": "使用 Authorization: Bearer 發送金鑰",
+  "settings.providerAuthHeaderHint": "適用於 MiniMax Global、Vercel AI Gateway 等不使用 x-api-key 的 Anthropic-compatible 閘道。",
   "settings.fetchingModels": "重新整理中…",
   "settings.fetchModelsSuccess": "已取得 {n} 個模型",
   "settings.fetchModelsManualFallbackAuth": "模型發現介面拒絕了金鑰或權限。請檢查 API Key；如果該服務不開放模型發現，也可以在下方手動填寫模型 ID。",
@@ -928,10 +1070,16 @@ export const zhTW: Record<DictKey, string> = {
   "settings.newProviderDraftName": "新供應商",
 	  "settings.reasoningProtocol": "模型能力模式",
 	  "settings.reasoningProtocol.auto": "自動識別（推薦）",
-	  "settings.reasoningProtocol.deepseek": "DeepSeek 思考",
-	  "settings.reasoningProtocol.openai": "OpenAI reasoning",
+	  "settings.reasoningProtocol.deepseek": "DeepSeek 思考（thinking + 深度）",
+	  "settings.reasoningProtocol.openai": "OpenAI reasoning（reasoning_effort）",
 	  "settings.reasoningProtocol.none": "普通聊天（不傳送思考參數）",
 	  "settings.reasoningProtocolHint": "預設自動識別。只有代理不相容，或你確定模型需要特定 reasoning 格式時再切換。",
+	  "settings.thinkingMode": "Thinking 覆蓋",
+	  "settings.thinkingMode.auto": "Auto（使用服務預設）",
+	  "settings.thinkingMode.enabled": "Enabled（開啟）",
+	  "settings.thinkingMode.disabled": "Disabled（關閉）",
+	  "settings.thinkingMode.adaptive": "Adaptive（自適應）",
+	  "settings.thinkingModeHint": "服務商專用覆蓋項。相容閘道可用 enabled/disabled；只有服務文件明確支援時才使用 adaptive。",
   "settings.updateKey": "更新 {env}（全域儲存）",
   "settings.updateKeyAction": "更新金鑰",
   "settings.keyStatus": "金鑰",
@@ -985,6 +1133,8 @@ export const zhTW: Record<DictKey, string> = {
   "msg.imageAttachment": "圖片",
   "msg.workspaceReference": "工作區引用",
   "msg.folderReference": "資料夾 · 工作區引用",
+  "msg.pastedExpandTooltip": "展開貼上文字",
+  "msg.pastedCollapseTooltip": "收起貼上文字",
   "turnActions.summary": "總結",
   "turnActions.rewind": "回溯",
   "transcript.jumpToBottom": "跳到底部",
@@ -1451,11 +1601,76 @@ export const zhTW: Record<DictKey, string> = {
   "settings.toggleOff": "關閉",
   "settings.botSetup": "機器人入口",
   "settings.botSetupHint": "先開啟入口，再選擇要連線的聊天頻道。",
-  "settings.botConnectPhoneTitle": "新增 IM Bot",
-  "settings.botConnectPhoneSubtitle": "QQ 手動配置，飛書、Lark 和微信掃碼接入。",
   "settings.botGateway": "機器人閘道器",
-  "settings.botGatewayHint": "桌面端統一管理 IM bot 的總開關、模型和執行參數。",
+  "settings.botGatewayHint": "桌面端統一管理 IM bot 的總開關和連線狀態。",
   "settings.botBrowserPreviewWarning": "目前是瀏覽器預覽，只會儲存 mock 設定，不會寫入本機配置或啟動 QQ/IM Bot gateway。請在桌面應用或 CLI 中配置後再到 QQ 用戶端測試。",
+  "settings.botSimpleEyebrow": "傻瓜式設定",
+  "settings.botSimpleTitle": "行動端與聊天機器人",
+  "settings.botSimpleHint": "先用手機掃碼遠端控制目前工作區；需要長時間從聊天工具存取時，再選擇一個 Bot Channel 配置。",
+  "settings.botMobileRemoteEyebrow": "獨立入口",
+  "settings.botMobileRemoteTitle": "手機遠端控制",
+  "settings.botMobileRemoteHint": "掃碼或複製連結，從行動端打開目前工作區的控制入口；它和 IM Bot 頻道管理分開配置。",
+  "settings.botMobileRemoteControlApi": "控制連結",
+  "settings.botMobileRemoteDisabled": "開啟後生成二維碼和連結",
+  "settings.botMobileRemoteEnableFirst": "開啟後生成二維碼",
+  "settings.botMobileRemoteReady": "等待手機連線",
+  "settings.botMobileRemoteWaiting": "手機掃碼連線",
+  "settings.botMobileRemoteLoopbackNotice": "目前控制 API 仍是本機 loopback，只適合本機預覽；跨裝置 relay 接入後會沿用這個入口。",
+  "settings.botMobileRemoteCopyLink": "複製連結",
+  "settings.botMobileRemoteCopied": "已複製",
+  "settings.botMobileRemoteCopyFailed": "複製失敗",
+  "settings.botChannelEntryTitle": "使用 Bot Channel",
+  "settings.botChannelEntryHint": "連接聊天 Bot，適合更長時間的行動端存取。",
+  "settings.botManageBots": "機器人管理",
+  "settings.botManageBotsHint": "左側選擇頻道；未配置時掃碼或填寫金鑰，配置後在右側管理這個機器人的執行設定。",
+  "settings.botBackToOverview": "返回入口",
+  "settings.botChannelOpenConfigured": "已配置，可繼續修改設定。",
+  "settings.botChannelNeedsSetup": "未配置",
+  "settings.botChannelTabsLabel": "機器人頻道",
+  "settings.botChannelSetupTitle": "配置 {provider}",
+  "settings.botChannelSetupHint": "未配置前只需要完成憑據或掃碼綁定；成功後會自動切換為具體設定。",
+  "settings.botSetupStatus": "機器人設定狀態",
+  "settings.botSetupStatusConnected": "{online}/{total} 在線",
+  "settings.botSetupStatusAccessTrusted": "受信任名單 {count}",
+  "settings.botSetupStatusAccessEveryone": "所有人可用",
+  "settings.botSetupStatusAccessPerBot": "{configured}/{total} 個 Bot 已配置",
+  "settings.botMode": "機器人設定模式",
+  "settings.botModeSimple": "普通",
+  "settings.botModeAdvanced": "高階",
+  "settings.botShowAdvancedSettings": "高階設定",
+  "settings.botHideAdvancedSettings": "收起高階設定",
+  "settings.botAdvancedSettingsHint": "低頻能力：路由、佇列細項和自身訊息過濾。",
+  "settings.botShowAddBot": "新增頻道",
+  "settings.botStepConnectTitle": "連接機器人",
+  "settings.botStepConnectHint": "QQ 手動填密鑰；飛書、Lark、微信掃碼即可接入。",
+  "settings.botStepConnectEmpty": "還沒有連接",
+  "settings.botSharedAllowlistHint": "所有渠道共用第 2 步「誰可以使用」的名單與配對設定，改一處即可全部生效。",
+  "settings.botSharedAllowlistGo": "前往第 2 步編輯",
+  "settings.botGlobalAllowlist": "舊全域白名單",
+  "settings.botGlobalAllowlistHint": "相容舊配置；新 Bot 建議在各自詳情裡設定誰可以使用。",
+  "settings.botAccessShowAllPlatforms": "顯示全部平台",
+  "settings.botAccessShowConnectedOnly": "只看已連接平台",
+  "settings.botListUsers": "使用者",
+  "settings.botListGroups": "群組",
+  "settings.botListApprovers": "審批人",
+  "settings.botListAdmins": "管理員",
+  "settings.botPlatformFeishuLark": "飛書 / Lark",
+  "settings.botQueueDropLabel": "佇列滿時",
+  "settings.botQueueDropHint": "佇列達到上限後，如何處理放不下的訊息。",
+  "settings.botPairingDetailHint": "配對開關在第 2 步「誰可以使用」裡；這裡只調整配對碼有效期和待批數量。",
+  "settings.botSimpleAccessTitle": "誰可以使用",
+  "settings.botSimpleAccessTrustedSummary": "目前使用受信任名單，已填寫 {count} 條成員或群組。",
+  "settings.botSimpleAccessEveryoneSummary": "目前允許所有人使用，適合短時間測試。",
+  "settings.botAccessTrusted": "受信任名單",
+  "settings.botAccessTrustedHint": "只允許填寫過的使用者或群組呼叫機器人。",
+  "settings.botAccessEveryone": "所有人可用",
+  "settings.botAccessEveryoneHint": "臨時測試方便，但不建議長期開啟。",
+  "settings.botAccessPairing": "私聊先配對",
+  "settings.botAccessPairingHint": "陌生私聊先發配對碼，本機批准後再放行。",
+  "settings.botRoleAccess": "管理員與審批人",
+  "settings.botRoleAccessHint": "可選。普通使用者不用填；需要遠端審批或管理時再配置。",
+  "settings.botQueueModeSimple": "新訊息怎麼處理",
+  "settings.botQueueModeSimpleHint": "當機器人還在處理上一條訊息時，選擇下一條訊息的處理方式。",
   "settings.saveBotSettings": "儲存機器人設定",
   "settings.botEnableBot": "啟用機器人",
   "settings.botGatewayMasterSwitch": "總開關",
@@ -1466,8 +1681,48 @@ export const zhTW: Record<DictKey, string> = {
   "settings.botModelHint": "留空時使用預設模型。",
   "settings.botModelDefault": "使用預設模型",
   "settings.botRuntime": "執行限制",
+  "settings.botRuntimeHint": "閘道器行程級限制，控制每次機器人任務最多走多遠。",
   "settings.botMaxSteps": "最大步數",
   "settings.botDebounceMs": "合併視窗 ms",
+  "settings.botQueueCap": "佇列上限",
+  "settings.botQueueMode.steer": "引導目前回合",
+  "settings.botQueueMode.followup": "追加跟進",
+  "settings.botQueueMode.collect": "收集等待",
+  "settings.botQueueMode.interrupt": "打斷目前",
+  "settings.botQueueDrop.summarize": "彙總溢出",
+  "settings.botQueueDrop.old": "丟棄最舊",
+  "settings.botQueueDrop.new": "丟棄最新",
+  "settings.botIgnoreSelfMessages": "忽略自身回聲",
+  "settings.botIgnoreSelfMessagesHint": "避免 gateway 再次處理由已配置 Bot 身份發出的訊息。",
+  "settings.botSelfUserIds": "Bot 自身 ID",
+  "settings.botSelfUserIdsHint": "可選填寫各平台 Bot 自己的 user/open ID，用於過濾回聲訊息。",
+  "settings.botPairing": "私聊配對",
+  "settings.botPairingTTL": "有效分鐘",
+  "settings.botPairingMaxPending": "待批上限",
+  "settings.botControlApi": "本機控制 API",
+  "settings.botControlApiHint": "僅 loopback 的 HTTP API，可查狀態和受控傳送；需要從指定環境變數讀取 Bearer token。",
+  "settings.botControlAddr": "控制地址",
+  "settings.botControlTokenEnv": "Token 變數",
+  "settings.botGatewayDefaults": "閘道器執行設定",
+  "settings.botGatewayDefaultsHint": "不用編輯 TOML，也能配置排隊、配對和自身訊息過濾。",
+  "settings.botRoutes": "聊天綁定到專案",
+  "settings.botRoutesHint": "把某個群組、聯絡人或執行緒固定到指定工作目錄、模型或審批方式。",
+  "settings.botRoutesEmpty": "還沒有綁定規則。新增一條規則，把某個聊天、使用者或執行緒綁定到專案或模型。",
+  "settings.botRouteTitle": "路由 {n}",
+  "settings.botAddRoute": "新增路由",
+  "settings.botRouteConnection": "來自哪個 Bot",
+  "settings.botRouteAny": "任意",
+  "settings.botRoutePlatform": "平台",
+  "settings.botRouteChatType": "聊天類型",
+  "settings.botRouteChatType.any": "任意",
+  "settings.botRouteChatType.dm": "私聊",
+  "settings.botRouteChatType.group": "群聊",
+  "settings.botRouteChatType.guild": "頻道",
+  "settings.botRouteChatType.direct": "Direct",
+  "settings.botRouteChatType.thread": "Thread",
+  "settings.botRouteChatId": "群組/聊天 ID",
+  "settings.botRouteUserId": "聯絡人 ID",
+  "settings.botRouteThreadId": "執行緒 ID",
   "settings.botChannels": "連線頻道",
   "settings.botChannelsHint": "選擇頻道；QQ 手動配置，飛書、Lark 和微信掃碼連線。",
   "settings.botQQHint": "QQ 官方 Bot API v2。",
@@ -1522,8 +1777,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.botInstallLarkHint": "掃碼建立 Lark PersonalAgent。",
   "settings.botInstallWeixinHint": "微信掃碼登入並儲存本地 token。",
   "settings.botConnectedBots": "已連線 Bot",
-  "settings.botConnectedBotsSummary": "{online}/{total} 線上",
-  "settings.botConnectionsEmpty": "還沒有連線的機器人頻道。",
+  "settings.botSelectBotHint": "選擇左側 Bot 進入配置。",
   "settings.botConnectionColumnChannel": "頻道",
   "settings.botConnectionColumnName": "機器人",
   "settings.botConnectionColumnRemote": "遠端 ID",
@@ -1541,16 +1795,17 @@ export const zhTW: Record<DictKey, string> = {
   "settings.botConnectionConnected": "已連線",
   "settings.botConnectionConfigured": "已配置",
   "settings.botConnectionDisconnected": "未連線",
-  "settings.botManage": "高階設定",
+  "settings.botManage": "修改配置",
   "settings.deleteBot": "刪除機器人",
+  "settings.deleteBotShort": "刪除",
   "settings.confirmDeleteBot": "確認刪除",
   "settings.deleteBotHint": "從已連線 Bot 中移除這個機器人；不會自動清除已儲存的金鑰。",
   "settings.botChannelModel": "模型",
-  "settings.botChannelModelHint": "留空時使用 Bot 預設模型。",
-  "settings.botChannelModelAuto": "自動",
+  "settings.botChannelModelHint": "這個機器人處理訊息時使用的模型。",
+  "settings.botChannelModelAuto": "使用桌面目前模型",
   "settings.botToolApprovalMode": "工具審批",
   "settings.botToolApprovalModeHint": "控制該頻道是否在工具執行前請求確認；計劃確認和 ask 問題仍會等待回覆。",
-  "settings.botToolApprovalMode.inherit": "跟隨預設",
+  "settings.botToolApprovalMode.inherit": "跟隨機器人",
   "settings.botToolApprovalMode.ask": "詢問",
   "settings.botToolApprovalMode.auto": "自動",
   "settings.botToolApprovalMode.yolo": "YOLO",
@@ -1573,29 +1828,20 @@ export const zhTW: Record<DictKey, string> = {
   "settings.botDiagnosticReportFailed": "傳送診斷報告失敗。",
   "settings.botDiagnosticPrivacy": "僅包含已脫敏的連線元資訊。",
   "settings.botAccessControl": "存取控制",
-  "settings.botAccessControlHint": "預設使用白名單；臨時測試可以改為允許所有人。",
-  "settings.botAccessMode": "誰可以使用",
-  "settings.botAccessAll": "所有人",
-  "settings.botAccessWhitelist": "僅白名單",
   "settings.botEditAllowlist": "編輯白名單",
   "settings.botAllowlist": "白名單",
-  "settings.botAllowlistHint": "限制哪些使用者或群可以呼叫機器人。",
   "settings.botAllowlistEnabled": "啟用白名單",
   "settings.botAllowAll": "允許所有人",
   "settings.botAllowAllWarn": "開啟後所有進入頻道的訊息都可以呼叫機器人。",
-  "settings.botQQAccessRequired": "啟用 QQ Bot 前，請先在高階設定裡新增 QQ 使用者/群白名單，或臨時改為允許所有人。",
-  "settings.botAllowlistEntries": "使用者與群",
+  "settings.botQQAccessRequired": "啟用 QQ Bot 前，請先在第 2 步「誰可以使用」裡新增 QQ 使用者/群組，或臨時改為所有人可用。",
   "settings.botQQUsers": "QQ 使用者",
-  "settings.botQQGroups": "QQ群",
   "settings.botFeishuUsers": "飛書使用者",
   "settings.botFeishuGroups": "飛書群",
   "settings.botFeishuLarkUsers": "飛書 / Lark 使用者",
-  "settings.botFeishuLarkGroups": "飛書 / Lark 群",
   "settings.botWeixinUsers": "微信使用者",
-  "settings.botWeixinGroups": "微信群",
   "settings.botListPlaceholder": "每行一個 ID，也可用逗號分隔",
   "settings.shellInterpreter": "Shell 直譯器",
-  "settings.shellAuto": "自動（優先 bash，Windows 無 bash 時用 PowerShell）",
+  "settings.shellAuto": "自動（bash 優先）",
   "settings.shellBash": "bash",
   "settings.shellPowershell": "PowerShell（Windows PowerShell）",
   "settings.shellPwsh": "pwsh（PowerShell 7+）",
@@ -1745,7 +1991,7 @@ export const zhTW: Record<DictKey, string> = {
   "settings.telemetryLabel": "匿名啟動統計",
   "settings.telemetryHint": "啟動時傳送隨機安裝 ID、版本號和作業系統用於統計活躍安裝量；若上次執行發生崩潰，則在本次啟動補發該崩潰報告。絕不包含對話、金鑰或檔案內容。",
   "settings.metricsLabel": "共享聚合品質指標",
-  "settings.metricsHint": "預設開啟。傳送匿名的輪次結束統計、設定偏好快照，以及 Memory v5 聚合桶，例如是否注入、大小桶、IR 數量桶和記憶圖規模桶。包含用於 DAU 去重的隨機安裝 ID。Bucket 可能包含正規化後的自訂 Provider 名和模型名——絕不包含對話、提示詞、金鑰、路徑、base URL、記憶正文、工具輸出或檔案內容。",
+  "settings.metricsHint": "預設開啟。傳送匿名的輪次結束統計、更新器錯誤類別、設定偏好快照，以及 Memory v5 聚合桶，例如是否注入、大小桶、IR 數量桶和記憶圖規模桶。包含用於 DAU 去重的隨機安裝 ID。Bucket 可能包含正規化後的自訂 Provider 名和模型名——絕不包含對話、提示詞、金鑰、路徑、base URL、記憶正文、工具輸出或檔案內容。",
   "context.windowTitle": "上下文視窗",
   "context.windowStatusHealthy": "上下文充足",
   "context.windowStatusWatch": "即將壓縮",

desktop/frontend/src/locales/zh.ts

@@ -206,6 +206,7 @@ export const zh: Record<DictKey, string> = {
   "caps.mcpSummary": "{connected} 个已连接 · {failed} 个异常 · {tools} 个工具",
   "caps.connectorsTab": "MCP 服务器",
   "caps.skillsTab": "Agent 技能",
+  "caps.pluginsTab": "插件",
   "caps.servers": "MCP 服务器",
   "caps.noServers": "未配置 MCP 服务器。用 /mcp add 添加。",
   "caps.availableServers": "可用 MCP",
@@ -312,6 +313,62 @@ export const zh: Record<DictKey, string> = {
   "caps.trustedReadOnlyTitle": "已记住为之后规划和只读 research 可用的只读工具",
   "caps.searchSkills": "搜索技能…",
   "caps.noSkillMatches": "没有匹配的技能。",
+  "caps.pluginRefresh": "刷新插件",
+  "caps.pluginInstallTitle": "安装插件",
+  "caps.pluginInstallHint": "选择本地插件目录，或从 Git 仓库安装。",
+  "caps.pluginInstallMethod": "插件安装方式",
+  "caps.pluginInstallLocal": "本地目录",
+  "caps.pluginInstallGit": "Git 仓库",
+  "caps.pluginSource": "插件来源",
+  "caps.pluginGitSource": "Git 仓库地址",
+  "caps.pluginSourcePlaceholder": "git:github.com/obra/superpowers",
+  "caps.pluginChooseLocalFolder": "选择插件目录",
+  "caps.pluginLocalFolder": "本地插件目录",
+  "caps.pluginNoLocalFolder": "尚未选择目录",
+  "caps.pluginInstallName": "安装名称（可选）",
+  "caps.pluginInstallNamePlaceholder": "安装名称（可选）",
+  "caps.pluginLink": "开发模式：链接源目录",
+  "caps.pluginLinkHint": "不复制插件，直接使用所选目录；适合调试插件。移动或删除该目录会导致插件失效。",
+  "caps.pluginReplace": "覆盖同名插件",
+  "caps.pluginReplaceHint": "插件名已存在时，允许用当前来源替换它。",
+  "caps.pluginPreview": "预检",
+  "caps.pluginInstall": "安装插件",
+  "caps.noPluginsTitle": "尚未安装插件",
+  "caps.noPluginsHint": "从 Git 或本地路径安装插件包，把技能、hooks 和 MCP 服务器作为一个统一单元管理。",
+  "caps.installedPlugins": "已安装插件",
+  "caps.pluginsSummary": "{enabled} 个启用 · 共 {total} 个 · {issues} 个异常",
+  "caps.pluginCounts": "{skills} 技能 · {hooks} hooks · {mcps} MCP",
+  "caps.pluginNoCapabilities": "未导出能力",
+  "caps.pluginWarnings": "{count} 个告警",
+  "caps.pluginEnable": "启用插件",
+  "caps.pluginDisable": "停用插件",
+  "caps.pluginEnabled": "已启用",
+  "caps.pluginDisabled": "已停用",
+  "caps.pluginVersion": "版本",
+  "caps.pluginRoot": "安装路径",
+  "caps.pluginUpdate": "更新",
+  "caps.pluginDoctor": "诊断",
+  "caps.pluginRemove": "移除插件",
+  "caps.pluginConfirmRemove": "确认移除",
+  "caps.pluginPlanError": "安装检查失败",
+  "caps.pluginPlanReady": "安装计划",
+  "caps.pluginAction": "操作",
+  "caps.pluginPlanStatus": "插件操作：{status}",
+  "caps.pluginPlanComplete": "插件操作完成",
+  "caps.pluginPlanInstalled": "插件已安装。展开下方插件，可查看它导出的 skills、hooks 和 MCP 服务器的使用方式。",
+  "caps.pluginUsageTitle": "使用方法",
+  "caps.pluginUsageEnabledHint": "启用的插件会加载到新会话中。可用 /skills 查看技能、用 /<skill> 直接调用，或自然描述任务；hooks 和 MCP 服务器会按正常会话流程自动参与。",
+  "caps.pluginUsageDisabledHint": "启用此插件后，它的 skills、hooks 或 MCP 服务器才会参与会话。",
+  "caps.pluginNoCapabilityDetails": "此插件暂无可展示的能力明细。",
+  "caps.pluginSkillsTitle": "Skills",
+  "caps.pluginSkillsHint": "用 /skills 浏览，直接调用，或在对话中说出工作流名称。",
+  "caps.pluginHooksTitle": "Hooks",
+  "caps.pluginHooksHint": "Hooks 会在匹配的生命周期事件中自动触发。",
+  "caps.pluginMCPTitle": "MCP 服务器",
+  "caps.pluginMCPHint": "这些服务器提供的工具会通过正常 MCP/工具流程可用。",
+  "caps.pluginNoDescription": "无描述",
+  "caps.pluginHookNoTarget": "未指定 hook 目标",
+  "caps.pluginMCPNoTarget": "未指定启动目标",
   "caps.sources": "来源",
   "caps.sourcesSummaryActive": "{active} 个有效",
   "caps.sourcesSummaryMissing": "{missing} 个缺失",
@@ -433,6 +490,9 @@ export const zh: Record<DictKey, string> = {
   "composer.pastedExpand": "展开",
   "composer.pastedRemove": "删除粘贴文本",
   "composer.removeImage": "移除图片",
+  "imageViewer.clickToPreview": "点击查看大图",
+  "imageViewer.closePreview": "关闭预览",
+  "imageViewer.title": "图片预览",
   "composer.attachImageFailed": "图片粘贴失败",
   "composer.attachFileFailed": "文件附加失败",
   "composer.attachDropFailed": "拖放文件附加失败",
@@ -672,6 +732,18 @@ export const zh: Record<DictKey, string> = {
   "history.historySessionActions": "历史会话操作",
   "history.trashActions": "回收站操作",
 
+  // 会话恢复
+  "recovery.branch": "恢复分支",
+  "recovery.badge": "恢复",
+  "recovery.open": "打开恢复分支",
+  "recovery.toast": "未保存的本地内容已保存为 {title}。",
+  "recovery.failedLease": "恢复分支被另一个运行时占用，未切换过去。",
+  "recovery.failedUnavailable": "无法打开恢复分支，当前标签仍停留在原会话。",
+  "recovery.banner": "当前会话来自保存冲突恢复。继续前建议与原会话对比。",
+  "recovery.reason": "原因：{reason}",
+  "recovery.digest": "摘要：{digest}",
+  "recovery.parent": "父会话：{parent}",
+
   // 项目树
   "projectTree.workspaceTitle": "项目",
   "projectTree.pinnedTitle": "置顶",
@@ -846,6 +918,7 @@ export const zh: Record<DictKey, string> = {
   "settings.tab.capabilities": "MCP 与技能",
   "settings.tab.mcp": "MCP 与工具",
   "settings.tab.skills": "技能",
+  "settings.tab.plugins": "插件",
   "settings.tab.memory": "记忆",
   "settings.tab.hooks": "Hooks",
   "settings.tab.shortcuts": "快捷键",
@@ -861,6 +934,7 @@ export const zh: Record<DictKey, string> = {
   "settings.tabSub.capabilities": "服务器、工具与技能",
   "settings.tabSub.mcp": "服务器与工具",
   "settings.tabSub.skills": "来源与启用",
+  "settings.tabSub.plugins": "包与诊断",
   "settings.tabSub.memory": "项目与全局",
   "settings.tabSub.hooks": "Shell 自动化",
   "settings.tabSub.shortcuts": "按键与帮助",
@@ -942,6 +1016,10 @@ export const zh: Record<DictKey, string> = {
   "settings.subagentEffort": "子代理 effort",
   "settings.subagentEffortDefault": "auto（模型服务默认）",
   "settings.subagentHint": "作用于 task 和 runAs=subagent skills；工具调用参数或单个 skill 配置会覆盖它。",
+  "settings.subagentDepth": "子代理嵌套深度",
+  "settings.subagentDepthHint": "控制第一层子代理能否再派发一层 reviewer 或 worker 子代理。",
+  "settings.subagentDepthOne": "1 层",
+  "settings.subagentDepthTwo": "2 层",
   "settings.defaultToolApprovalMode": "新会话默认审批",
   "settings.defaultToolApprovalModeHint": "保存在全局设置中，仅用于新建会话；切换或恢复已有会话时不会覆盖其 Ask / Auto / Yolo 状态。",
   "settings.defaultToolApprovalMode.ask": "询问",
@@ -964,7 +1042,7 @@ export const zh: Record<DictKey, string> = {
   "settings.coldResumePrune.on": "开",
   "settings.coldResumePrune.off": "关",
   "settings.reasoningLanguage": "思考语言",
-  "settings.reasoningLanguageHint": "只影响可见思考过程。自动跟随对话语言，不额外注入语言指令。",
+  "settings.reasoningLanguageHint": "只影响可见思考过程。自动在原始提问明显为中文时锚定中文，引用内容不参与判断。",
   "settings.reasoningLanguage.auto": "自动",
   "settings.reasoningLanguage.zh": "中文",
   "settings.reasoningLanguage.en": "English",
@@ -1071,11 +1149,76 @@ export const zh: Record<DictKey, string> = {
   "settings.toggleOff": "关闭",
   "settings.botSetup": "机器人入口",
   "settings.botSetupHint": "先开启入口，再选择要连接的聊天渠道。",
-  "settings.botConnectPhoneTitle": "添加 IM Bot",
-  "settings.botConnectPhoneSubtitle": "QQ 手动配置，飞书、Lark 和微信扫码接入。",
   "settings.botGateway": "机器人网关",
-  "settings.botGatewayHint": "桌面端统一管理 IM bot 的总开关、模型和运行参数。",
+  "settings.botGatewayHint": "桌面端统一管理 IM bot 的总开关和连接状态。",
   "settings.botBrowserPreviewWarning": "当前是浏览器预览，只会保存 mock 设置，不会写入本机配置或启动 QQ/IM Bot gateway。请在桌面应用或 CLI 中配置后再到 QQ 客户端测试。",
+  "settings.botSimpleEyebrow": "傻瓜式配置",
+  "settings.botSimpleTitle": "移动端与聊天机器人",
+  "settings.botSimpleHint": "先用手机扫码远程控制当前工作区；需要长期从聊天工具访问时，再选择一个 Bot Channel 配置。",
+  "settings.botMobileRemoteEyebrow": "独立入口",
+  "settings.botMobileRemoteTitle": "手机远程控制",
+  "settings.botMobileRemoteHint": "扫码或复制链接，从移动端打开当前工作区的控制入口；它和 IM Bot 渠道管理分开配置。",
+  "settings.botMobileRemoteControlApi": "控制链接",
+  "settings.botMobileRemoteDisabled": "开启后生成二维码和链接",
+  "settings.botMobileRemoteEnableFirst": "开启后生成二维码",
+  "settings.botMobileRemoteReady": "等待手机连接",
+  "settings.botMobileRemoteWaiting": "手机扫码连接",
+  "settings.botMobileRemoteLoopbackNotice": "当前控制 API 仍是本机 loopback，只适合本机预览；跨设备 relay 接入后会沿用这个入口。",
+  "settings.botMobileRemoteCopyLink": "复制链接",
+  "settings.botMobileRemoteCopied": "已复制",
+  "settings.botMobileRemoteCopyFailed": "复制失败",
+  "settings.botChannelEntryTitle": "使用 Bot Channel",
+  "settings.botChannelEntryHint": "连接聊天 Bot，适合更长时间的移动端访问。",
+  "settings.botManageBots": "机器人管理",
+  "settings.botManageBotsHint": "左侧选择渠道；未配置时扫码或填写密钥，配置后在右侧管理这个机器人的运行设置。",
+  "settings.botBackToOverview": "返回入口",
+  "settings.botChannelOpenConfigured": "已配置，可继续修改设置。",
+  "settings.botChannelNeedsSetup": "未配置",
+  "settings.botChannelTabsLabel": "机器人渠道",
+  "settings.botChannelSetupTitle": "配置 {provider}",
+  "settings.botChannelSetupHint": "未配置前只需要完成凭据或扫码绑定；成功后会自动切换为具体设置。",
+  "settings.botSetupStatus": "机器人设置状态",
+  "settings.botSetupStatusConnected": "{online}/{total} 在线",
+  "settings.botSetupStatusAccessTrusted": "受信任名单 {count}",
+  "settings.botSetupStatusAccessEveryone": "所有人可用",
+  "settings.botSetupStatusAccessPerBot": "{configured}/{total} 个 Bot 已配置",
+  "settings.botMode": "机器人设置模式",
+  "settings.botModeSimple": "普通",
+  "settings.botModeAdvanced": "高级",
+  "settings.botShowAdvancedSettings": "高级设置",
+  "settings.botHideAdvancedSettings": "收起高级设置",
+  "settings.botAdvancedSettingsHint": "低频能力：路由、队列细项和自身消息过滤。",
+  "settings.botShowAddBot": "新增渠道",
+  "settings.botStepConnectTitle": "连接机器人",
+  "settings.botStepConnectHint": "QQ 手动填密钥；飞书、Lark、微信扫码即可接入。",
+  "settings.botStepConnectEmpty": "还没有连接",
+  "settings.botSharedAllowlistHint": "所有渠道共用第 2 步「谁可以使用」的名单与配对设置，改一处即可全部生效。",
+  "settings.botSharedAllowlistGo": "前往第 2 步编辑",
+  "settings.botGlobalAllowlist": "旧全局白名单",
+  "settings.botGlobalAllowlistHint": "兼容旧配置；新 Bot 推荐在各自详情里设置谁可以使用。",
+  "settings.botAccessShowAllPlatforms": "显示全部平台",
+  "settings.botAccessShowConnectedOnly": "只看已连接平台",
+  "settings.botListUsers": "用户",
+  "settings.botListGroups": "群",
+  "settings.botListApprovers": "审批人",
+  "settings.botListAdmins": "管理员",
+  "settings.botPlatformFeishuLark": "飞书 / Lark",
+  "settings.botQueueDropLabel": "队列满时",
+  "settings.botQueueDropHint": "队列达到上限后，如何处理放不下的消息。",
+  "settings.botPairingDetailHint": "配对开关在第 2 步「谁可以使用」里；这里只调整配对码有效期和待批数量。",
+  "settings.botSimpleAccessTitle": "谁可以使用",
+  "settings.botSimpleAccessTrustedSummary": "当前使用受信任名单，已填写 {count} 条成员或群。",
+  "settings.botSimpleAccessEveryoneSummary": "当前允许所有人使用，适合短时间测试。",
+  "settings.botAccessTrusted": "受信任名单",
+  "settings.botAccessTrustedHint": "只允许填写过的用户或群调用机器人。",
+  "settings.botAccessEveryone": "所有人可用",
+  "settings.botAccessEveryoneHint": "临时测试方便，但不建议长期打开。",
+  "settings.botAccessPairing": "私聊先配对",
+  "settings.botAccessPairingHint": "陌生私聊先发配对码，本机批准后再放行。",
+  "settings.botRoleAccess": "管理员与审批人",
+  "settings.botRoleAccessHint": "可选。普通用户不用填；需要远程审批或管理时再配置。",
+  "settings.botQueueModeSimple": "新消息怎么处理",
+  "settings.botQueueModeSimpleHint": "当机器人还在处理上一条消息时，选择下一条消息的处理方式。",
   "settings.saveBotSettings": "保存机器人设置",
   "settings.botEnableBot": "启用机器人",
   "settings.botGatewayMasterSwitch": "总开关",
@@ -1086,8 +1229,48 @@ export const zh: Record<DictKey, string> = {
   "settings.botModelHint": "留空时使用默认模型。",
   "settings.botModelDefault": "使用默认模型",
   "settings.botRuntime": "运行限制",
+  "settings.botRuntimeHint": "网关进程级限制，控制每次机器人任务最多走多远。",
   "settings.botMaxSteps": "最大步数",
   "settings.botDebounceMs": "合并窗口 ms",
+  "settings.botQueueCap": "队列上限",
+  "settings.botQueueMode.steer": "引导当前回合",
+  "settings.botQueueMode.followup": "追加跟进",
+  "settings.botQueueMode.collect": "收集等待",
+  "settings.botQueueMode.interrupt": "打断当前",
+  "settings.botQueueDrop.summarize": "汇总溢出",
+  "settings.botQueueDrop.old": "丢弃最旧",
+  "settings.botQueueDrop.new": "丢弃最新",
+  "settings.botIgnoreSelfMessages": "忽略自身回声",
+  "settings.botIgnoreSelfMessagesHint": "避免 gateway 再次处理由已配置 Bot 身份发出的消息。",
+  "settings.botSelfUserIds": "Bot 自身 ID",
+  "settings.botSelfUserIdsHint": "可选填写各平台 Bot 自己的 user/open ID，用于过滤回声消息。",
+  "settings.botPairing": "私聊配对",
+  "settings.botPairingTTL": "有效分钟",
+  "settings.botPairingMaxPending": "待批上限",
+  "settings.botControlApi": "本机控制 API",
+  "settings.botControlApiHint": "仅 loopback 的 HTTP API，可查状态和受控发送；需要从指定环境变量读取 Bearer token。",
+  "settings.botControlAddr": "控制地址",
+  "settings.botControlTokenEnv": "Token 变量",
+  "settings.botGatewayDefaults": "网关运行设置",
+  "settings.botGatewayDefaultsHint": "不用编辑 TOML，也能配置排队、配对和自身消息过滤。",
+  "settings.botRoutes": "聊天绑定到项目",
+  "settings.botRoutesHint": "把某个群、联系人或线程固定到指定工作目录、模型或审批方式。",
+  "settings.botRoutesEmpty": "还没有绑定规则。添加一条规则，把某个聊天、用户或线程绑定到项目或模型。",
+  "settings.botRouteTitle": "路由 {n}",
+  "settings.botAddRoute": "添加路由",
+  "settings.botRouteConnection": "来自哪个 Bot",
+  "settings.botRouteAny": "任意",
+  "settings.botRoutePlatform": "平台",
+  "settings.botRouteChatType": "聊天类型",
+  "settings.botRouteChatType.any": "任意",
+  "settings.botRouteChatType.dm": "私聊",
+  "settings.botRouteChatType.group": "群聊",
+  "settings.botRouteChatType.guild": "频道",
+  "settings.botRouteChatType.direct": "Direct",
+  "settings.botRouteChatType.thread": "Thread",
+  "settings.botRouteChatId": "群/聊天 ID",
+  "settings.botRouteUserId": "联系人 ID",
+  "settings.botRouteThreadId": "线程 ID",
   "settings.botChannels": "连接渠道",
   "settings.botChannelsHint": "选择渠道；QQ 手动配置，飞书、Lark 和微信扫码连接。",
   "settings.botQQHint": "QQ 官方 Bot API v2。",
@@ -1142,8 +1325,7 @@ export const zh: Record<DictKey, string> = {
   "settings.botInstallLarkHint": "扫码创建 Lark PersonalAgent。",
   "settings.botInstallWeixinHint": "微信扫码登录并保存本地 token。",
   "settings.botConnectedBots": "已连接 Bot",
-  "settings.botConnectedBotsSummary": "{online}/{total} 在线",
-  "settings.botConnectionsEmpty": "还没有连接的机器人渠道。",
+  "settings.botSelectBotHint": "选择左侧 Bot 进入配置。",
   "settings.botConnectionColumnChannel": "渠道",
   "settings.botConnectionColumnName": "机器人",
   "settings.botConnectionColumnRemote": "远端 ID",
@@ -1161,16 +1343,17 @@ export const zh: Record<DictKey, string> = {
   "settings.botConnectionConnected": "已连接",
   "settings.botConnectionConfigured": "已配置",
   "settings.botConnectionDisconnected": "未连接",
-  "settings.botManage": "高级设置",
+  "settings.botManage": "修改配置",
   "settings.deleteBot": "删除机器人",
+  "settings.deleteBotShort": "删除",
   "settings.confirmDeleteBot": "确认删除",
   "settings.deleteBotHint": "从已连接 Bot 中移除这个机器人；不会自动清除已保存的密钥。",
   "settings.botChannelModel": "模型",
-  "settings.botChannelModelHint": "留空时使用 Bot 默认模型。",
-  "settings.botChannelModelAuto": "自动",
+  "settings.botChannelModelHint": "这个机器人处理消息时使用的模型。",
+  "settings.botChannelModelAuto": "使用桌面当前模型",
   "settings.botToolApprovalMode": "工具审批",
   "settings.botToolApprovalModeHint": "控制该渠道是否在工具执行前请求确认；计划确认和 ask 问题仍会等待回复。",
-  "settings.botToolApprovalMode.inherit": "跟随默认",
+  "settings.botToolApprovalMode.inherit": "跟随机器人",
   "settings.botToolApprovalMode.ask": "询问",
   "settings.botToolApprovalMode.auto": "自动",
   "settings.botToolApprovalMode.yolo": "YOLO",
@@ -1193,26 +1376,17 @@ export const zh: Record<DictKey, string> = {
   "settings.botDiagnosticReportFailed": "发送诊断报告失败。",
   "settings.botDiagnosticPrivacy": "仅包含已脱敏的连接元信息。",
   "settings.botAccessControl": "访问控制",
-  "settings.botAccessControlHint": "默认使用白名单；临时测试可以改为允许所有人。",
-  "settings.botAccessMode": "谁可以使用",
-  "settings.botAccessAll": "所有人",
-  "settings.botAccessWhitelist": "仅白名单",
   "settings.botEditAllowlist": "编辑白名单",
   "settings.botAllowlist": "白名单",
-  "settings.botAllowlistHint": "限制哪些用户或群可以调用机器人。",
   "settings.botAllowlistEnabled": "启用白名单",
   "settings.botAllowAll": "允许所有人",
   "settings.botAllowAllWarn": "开启后所有进入渠道的消息都可以调用机器人。",
-  "settings.botQQAccessRequired": "启用 QQ Bot 前，请先在高级设置里添加 QQ 用户/群白名单，或临时改为允许所有人。",
-  "settings.botAllowlistEntries": "用户与群",
+  "settings.botQQAccessRequired": "启用 QQ Bot 前，请先在第 2 步「谁可以使用」里添加 QQ 用户/群，或临时改为所有人可用。",
   "settings.botQQUsers": "QQ 用户",
-  "settings.botQQGroups": "QQ群",
   "settings.botFeishuUsers": "飞书用户",
   "settings.botFeishuGroups": "飞书群",
   "settings.botFeishuLarkUsers": "飞书 / Lark 用户",
-  "settings.botFeishuLarkGroups": "飞书 / Lark 群",
   "settings.botWeixinUsers": "微信用户",
-  "settings.botWeixinGroups": "微信群",
   "settings.botListPlaceholder": "每行一个 ID，也可用逗号分隔",
   "settings.permissions": "权限",
   "settings.writerMode": "写操作模式",
@@ -1224,8 +1398,11 @@ export const zh: Record<DictKey, string> = {
   "settings.yoloHint": "Yolo批准会跳过所有工具权限审批（写文件、bash）。deny 规则仍会拦截，ask 问题和计划确认仍会等待你选择。仅运行时生效，重启后失效。",
   "settings.addRule": "添加 {list} 规则…",
   "settings.sandboxTitle": "沙箱与工作区",
+  "settings.sandboxBoundaryHint": "沙箱可写根只限制已批准的 Bash 和文件工具能写到哪里；它不会自动批准命令。mvn deploy 这类命令还需要权限规则或审批模式放行。",
+  "settings.reloadSessionConfig": "应用手动配置修改",
+  "settings.reloadSessionConfigHint": "设置页内修改会自动生效；此按钮用于应用手动编辑 config.toml / reasonix.toml 后的变更。",
   "settings.shellInterpreter": "Shell 解释器",
-  "settings.shellAuto": "自动（优先 bash，Windows 无 bash 时用 PowerShell）",
+  "settings.shellAuto": "自动（bash 优先）",
   "settings.shellBash": "bash",
   "settings.shellPowershell": "PowerShell（Windows PowerShell）",
   "settings.shellPwsh": "pwsh（PowerShell 7+）",
@@ -1235,6 +1412,9 @@ export const zh: Record<DictKey, string> = {
   "settings.allowNetwork": "允许沙箱内 bash 访问网络",
   "settings.workspaceRoot": "工作区根目录",
   "settings.workspaceDefault": "（默认：当前目录）",
+  "settings.effectiveWriteRoots": "实际可写根",
+  "settings.effectiveWriteRootsHint": "按当前工作区展开。设置页内修改会自动生效；手动编辑配置文件后可用上方按钮刷新当前会话。",
+  "settings.noEffectiveWriteRoots": "暂无可写根",
   "settings.appearance": "外观",
   "settings.theme": "主题",
   "settings.themeStyle": "视觉风格",
@@ -1342,6 +1522,7 @@ export const zh: Record<DictKey, string> = {
   "settings.pageDesc.bots": "配置 QQ、飞书、Lark 和微信机器人渠道，管理每个 Bot 的模型和运行参数。",
   "settings.pageDesc.mcp": "管理 MCP 服务器连接和工具发现。",
   "settings.pageDesc.skills": "浏览、启用和管理智能体技能与来源。",
+  "settings.pageDesc.plugins": "安装和管理插件包，用统一单元提供技能、hooks 与 MCP 服务器。",
   "settings.pageDesc.memory": "管理 Reasonix 会长期参考的偏好、项目约定和指令文件。",
   "settings.pageDesc.hooks": "配置围绕对话、工具调用、压缩和会话生命周期执行的 shell hooks。",
   "settings.pageDesc.shortcuts": "自定义键盘快捷键，并让帮助表保持同步。",
@@ -1356,7 +1537,7 @@ export const zh: Record<DictKey, string> = {
   "settings.customProviders": "自定义供应商",
   "settings.customProvidersHint": "你自己添加的 OpenAI-compatible 或代理接入，可以编辑、删除并持久化模型列表。",
   "settings.customProvidersEmpty": "还没有自定义供应商。常用服务优先使用上方内置接入。",
-  "settings.providerAccessEmptyHint": "点击“添加供应商”，选择 DeepSeek 或自定义 OpenAI-compatible 接入。",
+  "settings.providerAccessEmptyHint": "使用“推荐预设”可快速接入常用供应商，通常只需填写 API Key；使用“自定义供应商”可手动配置 OpenAI-compatible 或 Anthropic-compatible 端点。",
   "settings.builtinProviderBadge": "内置",
   "settings.customProviderBadge": "自定义",
   "settings.configureProvider": "配置",
@@ -1368,24 +1549,75 @@ export const zh: Record<DictKey, string> = {
   "settings.searchModels": "搜索模型...",
   "settings.noMatchingModels": "没有匹配的模型",
   "settings.addProvider.chooseTitle": "添加供应商",
-  "settings.addProvider.chooseHint": "选择官方预设，或添加一个自定义 OpenAI-compatible 接入。",
-  "settings.addProvider.officialChoice": "官方供应商",
-  "settings.addProvider.officialChoiceHint": "从 Reasonix 预设的 DeepSeek 接入中选择。",
+  "settings.addProvider.chooseHint": "选择推荐预设，或手动添加一个自定义供应商。",
+  "settings.addProvider.officialChoice": "推荐预设",
+  "settings.addProvider.officialChoiceHint": "选择 DeepSeek 或已适配的自定义供应商预设。",
   "settings.addProvider.customChoice": "自定义供应商",
   "settings.addProvider.customChoiceHint": "按 base_url、模型、密钥环境变量等参数手动配置。",
   "settings.addProvider.officialTitle": "添加官方供应商",
-  "settings.addProvider.officialHint": "官方预设会写入固定端点和默认参数，只需要选择供应商并配置密钥。",
+  "settings.addProvider.officialHint": "推荐预设会填好端点、模型、视觉和推理默认值，通常只需要填 API Key。",
   "settings.addProvider.customTitle": "添加自定义供应商",
   "settings.addProvider.customHint": "适合代理、聚合平台或其它 OpenAI-compatible 服务。",
   "settings.addProvider.official.deepseek": "DeepSeek 官方",
   "settings.addProvider.official.deepseekDesc": "官方 API Key 接入，可用 DeepSeek V4 Flash / Pro。",
+  "settings.addProvider.preset.kimiCnDesc": "Kimi 国内 API，走 Moonshot CN 端点，预置 K2 视觉模型。",
+  "settings.addProvider.preset.kimiGlobalDesc": "Kimi 国际 API，走 Moonshot global 端点，预置 K2 视觉模型。",
+  "settings.addProvider.preset.kimiCodingPlanDesc": "Kimi Coding Plan，走独立 Anthropic-compatible 端点，不归属 CN/Global。",
+  "settings.addProvider.preset.mimoApiDesc": "小米 MiMo 直连 API，适合 v2.5 按量付费模型。",
+  "settings.addProvider.preset.mimoAnthropicDesc": "小米 MiMo 直连 Anthropic-compatible 端点，自动使用 MiMo v2.5 模型。",
+  "settings.addProvider.preset.mimoTokenPlanCnDesc": "小米 MiMo token plan 中国区端点。",
+  "settings.addProvider.preset.mimoTokenPlanCnAnthropicDesc": "小米 MiMo token plan 中国区 Anthropic-compatible 端点。",
+  "settings.addProvider.preset.mimoTokenPlanSgpDesc": "小米 MiMo token plan 新加坡端点。",
+  "settings.addProvider.preset.mimoTokenPlanSgpAnthropicDesc": "小米 MiMo token plan 新加坡 Anthropic-compatible 端点。",
+  "settings.addProvider.preset.mimoTokenPlanAmsDesc": "小米 MiMo token plan 阿姆斯特丹端点。",
+  "settings.addProvider.preset.mimoTokenPlanAmsAnthropicDesc": "小米 MiMo token plan 阿姆斯特丹 Anthropic-compatible 端点。",
+  "settings.addProvider.preset.minimaxCnApiDesc": "MiniMax 国内 OpenAI-compatible M 系列 API 端点。",
+  "settings.addProvider.preset.minimaxGlobalApiDesc": "MiniMax 国际 OpenAI-compatible M 系列 API 端点。",
+  "settings.addProvider.preset.minimaxCnAnthropicDesc": "MiniMax 国内 Anthropic-compatible M 系列端点。",
+  "settings.addProvider.preset.minimaxGlobalAnthropicDesc": "MiniMax 国际 Anthropic-compatible 端点，自动使用 Bearer 认证。",
+  "settings.addProvider.preset.glmCnDesc": "智谱 GLM 国内 API，预置 GLM thinking 控制。",
+  "settings.addProvider.preset.zaiGlobalDesc": "Z.AI 国际 GLM API。",
+  "settings.addProvider.preset.glmCodingPlanCnDesc": "智谱 GLM 国内 coding plan 端点，默认 GLM-5.2。",
+  "settings.addProvider.preset.glmCodingPlanCnAnthropicDesc": "智谱 GLM 国内 Anthropic-compatible coding plan 端点，默认 GLM-5.2[1m]。",
+  "settings.addProvider.preset.zaiCodingPlanGlobalDesc": "Z.AI 国际 coding plan 端点，默认 GLM-5.2。",
+  "settings.addProvider.preset.zaiCodingPlanGlobalAnthropicDesc": "Z.AI 国际 Anthropic-compatible coding plan 端点，默认 GLM-5.2[1m]。",
+  "settings.addProvider.preset.opencodeGoDesc": "OpenCode Go 混合模型网关，内置每模型能力覆盖。",
+  "settings.addProvider.preset.opencodeGoAnthropicDesc": "OpenCode Go Anthropic 线路，预置 Qwen 与 MiniMax 模型。",
+  "settings.addProvider.preset.opencodeZenAnthropicDesc": "OpenCode Zen Anthropic 线路，预置 Claude 与 Qwen 模型。",
+  "settings.addProvider.preset.qwenCnDesc": "Qwen/DashScope 国内标准 API 端点。",
+  "settings.addProvider.preset.qwenGlobalDesc": "Qwen/DashScope 国际标准 API 端点。",
+  "settings.addProvider.preset.qwenCodingPlanCnDesc": "Qwen Coding Plan 国内 OpenAI-compatible 端点，默认 qwen3.7-plus。",
+  "settings.addProvider.preset.qwenCodingPlanCnAnthropicDesc": "Qwen Coding Plan 国内 Anthropic-compatible 端点，默认 qwen3.7-plus。",
+  "settings.addProvider.preset.qwenCodingPlanGlobalDesc": "Qwen Coding Plan 国际 OpenAI-compatible 端点，默认 qwen3.7-plus。",
+  "settings.addProvider.preset.qwenCodingPlanGlobalAnthropicDesc": "Qwen Coding Plan 国际 Anthropic-compatible 端点，默认 qwen3.7-plus。",
+  "settings.addProvider.preset.stepfunDesc": "阶跃星辰 StepFun coding plan OpenAI-compatible 端点，默认 step-3.7-flash。",
+  "settings.addProvider.preset.stepfunAnthropicDesc": "阶跃星辰 StepFun coding plan Anthropic-compatible 端点，默认 step-3.7-flash。",
+  "settings.addProvider.preset.novitaDesc": "NovitaAI 多模型网关，预置 Kimi、MiniMax、GLM、DeepSeek、Qwen。",
+  "settings.addProvider.preset.gmiDesc": "GMI Cloud 多模型直连网关，预置 GLM、DeepSeek、Kimi、Gemini、Claude、GPT。",
+  "settings.addProvider.preset.vercelAiGatewayDesc": "Vercel AI Gateway，走 Anthropic-compatible Messages API。",
+  "settings.addProvider.preset.huggingfaceDesc": "HuggingFace Inference Router，预置 Qwen 与 DeepSeek 模型。",
+  "settings.addProvider.preset.nvidiaDesc": "NVIDIA NIM 加速推理端点，预置 Nemotron 模型。",
+  "settings.addProvider.preset.kilocodeDesc": "Kilo Code 网关，预置 kilo/auto 自动路由模型。",
+  "settings.addProvider.preset.ollamaCloudDesc": "Ollama Cloud 托管端点，支持 none 到 max 的 reasoning effort。",
+  "settings.addProvider.addedBadge": "已添加",
+  "settings.addProvider.modifiedBadge": "已修改",
+  "settings.addProvider.nameConflictBadge": "同名占用",
+  "settings.addProvider.similarExistingBadge": "有相似配置",
   "settings.addProvider.confirm": "添加供应商",
+  "settings.addProvider.alreadyAddedAction": "已添加",
+  "settings.addProvider.nameConflictAction": "先处理同名配置",
+  "settings.addProvider.viewPresetProvider": "查看配置",
+  "settings.addProvider.viewConflictProvider": "查看同名配置",
+  "settings.addProvider.resetPreset": "重置为预设",
+  "settings.addProvider.confirmResetPreset": "确认重置",
   "settings.customProviderName": "自定义供应商名称",
   "settings.customProviderNamePlaceholder": "",
   "settings.providerNamePlaceholder": "供应商名称（如 deepseek-proxy）",
   "settings.providerProtocol": "接入协议",
   "settings.providerProtocolOpenAI": "OpenAI-compatible",
   "settings.providerProtocolOpenAIHint": "适用于兼容 OpenAI API 的代理、聚合平台和自建服务。",
+  "settings.providerProtocolAnthropic": "Anthropic-compatible",
+  "settings.providerProtocolAnthropicHint": "适用于 Claude Messages API 风格的服务，例如部分 coding plan 端点。",
   "settings.providerBaseUrlLabel": "API 地址",
   "settings.providerChatUrlLabel": "完整请求 URL",
   "settings.providerChatUrlPlaceholder": "例如 https://api.openai.com/v1/chat/completions",
@@ -1408,6 +1640,12 @@ export const zh: Record<DictKey, string> = {
   "settings.providerHeaders": "额外请求头（可选）",
   "settings.providerHeadersPlaceholder": "HTTP-Referer: https://example.com\nX-Title: Reasonix",
   "settings.providerHeadersHint": "用于 OpenRouter 等要求站点来源的中转站。一行一个 Header: value；API Key 仍通过上方密钥保存。",
+  "settings.providerExtraBody": "额外请求体（可选）",
+  "settings.providerExtraBodyPlaceholder": "{\n  \"enable_thinking\": true\n}",
+  "settings.providerExtraBodyHint": "会合并到聊天请求体顶层 JSON。model、messages、tools、stream、thinking 等核心字段仍由 Reasonix 控制。",
+  "settings.providerExtraBodyError": "请输入不含 null 值的 JSON 对象，例如 { \"enable_thinking\": true }。",
+  "settings.providerAuthHeader": "使用 Authorization: Bearer 发送密钥",
+  "settings.providerAuthHeaderHint": "适用于 MiniMax Global、Vercel AI Gateway 等不使用 x-api-key 的 Anthropic-compatible 网关。",
   "settings.fetchingModels": "刷新中…",
   "settings.fetchModelsSuccess": "已获取 {n} 个模型",
   "settings.fetchModelsManualFallbackAuth": "模型发现接口拒绝了密钥或权限。请检查 API Key；如果该服务不开放模型发现，也可以在下方手动填写模型 ID。",
@@ -1427,10 +1665,16 @@ export const zh: Record<DictKey, string> = {
   "settings.newProviderDraftName": "新供应商",
 	  "settings.reasoningProtocol": "模型能力模式",
 	  "settings.reasoningProtocol.auto": "自动识别（推荐）",
-	  "settings.reasoningProtocol.deepseek": "DeepSeek 思考",
-	  "settings.reasoningProtocol.openai": "OpenAI reasoning",
+	  "settings.reasoningProtocol.deepseek": "DeepSeek 思考（thinking + 深度）",
+	  "settings.reasoningProtocol.openai": "OpenAI reasoning（reasoning_effort）",
 	  "settings.reasoningProtocol.none": "普通聊天（不发送思考参数）",
 	  "settings.reasoningProtocolHint": "默认自动识别。只有代理不兼容，或你确定模型需要特定 reasoning 格式时再切换。",
+	  "settings.thinkingMode": "Thinking 覆盖",
+	  "settings.thinkingMode.auto": "Auto（使用服务默认）",
+	  "settings.thinkingMode.enabled": "Enabled（开启）",
+	  "settings.thinkingMode.disabled": "Disabled（关闭）",
+	  "settings.thinkingMode.adaptive": "Adaptive（自适应）",
+	  "settings.thinkingModeHint": "服务商专用覆盖项。兼容网关可用 enabled/disabled；只有服务文档明确支持时才使用 adaptive。",
   "settings.updateKey": "更新 {env}（全局保存）",
   "settings.updateKeyAction": "更新密钥",
   "settings.keyStatus": "密钥",
@@ -1545,6 +1789,8 @@ export const zh: Record<DictKey, string> = {
   "msg.imageAttachment": "图片",
   "msg.workspaceReference": "工作区引用",
   "msg.folderReference": "文件夹 · 工作区引用",
+  "msg.pastedExpandTooltip": "展开粘贴文本",
+  "msg.pastedCollapseTooltip": "收起粘贴文本",
   "turnActions.summary": "总结",
   "turnActions.rewind": "回溯",
   "transcript.jumpToBottom": "跳到底部",
@@ -1626,7 +1872,7 @@ export const zh: Record<DictKey, string> = {
   "settings.telemetryLabel": "匿名启动统计",
   "settings.telemetryHint": "启动时发送随机安装 ID、版本号和操作系统用于统计活跃安装量；若上次运行发生崩溃，则在本次启动补发该崩溃报告。绝不包含对话、密钥或文件内容。",
   "settings.metricsLabel": "共享聚合质量指标",
-  "settings.metricsHint": "默认开启。发送匿名的轮次结束统计、设置偏好快照，以及 Memory v5 聚合桶，例如是否注入、大小桶、IR 数量桶和记忆图规模桶。包含用于 DAU 去重的随机安装 ID。Bucket 可能包含归一化后的自定义 Provider 名和模型名——绝不包含对话、提示词、密钥、路径、base URL、记忆正文、工具输出或文件内容。",
+  "settings.metricsHint": "默认开启。发送匿名的轮次结束统计、更新器错误类别、设置偏好快照，以及 Memory v5 聚合桶，例如是否注入、大小桶、IR 数量桶和记忆图规模桶。包含用于 DAU 去重的随机安装 ID。Bucket 可能包含归一化后的自定义 Provider 名和模型名——绝不包含对话、提示词、密钥、路径、base URL、记忆正文、工具输出或文件内容。",
   "updater.currentVersion": "当前版本：{v}",
   "updater.channelLabel": "更新渠道：{channel}",
   "updater.checkButton": "检查更新",

desktop/go.mod

@@ -25,7 +25,9 @@ require (
 require (
 	git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3 // indirect
 	github.com/BurntSushi/toml v1.6.0 // indirect
+	github.com/aymanbagabas/go-udiff v0.4.1 // indirect
 	github.com/bep/debounce v1.2.1 // indirect
+	github.com/bmatcuk/doublestar/v4 v4.10.0 // indirect
 	github.com/clipperhouse/uax29/v2 v2.7.0 // indirect
 	github.com/danieljoos/wincred v1.2.3 // indirect
 	github.com/go-ole/go-ole v1.3.0 // indirect
@@ -33,6 +35,7 @@ require (
 	github.com/google/uuid v1.6.0 // indirect
 	github.com/gorilla/websocket v1.5.3 // indirect
 	github.com/jchv/go-winloader v0.0.0-20250406163304-c1995be93bd1 // indirect
+	github.com/joho/godotenv v1.5.1 // indirect
 	github.com/labstack/echo/v4 v4.15.2 // indirect
 	github.com/labstack/gommon v0.5.0 // indirect
 	github.com/larksuite/oapi-sdk-go/v3 v3.9.5 // indirect
@@ -42,6 +45,7 @@ require (
 	github.com/leaanthony/u v1.1.1 // indirect
 	github.com/mattn/go-colorable v0.1.15 // indirect
 	github.com/mattn/go-isatty v0.0.22 // indirect
+	github.com/mattn/go-pointer v0.0.1 // indirect
 	github.com/mattn/go-runewidth v0.0.24 // indirect
 	github.com/pkg/browser v0.0.0-20240102092130-5ac0b6a4141c // indirect
 	github.com/pkg/errors v0.9.1 // indirect
@@ -49,6 +53,11 @@ require (
 	github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06 // indirect
 	github.com/samber/lo v1.53.0 // indirect
 	github.com/tkrajina/go-reflector v0.5.8 // indirect
+	github.com/tree-sitter/go-tree-sitter v0.25.0 // indirect
+	github.com/tree-sitter/tree-sitter-javascript v0.25.0 // indirect
+	github.com/tree-sitter/tree-sitter-python v0.25.0 // indirect
+	github.com/tree-sitter/tree-sitter-rust v0.24.2 // indirect
+	github.com/tree-sitter/tree-sitter-typescript v0.23.2 // indirect
 	github.com/valyala/bytebufferpool v1.0.0 // indirect
 	github.com/valyala/fasttemplate v1.2.2 // indirect
 	github.com/wailsapp/go-webview2 v1.0.23 // indirect
@@ -57,6 +66,8 @@ require (
 	golang.org/x/crypto v0.53.0 // indirect
 	golang.org/x/image v0.43.0 // indirect
 	golang.org/x/net v0.56.0 // indirect
+	gopkg.in/yaml.v3 v3.0.1 // indirect
+	mvdan.cc/sh/v3 v3.13.1 // indirect
 )
 
 replace reasonix => ../

desktop/go.sum

@@ -7,8 +7,12 @@ git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3 h1:N3IGoHHp9pb6mj1cbXbuaSXV/UMKwmbKLf
 git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3/go.mod h1:QtOLZGz8olr4qH2vWK0QH0w0O4T9fEIjMuWpKUsH7nc=
 github.com/BurntSushi/toml v1.6.0 h1:dRaEfpa2VI55EwlIW72hMRHdWouJeRF7TPYhI+AUQjk=
 github.com/BurntSushi/toml v1.6.0/go.mod h1:ukJfTF/6rtPPRCnwkur4qwRxa8vTRFBF0uk2lLoLwho=
+github.com/aymanbagabas/go-udiff v0.4.1 h1:OEIrQ8maEeDBXQDoGCbbTTXYJMYRCRO1fnodZ12Gv5o=
+github.com/aymanbagabas/go-udiff v0.4.1/go.mod h1:0L9PGwj20lrtmEMeyw4WKJ/TMyDtvAoK9bf2u/mNo3w=
 github.com/bep/debounce v1.2.1 h1:v67fRdBA9UQu2NhLFXrSg0Brw7CexQekrBwDMM8bzeY=
 github.com/bep/debounce v1.2.1/go.mod h1:H8yggRPQKLUhUoqrJC1bO2xNya7vanpDl7xR3ISbCJ0=
+github.com/bmatcuk/doublestar/v4 v4.10.0 h1:zU9WiOla1YA122oLM6i4EXvGW62DvKZVxIe6TYWexEs=
+github.com/bmatcuk/doublestar/v4 v4.10.0/go.mod h1:xBQ8jztBU6kakFMg+8WGxn0c6z1fTSPVIjEY1Wr7jzc=
 github.com/clipperhouse/uax29/v2 v2.7.0 h1:+gs4oBZ2gPfVrKPthwbMzWZDaAFPGYK72F0NJv2v7Vk=
 github.com/clipperhouse/uax29/v2 v2.7.0/go.mod h1:EFJ2TJMRUaplDxHKj1qAEhCtQPW2tJSwu5BF98AuoVM=
 github.com/danieljoos/wincred v1.2.3 h1:v7dZC2x32Ut3nEfRH+vhoZGvN72+dQ/snVXo/vMFLdQ=
@@ -18,10 +22,14 @@ github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/go-ole/go-ole v1.3.0 h1:Dt6ye7+vXGIKZ7Xtk4s6/xVdGDQynvom7xCFEdWr6uE=
 github.com/go-ole/go-ole v1.3.0/go.mod h1:5LS6F96DhAwUc7C+1HLexzMXY1xGRSryjyPPKW6zv78=
+github.com/go-quicktest/qt v1.101.0 h1:O1K29Txy5P2OK0dGo59b7b0LR6wKfIhttaAhHUyn7eI=
+github.com/go-quicktest/qt v1.101.0/go.mod h1:14Bz/f7NwaXPtdYEgzsx46kqSxVwTbzVZsDC26tQJow=
 github.com/godbus/dbus/v5 v5.2.2 h1:TUR3TgtSVDmjiXOgAAyaZbYmIeP3DPkld3jgKGV8mXQ=
 github.com/godbus/dbus/v5 v5.2.2/go.mod h1:3AAv2+hPq5rdnr5txxxRwiGjPXamgoIHgz9FPBfOp3c=
 github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q=
 github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q=
+github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
+github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/gorilla/websocket v1.5.0/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
@@ -29,8 +37,14 @@ github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aN
 github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
 github.com/jchv/go-winloader v0.0.0-20250406163304-c1995be93bd1 h1:njuLRcjAuMKr7kI3D85AXWkw6/+v9PwtV6M6o11sWHQ=
 github.com/jchv/go-winloader v0.0.0-20250406163304-c1995be93bd1/go.mod h1:alcuEEnZsY1WQsagKhZDsoPCRoOijYqhZvPwLG0kzVs=
+github.com/joho/godotenv v1.5.1 h1:7eLL/+HRGLY0ldzfGMeQkb7vMd0as4CfYvUVzLqw0N0=
+github.com/joho/godotenv v1.5.1/go.mod h1:f4LDr5Voq0i2e/R5DDNOoa2zzDfwtkZa6DnEwAbqwq4=
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
+github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
+github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/labstack/echo/v4 v4.15.2 h1:nnh2sCzGCVYnU+wCisMPiYapEg/QVo/gcI9ePKg5/T4=
 github.com/labstack/echo/v4 v4.15.2/go.mod h1:Xzp1Ns1RA2c9fY7nSgUJkpkUZGNbEIVHZbtbOMPktBI=
 github.com/labstack/gommon v0.5.0 h1:6VSQ2NOzsnEJ5W6+84E0RbcaDDmgB6NIAzWCczTEe6c=
@@ -54,6 +68,8 @@ github.com/mattn/go-colorable v0.1.15 h1:+u9SLTRGnXv73cEsnsmoZBom+dMU88B2M0aDcWy
 github.com/mattn/go-colorable v0.1.15/go.mod h1:6LmQG8QLFO4G5z1gPvYEzlUgJ2wF+stgPZH1UqBm1s8=
 github.com/mattn/go-isatty v0.0.22 h1:j8l17JJ9i6VGPUFUYoTUKPSgKe/83EYU2zBC7YNKMw4=
 github.com/mattn/go-isatty v0.0.22/go.mod h1:ZXfXG4SQHsB/w3ZeOYbR0PrPwLy+n6xiMrJlRFqopa4=
+github.com/mattn/go-pointer v0.0.1 h1:n+XhsuGeVO6MEAp7xyEukFINEa+Quek5psIR/ylA6o0=
+github.com/mattn/go-pointer v0.0.1/go.mod h1:2zXcozF6qYGgmsG+SeTZz3oAbFLdD3OWqnUbNvJZAlc=
 github.com/mattn/go-runewidth v0.0.24 h1:cpokDiIn0MGnhdHwuWnJBITySJ20QyNGnY2kR/ay2DU=
 github.com/mattn/go-runewidth v0.0.24/go.mod h1:XBkDxAl56ILZc9knddidhrOlY5R/pDhgLpndooCuJAs=
 github.com/minio/selfupdate v0.6.0 h1:i76PgT0K5xO9+hjzKcacQtO7+MjJ4JKA8Ak8XQ9DDwU=
@@ -67,6 +83,8 @@ github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZN
 github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
 github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
 github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
+github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
+github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
 github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06 h1:OkMGxebDjyw0ULyrTYWeN0UNCCkmCWfjPnIA2W6oviI=
 github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06/go.mod h1:+ePHsJ1keEjQtpvf9HHw0f4ZeJ0TLRsxhunSI2hYJSs=
 github.com/samber/lo v1.53.0 h1:t975lj2py4kJPQ6haz1QMgtId2gtmfktACxIXArw3HM=
@@ -79,6 +97,34 @@ github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
 github.com/tkrajina/go-reflector v0.5.8 h1:yPADHrwmUbMq4RGEyaOUpz2H90sRsETNVpjzo3DLVQQ=
 github.com/tkrajina/go-reflector v0.5.8/go.mod h1:ECbqLgccecY5kPmPmXg1MrHW585yMcDkVl6IvJe64T4=
+github.com/tree-sitter/go-tree-sitter v0.25.0 h1:sx6kcg8raRFCvc9BnXglke6axya12krCJF5xJ2sftRU=
+github.com/tree-sitter/go-tree-sitter v0.25.0/go.mod h1:r77ig7BikoZhHrrsjAnv8RqGti5rtSyvDHPzgTPsUuU=
+github.com/tree-sitter/tree-sitter-c v0.23.4 h1:nBPH3FV07DzAD7p0GfNvXM+Y7pNIoPenQWBpvM++t4c=
+github.com/tree-sitter/tree-sitter-c v0.23.4/go.mod h1:MkI5dOiIpeN94LNjeCp8ljXN/953JCwAby4bClMr6bw=
+github.com/tree-sitter/tree-sitter-cpp v0.23.4 h1:LaWZsiqQKvR65yHgKmnaqA+uz6tlDJTJFCyFIeZU/8w=
+github.com/tree-sitter/tree-sitter-cpp v0.23.4/go.mod h1:doqNW64BriC7WBCQ1klf0KmJpdEvfxyXtoEybnBo6v8=
+github.com/tree-sitter/tree-sitter-embedded-template v0.23.2 h1:nFkkH6Sbe56EXLmZBqHHcamTpmz3TId97I16EnGy4rg=
+github.com/tree-sitter/tree-sitter-embedded-template v0.23.2/go.mod h1:HNPOhN0qF3hWluYLdxWs5WbzP/iE4aaRVPMsdxuzIaQ=
+github.com/tree-sitter/tree-sitter-go v0.23.4 h1:yt5KMGnTHS+86pJmLIAZMWxukr8W7Ae1STPvQUuNROA=
+github.com/tree-sitter/tree-sitter-go v0.23.4/go.mod h1:Jrx8QqYN0v7npv1fJRH1AznddllYiCMUChtVjxPK040=
+github.com/tree-sitter/tree-sitter-html v0.23.2 h1:1UYDV+Yd05GGRhVnTcbP58GkKLSHHZwVaN+lBZV11Lc=
+github.com/tree-sitter/tree-sitter-html v0.23.2/go.mod h1:gpUv/dG3Xl/eebqgeYeFMt+JLOY9cgFinb/Nw08a9og=
+github.com/tree-sitter/tree-sitter-java v0.23.5 h1:J9YeMGMwXYlKSP3K4Us8CitC6hjtMjqpeOf2GGo6tig=
+github.com/tree-sitter/tree-sitter-java v0.23.5/go.mod h1:NRKlI8+EznxA7t1Yt3xtraPk1Wzqh3GAIC46wxvc320=
+github.com/tree-sitter/tree-sitter-javascript v0.25.0 h1:ZkWETb66/w8cc13yhfnNuHOLDQWl3BnKlH6f9AdR88c=
+github.com/tree-sitter/tree-sitter-javascript v0.25.0/go.mod h1:lmGD1EJdCA+v0S1u2fFgepMg/opzSg/4pgFym2FPGAs=
+github.com/tree-sitter/tree-sitter-json v0.24.8 h1:tV5rMkihgtiOe14a9LHfDY5kzTl5GNUYe6carZBn0fQ=
+github.com/tree-sitter/tree-sitter-json v0.24.8/go.mod h1:F351KK0KGvCaYbZ5zxwx/gWWvZhIDl0eMtn+1r+gQbo=
+github.com/tree-sitter/tree-sitter-php v0.23.11 h1:iHewsLNDmznh8kgGyfWfujsZxIz1YGbSd2ZTEM0ZiP8=
+github.com/tree-sitter/tree-sitter-php v0.23.11/go.mod h1:T/kbfi+UcCywQfUNAJnGTN/fMSUjnwPXA8k4yoIks74=
+github.com/tree-sitter/tree-sitter-python v0.25.0 h1:O6XD9v8U1LOcRc3cNj9nM7XufrtEBezE6VrpRrHZDf0=
+github.com/tree-sitter/tree-sitter-python v0.25.0/go.mod h1:cpdthSy/Yoa28aJFBscFHlGiU+cnSiSh1kuDVtI8YeM=
+github.com/tree-sitter/tree-sitter-ruby v0.23.1 h1:T/NKHUA+iVbHM440hFx+lzVOzS4dV6z8Qw8ai+72bYo=
+github.com/tree-sitter/tree-sitter-ruby v0.23.1/go.mod h1:kUS4kCCQloFcdX6sdpr8p6r2rogbM6ZjTox5ZOQy8cA=
+github.com/tree-sitter/tree-sitter-rust v0.24.2 h1:NL4nF67ib21RMzzfvkmXlVwe45vvhW10DVyO+D0z/W0=
+github.com/tree-sitter/tree-sitter-rust v0.24.2/go.mod h1:hfeGWic9BAfgTrc7Xf6FaOAguCFJRo3RBbs7QJ6D7MI=
+github.com/tree-sitter/tree-sitter-typescript v0.23.2 h1:/Odvphn18PniVixb9e97X0DbNVsU6Qocv9mfkyzdXwU=
+github.com/tree-sitter/tree-sitter-typescript v0.23.2/go.mod h1:zjzMXT/Ulffel2xfOcAkQQkiAkmgnbtPGlFQw/5X4xA=
 github.com/valyala/bytebufferpool v1.0.0 h1:GqA5TC/0021Y/b9FG4Oi9Mr3q7XYx6KllzawFIhcdPw=
 github.com/valyala/bytebufferpool v1.0.0/go.mod h1:6bBcMArwyJ5K/AmCkWv1jt77kVWyCJ6HpOuEn7z0Csc=
 github.com/valyala/fasttemplate v1.2.2 h1:lxLXG0uE3Qnshl9QyaK6XJxMXlQZELvChBOCmQD0Loo=
@@ -146,7 +192,10 @@ golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8T
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+mvdan.cc/sh/v3 v3.13.1 h1:DP3TfgZhDkT7lerUdnp6PTGKyxxzz6T+cOlY/xEvfWk=
+mvdan.cc/sh/v3 v3.13.1/go.mod h1:lXJ8SexMvEVcHCoDvAGLZgFJ9Wsm2sulmoNEXGhYZD0=

desktop/history_test.go

@@ -88,6 +88,25 @@ func TestHistoryMessagesDoNotReplayMemoryCompilerContract(t *testing.T) {
 	assertNoHistoryMemoryContract(t, got[0].Content)
 }
 
+func TestHistoryMessagesStripActiveGoalFromVisibleUserContent(t *testing.T) {
+	raw := "<active-goal>\nship the approval redesign\n</active-goal>\n\ncontinue implementation"
+	msgs := []provider.Message{
+		{Role: provider.RoleUser, Content: raw},
+		{Role: provider.RoleAssistant, Content: "done"},
+	}
+
+	got := historyMessages(msgs, control.StripComposePrefixes)
+	if len(got) != 2 {
+		t.Fatalf("history length = %d, want 2: %+v", len(got), got)
+	}
+	if got[0].Content != "continue implementation" {
+		t.Fatalf("visible user content = %q, want active-goal stripped", got[0].Content)
+	}
+	if strings.Contains(got[0].Content, "<active-goal>") || strings.Contains(got[0].Content, "ship the approval redesign") {
+		t.Fatalf("active-goal leaked into visible history content: %+v", got[0])
+	}
+}
+
 func TestHistoryMessagesCarryCheckpointTurnsAcrossHiddenSyntheticUsers(t *testing.T) {
 	msgs := []provider.Message{
 		{Role: provider.RoleSystem, Content: "sys"},

desktop/notifications_test.go

@@ -0,0 +1,69 @@
+package main
+
+import (
+	"testing"
+
+	"reasonix/internal/config"
+	"reasonix/internal/event"
+	"reasonix/internal/notify"
+)
+
+type desktopRecordSink struct {
+	events []event.Kind
+}
+
+func (s *desktopRecordSink) Emit(e event.Event) {
+	s.events = append(s.events, e.Kind)
+}
+
+type desktopRecordSender struct {
+	messages []notify.Message
+}
+
+func (s *desktopRecordSender) Send(m notify.Message) error {
+	s.messages = append(s.messages, m)
+	return nil
+}
+
+func TestDesktopControllerSinkSendsConfiguredSystemNotifications(t *testing.T) {
+	inner := &desktopRecordSink{}
+	sender := &desktopRecordSender{}
+	app := &App{notificationSender: sender}
+	sink := app.desktopControllerSink(inner, config.NotificationsConfig{
+		Enabled:         true,
+		TurnDone:        true,
+		ApprovalRequest: true,
+		AskRequest:      true,
+	})
+
+	sink.Emit(event.Event{Kind: event.TurnDone})
+
+	if len(inner.events) != 1 || inner.events[0] != event.TurnDone {
+		t.Fatalf("forwarded events = %+v, want [TurnDone]", inner.events)
+	}
+	if len(sender.messages) != 1 {
+		t.Fatalf("notifications = %d, want 1", len(sender.messages))
+	}
+	if sender.messages[0].Body != "Turn finished" {
+		t.Fatalf("notification body = %q, want Turn finished", sender.messages[0].Body)
+	}
+}
+
+func TestDesktopControllerSinkSkipsSystemNotificationsWhenDisabled(t *testing.T) {
+	inner := &desktopRecordSink{}
+	sender := &desktopRecordSender{}
+	app := &App{notificationSender: sender}
+	sink := app.desktopControllerSink(inner, config.NotificationsConfig{
+		Enabled:  false,
+		TurnDone: true,
+	})
+
+	sink.Emit(event.Event{Kind: event.TurnDone})
+
+	if len(inner.events) != 1 || inner.events[0] != event.TurnDone {
+		t.Fatalf("forwarded events = %+v, want [TurnDone]", inner.events)
+	}
+	if len(sender.messages) != 0 {
+		t.Fatalf("notifications = %d, want 0", len(sender.messages))
+	}
+}

desktop/plugin_packages_app.go

@@ -0,0 +1,231 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"strings"
+
+	"reasonix/internal/config"
+	"reasonix/internal/installsource"
+	"reasonix/internal/pluginpkg"
+)
+
+type PluginView struct {
+	Name             string                `json:"name"`
+	Version          string                `json:"version,omitempty"`
+	Description      string                `json:"description,omitempty"`
+	Source           string                `json:"source,omitempty"`
+	Root             string                `json:"root"`
+	ManifestKind     string                `json:"manifestKind,omitempty"`
+	Enabled          bool                  `json:"enabled"`
+	Skills           int                   `json:"skills"`
+	Hooks            int                   `json:"hooks"`
+	MCPServers       int                   `json:"mcpServers"`
+	SkillDetails     []PluginSkillView     `json:"skillDetails,omitempty"`
+	HookDetails      []PluginHookView      `json:"hookDetails,omitempty"`
+	MCPServerDetails []PluginMCPServerView `json:"mcpServerDetails,omitempty"`
+	Warnings         []string              `json:"warnings,omitempty"`
+	Error            string                `json:"error,omitempty"`
+}
+
+type PluginInstallOptions struct {
+	DryRun  bool   `json:"dryRun,omitempty"`
+	Link    bool   `json:"link,omitempty"`
+	Replace bool   `json:"replace,omitempty"`
+	Name    string `json:"name,omitempty"`
+}
+
+type PluginSkillView struct {
+	Name        string `json:"name"`
+	Description string `json:"description,omitempty"`
+	Path        string `json:"path,omitempty"`
+	Invocation  string `json:"invocation,omitempty"`
+	RunAs       string `json:"runAs,omitempty"`
+}
+
+type PluginHookView struct {
+	Event       string `json:"event"`
+	Match       string `json:"match,omitempty"`
+	Command     string `json:"command,omitempty"`
+	ContextFile string `json:"contextFile,omitempty"`
+	Description string `json:"description,omitempty"`
+}
+
+type PluginMCPServerView struct {
+	Name      string `json:"name"`
+	Transport string `json:"transport,omitempty"`
+	Command   string `json:"command,omitempty"`
+	URL       string `json:"url,omitempty"`
+}
+
+func (a *App) Plugins() []PluginView {
+	st, err := pluginpkg.LoadState(config.ReasonixHomeDir())
+	if err != nil {
+		return []PluginView{{Error: err.Error()}}
+	}
+	out := make([]PluginView, 0, len(st.Plugins))
+	for _, p := range st.Plugins {
+		view := PluginView{
+			Name:         p.Name,
+			Version:      p.Version,
+			Description:  p.Description,
+			Source:       p.Source,
+			Root:         pluginpkg.ResolveRoot(config.ReasonixHomeDir(), p.Root),
+			ManifestKind: p.ManifestKind,
+			Enabled:      p.Enabled,
+		}
+		if pkg, warnings, err := pluginpkg.ParseDir(view.Root); err == nil {
+			applyPluginPackageDetails(&view, pkg, warnings)
+		} else {
+			view.Error = err.Error()
+		}
+		out = append(out, view)
+	}
+	return out
+}
+
+func applyPluginPackageDetails(view *PluginView, pkg pluginpkg.Package, warnings []string) {
+	view.Skills, view.Hooks, view.MCPServers = pkg.CapabilityCounts()
+	view.Warnings = warnings
+	inv := pkg.Inventory()
+	view.SkillDetails = make([]PluginSkillView, 0, len(inv.Skills))
+	for _, sk := range inv.Skills {
+		view.SkillDetails = append(view.SkillDetails, PluginSkillView{
+			Name:        sk.Name,
+			Description: sk.Description,
+			Path:        sk.Path,
+			Invocation:  sk.Invocation,
+			RunAs:       sk.RunAs,
+		})
+	}
+	view.HookDetails = make([]PluginHookView, 0, len(inv.Hooks))
+	for _, hook := range inv.Hooks {
+		view.HookDetails = append(view.HookDetails, PluginHookView{
+			Event:       hook.Event,
+			Match:       hook.Match,
+			Command:     hook.Command,
+			ContextFile: hook.ContextFile,
+			Description: hook.Description,
+		})
+	}
+	view.MCPServerDetails = make([]PluginMCPServerView, 0, len(inv.MCPServers))
+	for _, server := range inv.MCPServers {
+		view.MCPServerDetails = append(view.MCPServerDetails, PluginMCPServerView{
+			Name:      server.Name,
+			Transport: server.Transport,
+			Command:   server.Command,
+			URL:       server.URL,
+		})
+	}
+}
+
+func (a *App) PlanPluginInstall(source string, opts PluginInstallOptions) (string, error) {
+	opts.DryRun = true
+	return a.runPluginInstallSource(source, opts, false)
+}
+
+func (a *App) InstallPlugin(source string, opts PluginInstallOptions) (string, error) {
+	if err := a.ensureActiveTabRebuildAllowed("plugins"); err != nil {
+		return "", err
+	}
+	out, err := a.runPluginInstallSource(source, opts, true)
+	if err != nil {
+		return "", err
+	}
+	a.invalidateSkillRootsCache()
+	if rebuildErr := a.rebuild(); rebuildErr != nil {
+		return out, rebuildErr
+	}
+	return out, nil
+}
+
+func (a *App) RemovePlugin(name string) error {
+	if err := a.ensureActiveTabRebuildAllowed("plugins"); err != nil {
+		return err
+	}
+	raw, _ := json.Marshal(map[string]any{"op": "uninstall", "kind": "plugin", "name": strings.TrimSpace(name), "scope": "global"})
+	tl := installsource.NewTool(installsource.Options{
+		ProjectRoot: a.activeWorkspaceRoot(),
+		OnDisconnect: func(serverName string) bool {
+			tab := a.activeTab()
+			if tab == nil || tab.Ctrl == nil {
+				return false
+			}
+			removed, _ := tab.Ctrl.RemoveMCPServer(serverName)
+			return removed
+		},
+	})
+	if _, err := tl.Execute(context.Background(), raw); err != nil {
+		return err
+	}
+	a.invalidateSkillRootsCache()
+	return a.rebuild()
+}
+
+func (a *App) SetPluginEnabled(name string, enabled bool) error {
+	if err := a.ensureActiveTabRebuildAllowed("plugins"); err != nil {
+		return err
+	}
+	if err := pluginpkg.SetEnabled(config.ReasonixHomeDir(), strings.TrimSpace(name), enabled); err != nil {
+		return err
+	}
+	a.invalidateSkillRootsCache()
+	return a.rebuild()
+}
+
+func (a *App) UpdatePlugin(name string) (string, error) {
+	name = strings.TrimSpace(name)
+	for _, p := range a.Plugins() {
+		if p.Name == name {
+			if strings.TrimSpace(p.Source) == "" {
+				return "", fmt.Errorf("plugin %q has no recorded source", name)
+			}
+			return a.InstallPlugin(p.Source, PluginInstallOptions{Name: name, Replace: true})
+		}
+	}
+	return "", fmt.Errorf("plugin %q is not installed", name)
+}
+
+func (a *App) PluginDoctor(name string) PluginView {
+	name = strings.TrimSpace(name)
+	for _, p := range a.Plugins() {
+		if p.Name != name {
+			continue
+		}
+		if p.Error != "" {
+			return p
+		}
+		if p.Root == "" {
+			p.Error = "missing plugin root"
+			return p
+		}
+		if _, err := os.Stat(p.Root); err != nil {
+			p.Error = err.Error()
+			return p
+		}
+		return p
+	}
+	return PluginView{Name: name, Error: "plugin is not installed"}
+}
+
+func (a *App) runPluginInstallSource(source string, opts PluginInstallOptions, apply bool) (string, error) {
+	mode := "copy"
+	if opts.Link {
+		mode = "link"
+	}
+	body := map[string]any{
+		"source":  strings.TrimSpace(source),
+		"kind":    "plugin",
+		"mode":    mode,
+		"replace": opts.Replace,
+		"apply":   apply && !opts.DryRun,
+	}
+	if strings.TrimSpace(opts.Name) != "" {
+		body["name"] = strings.TrimSpace(opts.Name)
+	}
+	raw, _ := json.Marshal(body)
+	tl := installsource.NewTool(installsource.Options{ProjectRoot: a.activeWorkspaceRoot()})
+	return tl.Execute(context.Background(), raw)
+}

desktop/settings_app_test.go

@@ -97,6 +97,16 @@ func TestProviderViewFromEntry_MigratesProviderWideVision(t *testing.T) {
 	}
 }
 
+func TestProviderViewFromEntryIncludesThinking(t *testing.T) {
+	view := providerViewFromEntry(config.ProviderEntry{
+		Name:     "anthropic",
+		Thinking: "ADAPTIVE",
+	}, false, true)
+	if view.Thinking != "adaptive" {
+		t.Fatalf("ProviderView.Thinking = %q, want adaptive", view.Thinking)
+	}
+}
+
 func TestProviderViewFromEntryShowsKeySource(t *testing.T) {
 	isolateDesktopUserDirs(t)
 	t.Setenv("TEST_PROVIDER_KEY_SOURCE", "")
@@ -120,6 +130,43 @@ func TestProviderViewFromEntryShowsKeySource(t *testing.T) {
 	}
 }
 
+func TestSettingsExposesEffectiveSandboxWriteRoots(t *testing.T) {
+	home := isolateDesktopUserDirs(t)
+	project := robustTempDir(t)
+	cfg := config.LoadForEdit(config.UserConfigPath())
+	cfg.Sandbox.AllowWrite = []string{
+		"${HOME}/.m2",
+		"${HOME}/.m2/repository",
+	}
+	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
+		t.Fatalf("save config: %v", err)
+	}
+
+	app := NewApp()
+	app.tabs = map[string]*WorkspaceTab{
+		"project": {ID: "project", Scope: "project", WorkspaceRoot: project, Ready: true},
+	}
+	app.activeTabID = "project"
+
+	got := app.Settings().Sandbox
+	if got.EffectiveWorkspaceRoot != project {
+		t.Fatalf("EffectiveWorkspaceRoot = %q, want %q", got.EffectiveWorkspaceRoot, project)
+	}
+	// Settings expose expanded configured roots; the writer confiner normalizes
+	// separators later when enforcing them.
+	want := []string{
+		project,
+		home + "/.m2",
+		home + "/.m2/repository",
+	}
+	if !reflect.DeepEqual(got.EffectiveWriteRoots, want) {
+		t.Fatalf("EffectiveWriteRoots = %v, want %v", got.EffectiveWriteRoots, want)
+	}
+	if !reflect.DeepEqual(got.AllowWrite, cfg.Sandbox.AllowWrite) {
+		t.Fatalf("AllowWrite = %v, want raw configured paths %v", got.AllowWrite, cfg.Sandbox.AllowWrite)
+	}
+}
+
 func TestProviderViewFromEntryExposesNoAuthAvailability(t *testing.T) {
 	isolateDesktopUserDirs(t)
 	t.Setenv("LOCAL_API_KEY", "")
@@ -382,6 +429,60 @@ func TestSaveProviderFiltersNonChatModels(t *testing.T) {
 	}
 }
 
+func TestSaveProviderPersistsThinkingOverride(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	app := NewApp()
+	if err := app.SaveProvider(ProviderView{
+		Name:      "glm-proxy",
+		Kind:      "openai",
+		BaseURL:   "https://proxy.example.com/v1",
+		Models:    []string{"glm-4.5-air"},
+		APIKeyEnv: "GLM_PROXY_API_KEY",
+		Thinking:  "DISABLED",
+	}); err != nil {
+		t.Fatalf("SaveProvider: %v", err)
+	}
+
+	cfg := config.LoadForEdit(config.UserConfigPath())
+	got, ok := cfg.Provider("glm-proxy")
+	if !ok {
+		t.Fatal("saved provider not found")
+	}
+	if got.Thinking != "disabled" {
+		t.Fatalf("saved provider thinking = %q, want disabled", got.Thinking)
+	}
+}
+
+func TestSaveProviderPersistsAuthHeader(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	app := NewApp()
+	if err := app.SaveProvider(ProviderView{
+		Name:       "minimax-global-anthropic",
+		Kind:       "anthropic",
+		BaseURL:    "https://api.minimax.io/anthropic",
+		Models:     []string{"MiniMax-M3"},
+		APIKeyEnv:  "MINIMAX_API_KEY",
+		AuthHeader: true,
+	}); err != nil {
+		t.Fatalf("SaveProvider: %v", err)
+	}
+
+	cfg := config.LoadForEdit(config.UserConfigPath())
+	got, ok := cfg.Provider("minimax-global-anthropic")
+	if !ok {
+		t.Fatal("saved provider not found")
+	}
+	if !got.AuthHeader {
+		t.Fatal("saved provider auth_header = false, want true")
+	}
+	view := providerViewFromEntry(*got, false, true)
+	if !view.AuthHeader {
+		t.Fatal("provider view authHeader = false, want true")
+	}
+}
+
 func TestSaveProviderPersistsCustomEndpointURLs(t *testing.T) {
 	isolateDesktopUserDirs(t)
 
@@ -443,6 +544,7 @@ func TestSaveProviderPreservesHiddenProviderFields(t *testing.T) {
 		Thinking:     "adaptive",
 		Effort:       "high",
 		VisionDetail: "low",
+		ExtraBody:    map[string]any{"enable_thinking": true},
 		NoProxy:      true,
 	}}
 	if err := cfg.SaveTo(config.UserConfigPath()); err != nil {
@@ -463,6 +565,9 @@ func TestSaveProviderPreservesHiddenProviderFields(t *testing.T) {
 	if !found {
 		t.Fatalf("Settings providers missing custom: %+v", settings.Providers)
 	}
+	if view.ExtraBody["enable_thinking"] != true {
+		t.Fatalf("settings extra_body = %+v, want enable_thinking=true", view.ExtraBody)
+	}
 
 	if err := app.SaveProvider(view); err != nil {
 		t.Fatalf("SaveProvider: %v", err)
@@ -485,6 +590,9 @@ func TestSaveProviderPreservesHiddenProviderFields(t *testing.T) {
 	if got.VisionDetail != "low" {
 		t.Fatalf("vision_detail = %q, want low", got.VisionDetail)
 	}
+	if got.ExtraBody["enable_thinking"] != true {
+		t.Fatalf("extra_body = %+v, want enable_thinking=true", got.ExtraBody)
+	}
 	if !got.NoProxy {
 		t.Fatal("no_proxy = false, want preserved true")
 	}
@@ -721,7 +829,7 @@ func TestSetReasoningLanguageUpdatesLiveTabControllers(t *testing.T) {
 	}
 
 	userComposed := userCtrl.Compose("hi")
-	if !strings.Contains(userComposed, "Simplified Chinese") {
+	if !strings.Contains(userComposed, "简体中文") {
 		t.Fatalf("user-level tab Compose = %q, want zh reasoning language", userComposed)
 	}
 	projectComposed := projectCtrl.Compose("hi")

desktop/tab_profile_test.go

@@ -122,7 +122,7 @@ api_key_env = "PROJECT_API_KEY"
 	if !got.Supported || got.Current != "auto" || got.Default != "high" {
 		t.Fatalf("EffortForTab model registry = %+v, want supported auto/high", got)
 	}
-	wantLevels := []string{"auto", "high", "max"}
+	wantLevels := []string{"auto", "disabled", "high", "max"}
 	if len(got.Levels) != len(wantLevels) {
 		t.Fatalf("levels = %v, want %v", got.Levels, wantLevels)
 	}

desktop/tabs_order_test.go

@@ -9,6 +9,7 @@ import (
 	"testing"
 	"time"
 
+	"reasonix/internal/agent"
 	"reasonix/internal/config"
 	"reasonix/internal/control"
 	"reasonix/internal/event"
@@ -679,6 +680,42 @@ func TestBuildTabControllerReusesOpenSessionPathRuntime(t *testing.T) {
 	}
 }
 
+func TestBuildTabControllerBlocksWhenSessionLeaseHeld(t *testing.T) {
+	isolateDesktopUserDirs(t)
+	dir := desktopSessionDir(globalTabWorkspaceRoot())
+	if err := os.MkdirAll(dir, 0o755); err != nil {
+		t.Fatalf("mkdir sessions: %v", err)
+	}
+	path := filepath.Join(dir, "leased.jsonl")
+	if err := os.WriteFile(path, nil, 0o644); err != nil {
+		t.Fatalf("write placeholder session: %v", err)
+	}
+	lease, err := agent.TryAcquireSessionLease(path)
+	if err != nil {
+		t.Fatalf("TryAcquireSessionLease: %v", err)
+	}
+	defer lease.Release()
+
+	app := NewApp()
+	tab := app.createTabEntryWithID("global", globalTabWorkspaceRoot(), "", "leased")
+	tab.SessionPath = path
+	tab.sink = &tabEventSink{tabID: "leased", app: app}
+	app.tabs[tab.ID] = tab
+	app.tabOrder = []string{tab.ID}
+	app.activeTabID = tab.ID
+
+	app.buildTabController(tab)
+	if tab.Ctrl != nil {
+		t.Fatalf("tab controller = %T, want nil when lease is held", tab.Ctrl)
+	}
+	if !tab.Ready {
+		t.Fatal("tab should be ready with startup error")
+	}
+	if !strings.Contains(tab.StartupErr, agent.ErrSessionLeaseHeld.Error()) {
+		t.Fatalf("startup error = %q, want lease-held error", tab.StartupErr)
+	}
+}
+
 func TestOpenGlobalTabResolvesTopicToLatestSessionRuntime(t *testing.T) {
 	isolateDesktopUserDirs(t)
 	dir := desktopSessionDir(globalTabWorkspaceRoot())

desktop/tabs_topic_test.go

@@ -1567,6 +1567,53 @@ func TestAutoTitleTopicStripsReasoningLanguagePrefix(t *testing.T) {
 	}
 }
 
+func TestAutoTitleTopicRefreshesOnThirdUserTurn(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectRoot := t.TempDir()
+	topic, err := NewApp().CreateTopic("project", projectRoot, "")
+	if err != nil {
+		t.Fatalf("create topic: %v", err)
+	}
+	sessionPath := filepath.Join(t.TempDir(), "session.jsonl")
+	firstTurn := strings.Join([]string{
+		`{"role":"user","content":"帮我看看"}`,
+		`{"role":"assistant","content":"可以"}`,
+	}, "\n") + "\n"
+	if err := os.WriteFile(sessionPath, []byte(firstTurn), 0o644); err != nil {
+		t.Fatalf("write first session: %v", err)
+	}
+
+	title, updated := autoTitleTopicFromSession(projectRoot, topic.ID, sessionPath)
+	if !updated || title != "帮我看看" {
+		t.Fatalf("first auto title = %q updated=%v, want 帮我看看/true", title, updated)
+	}
+
+	thirdTurn := strings.Join([]string{
+		`{"role":"user","content":"帮我看看"}`,
+		`{"role":"assistant","content":"可以"}`,
+		`{"role":"user","content":"继续"}`,
+		`{"role":"assistant","content":"继续分析"}`,
+		`{"role":"user","content":"实现自动更新会话标题"}`,
+		`{"role":"assistant","content":"已实现"}`,
+	}, "\n") + "\n"
+	if err := os.WriteFile(sessionPath, []byte(thirdTurn), 0o644); err != nil {
+		t.Fatalf("write third session: %v", err)
+	}
+
+	title, updated = autoTitleTopicFromSession(projectRoot, topic.ID, sessionPath)
+	if !updated || title != "实现自动更新会话标题" {
+		t.Fatalf("third-turn auto title = %q updated=%v, want 实现自动更新会话标题/true", title, updated)
+	}
+	if got := loadTopicTitle(projectRoot, topic.ID); got != "实现自动更新会话标题" {
+		t.Fatalf("stored title = %q, want 实现自动更新会话标题", got)
+	}
+	meta := loadTopicAutoTitleMeta(projectRoot)[topic.ID]
+	if meta.Stage != 3 {
+		t.Fatalf("auto title stage = %d, want 3", meta.Stage)
+	}
+}
+
 func TestAutoTitleDoesNotOverrideManualTopicTitle(t *testing.T) {
 	isolateDesktopUserDirs(t)
 
@@ -1592,6 +1639,50 @@ func TestAutoTitleDoesNotOverrideManualTopicTitle(t *testing.T) {
 	}
 }
 
+func TestAutoTitleDoesNotOverrideManualSessionTitle(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectRoot := t.TempDir()
+	topic, err := NewApp().CreateTopic("project", projectRoot, "")
+	if err != nil {
+		t.Fatalf("create topic: %v", err)
+	}
+	sessionPath := filepath.Join(t.TempDir(), "session.jsonl")
+	if err := os.WriteFile(sessionPath, []byte(`{"role":"user","content":"讲讲这个代码库的架构"}`+"\n"), 0o644); err != nil {
+		t.Fatalf("write session: %v", err)
+	}
+	if err := agent.SaveBranchMetaPreserveUpdated(sessionPath, agent.BranchMeta{CustomTitle: "手动会话标题"}); err != nil {
+		t.Fatalf("save branch meta: %v", err)
+	}
+
+	if title, updated := autoTitleTopicFromSession(projectRoot, topic.ID, sessionPath); updated || title != "" {
+		t.Fatalf("manual session title should not auto-update, title=%q updated=%v", title, updated)
+	}
+	if got := loadTopicTitle(projectRoot, topic.ID); got != defaultTopicTitle {
+		t.Fatalf("stored title = %q, want default title", got)
+	}
+}
+
+func TestRenameTopicBlankKeepsManualTitleSource(t *testing.T) {
+	isolateDesktopUserDirs(t)
+
+	projectRoot := t.TempDir()
+	app := NewApp()
+	topic, err := app.CreateTopic("project", projectRoot, "")
+	if err != nil {
+		t.Fatalf("create topic: %v", err)
+	}
+	if err := app.RenameTopic(topic.ID, "   "); err != nil {
+		t.Fatalf("rename blank topic: %v", err)
+	}
+	if got := loadTopicTitle(projectRoot, topic.ID); got != defaultTopicTitle {
+		t.Fatalf("stored title = %q, want %q", got, defaultTopicTitle)
+	}
+	if got := loadTopicTitleSource(projectRoot, topic.ID); got != topicTitleSourceManual {
+		t.Fatalf("title source = %q, want manual", got)
+	}
+}
+
 func TestTrashTopicMovesRelatedSessionsToTrash(t *testing.T) {
 	isolateDesktopUserDirs(t)
 

desktop/updater.go

@@ -33,36 +33,38 @@ import (
 // has no Wails dependency so the logic is unit-tested directly; updater_app.go is
 // the thin Wails binding that wires these into App methods and progress events.
 
-// Manifest endpoints — R2 CDN first (fast, especially in CN), GitHub releases as
-// fallback. The build channel picks the rolling pointer so a canary build polls
-// the canary line and a stable build polls latest; the two never cross.
+// Manifest endpoints — R2 CDN first (fast, especially in CN), then the crash
+// worker release gateway. The build channel picks the rolling pointer so a
+// canary build polls the canary line and a stable build polls latest; the two
+// never cross. The gateway deliberately avoids GitHub's repository-wide
+// /releases/latest shortcut because CLI tags (v*) and desktop tags (desktop-v*)
+// are separate release lines in the same repo.
 const (
-	r2Base         = "https://dl.reasonix.io"
-	ghReleasesBase = "https://github.com/esengine/reasonix/releases"
-	httpTimeout    = 15 * time.Second
+	r2Base             = "https://dl.reasonix.io"
+	releaseGatewayBase = "https://crash.reasonix.io/v1/desktop/releases"
+	downloadPageURL    = "https://reasonix.io/#start"
+	httpTimeout        = 15 * time.Second
 )
 
 // manifestEndpoints returns the primary (R2) then fallback (GitHub) manifest URLs
 // for the running build's channel.
 func manifestEndpoints() []string {
 	if channel == "canary" {
-		// Canary publishes only to R2 (no GitHub release), so there is no
-		// GitHub fallback for this channel.
-		return []string{r2Base + "/canary/latest.json"}
+		return []string{
+			r2Base + "/canary/latest.json",
+			releaseGatewayBase + "/canary/latest.json",
+		}
 	}
 	return []string{
 		r2Base + "/latest/latest.json",
-		ghReleasesBase + "/latest/download/latest.json",
+		releaseGatewayBase + "/stable/latest.json",
 	}
 }
 
 // downloadPage is the human-facing releases page shown when self-update is
 // unavailable (macOS) or the manifest omits its own link.
 func downloadPage() string {
-	if channel == "canary" {
-		return ghReleasesBase // lists pre-releases too
-	}
-	return ghReleasesBase + "/latest"
+	return downloadPageURL
 }
 
 // UpdateInfo is the CheckUpdate result that drives the frontend's update banner.

desktop/updater_app.go

@@ -27,6 +27,7 @@ func (a *App) Version() string { return version }
 func (a *App) CheckUpdate() (*UpdateInfo, error) {
 	c, err := httpClient()
 	if err != nil {
+		a.recordUpdateError(err)
 		return &UpdateInfo{
 			Current:       version,
 			Channel:       channel,
@@ -41,6 +42,7 @@ func (a *App) CheckUpdate() (*UpdateInfo, error) {
 	defer cancel()
 	m, err := fetchManifest(ctx, c)
 	if err != nil {
+		a.recordUpdateError(err)
 		return &UpdateInfo{
 			Current:       version,
 			Channel:       channel,
@@ -55,7 +57,7 @@ func (a *App) CheckUpdate() (*UpdateInfo, error) {
 	return &info, nil
 }
 
-// OpenDownloadPage opens the releases page in the browser — the macOS manual-update
+// OpenDownloadPage opens the install page in the browser — the macOS manual-update
 // path and a fallback link elsewhere.
 func (a *App) OpenDownloadPage() {
 	page := downloadPage()
@@ -208,6 +210,16 @@ func (a *App) emitProgress(phase string, received, total int64, errMsg string) {
 
 // failUpdate emits an error progress event and returns the error to the caller.
 func (a *App) failUpdate(err error) error {
+	a.recordUpdateError(err)
 	a.emitProgress("error", 0, 0, err.Error())
 	return err
 }
+
+func (a *App) recordUpdateError(err error) {
+	if err == nil || version == "dev" {
+		return
+	}
+	if m := a.metrics.Load(); m != nil {
+		m.inc("updater_error", errorClass(err.Error()))
+	}
+}

desktop/updater_test.go

@@ -98,6 +98,14 @@ func TestChannelSelectsDistinctPointers(t *testing.T) {
 	if !strings.Contains(stable[0], "/latest/latest.json") {
 		t.Errorf("stable primary = %q, want the latest/ pointer", stable[0])
 	}
+	if stable[1] != releaseGatewayBase+"/stable/latest.json" {
+		t.Errorf("stable fallback = %q, want the release gateway", stable[1])
+	}
+	for _, u := range append(stable, canary...) {
+		if strings.Contains(u, "/releases/latest") {
+			t.Errorf("manifest endpoint uses GitHub's repository-wide latest release: %q", u)
+		}
+	}
 	for _, u := range canary {
 		if strings.Contains(u, "/latest/") {
 			t.Errorf("canary endpoint hits the stable latest/ pointer: %q", u)
@@ -106,8 +114,11 @@ func TestChannelSelectsDistinctPointers(t *testing.T) {
 	if !strings.Contains(canary[0], "/canary/latest.json") {
 		t.Errorf("canary primary = %q, want the canary/ pointer", canary[0])
 	}
-	if downloadPage() == (ghReleasesBase + "/latest") {
-		t.Error("canary download page should not be the stable latest releases page")
+	if canary[1] != releaseGatewayBase+"/canary/latest.json" {
+		t.Errorf("canary fallback = %q, want the release gateway", canary[1])
+	}
+	if strings.Contains(downloadPage(), "/releases/latest") {
+		t.Errorf("download page should not use GitHub's repository-wide latest release: %q", downloadPage())
 	}
 }
 

docs/BOT_GUIDE.md

@@ -6,13 +6,14 @@
  · 
 <a href="./GUIDE.md">General guide</a>
 
-> For desktop users. This guide explains how to connect Feishu, Lark, WeChat, and QQ
-> bots, how to use Reasonix from IM, and how approvals, Ask questions, YOLO, and
-> bot commands work.
+> For desktop and CLI users. This guide explains how to connect Feishu, Lark,
+> WeChat, and QQ bots, how to use Reasonix from IM, and how approvals, Ask
+> questions, YOLO, and bot commands work.
 
 ## Contents
 
 - [What the bot does](#what-the-bot-does)
+- [Where it runs](#where-it-runs)
 - [Connect the four channels](#connect-the-four-channels)
 - [Run the bot headlessly](#run-the-bot-headlessly)
 - [Usage flow](#usage-flow)
@@ -24,9 +25,10 @@
 
 ## What the bot does
 
-After a bot is connected, you can send Reasonix messages from Feishu, Lark, or
-WeChat. The desktop app handles the model, tools, permissions, sandboxing, and
-local context, then sends progress and results back to the IM channel.
+After a bot is connected, you can send Reasonix messages from Feishu, Lark,
+WeChat, or QQ. The desktop app or `reasonix bot start` process handles the
+model, tools, permissions, sandboxing, and local context, then sends progress
+and results back to the IM channel.
 
 Common uses:
 
@@ -37,6 +39,26 @@ Common uses:
 - Open the matching desktop IM session to inspect context, cost, tokens, and tool
   traces.
 
+## Where it runs
+
+The bot gateway is a shared Go runtime. The same core behavior works on
+Windows, macOS, and Linux; platform differences mostly come from each IM
+provider's credentials, network reachability, callback/WebSocket setup, and
+saved local account state.
+
+There are two supported entry points:
+
+- **Desktop runtime**: configure bots in **Settings -> Bots**. The desktop app
+  starts the gateway, keeps status in the app, persists per-connection tool
+  approval mode changes, and lets you open matching local IM sessions.
+- **CLI runtime**: run `reasonix bot start` for a headless long-lived process.
+  It uses the same config, allowlist, routes, queue settings, pairing store,
+  adapters, and project/session index as the desktop runtime.
+
+The normal `reasonix run` command does not automatically start the IM gateway.
+Remote bot behavior is active only while the desktop bot runtime is running or
+while a `reasonix bot start` process is alive.
+
 ## Connect the four channels
 
 Open the Reasonix desktop app and go to **Settings -> Bots**. In **Add IM Bot**,
@@ -77,7 +99,14 @@ flowchart LR
 
 Feishu and Lark share the same capability set, but they are saved as separate
 connections. You can give them different models, working directories, or tool
-approval modes.
+approval modes. Bot text replies are sent as standalone Interactive Card JSON
+2.0 markdown, which avoids Feishu/Lark platform quote prefixes while preserving
+CommonMark formatting. If a card is too large for the platform limit, Reasonix
+falls back to plain text automatically.
+
+For webhook mode, configure a verification token. Incoming webhook events are
+verified fail-closed: an empty or missing configured token rejects callers
+instead of silently opening the webhook.
 
 ### WeChat
 
@@ -87,8 +116,9 @@ approval modes.
 4. Wait until the page shows the connection as connected.
 5. Send the WeChat bot a message.
 
-WeChat does not provide interactive card buttons here, so approvals and Ask
-questions are handled through text commands.
+WeChat does not provide interactive card buttons here, so approvals use numeric
+or text commands. Ask questions can be answered by replying with normal text,
+option numbers, or `/answer <id> <answer>`.
 
 ### QQ
 
@@ -99,13 +129,16 @@ questions are handled through text commands.
 5. Send the QQ bot a message.
 
 QQ Bot uses the official QQ Bot platform API. It supports inline keyboard
-buttons for approvals. Ask questions are sent as text; for single-choice
-questions you can reply with the option number, or use `/answer <id> <option>`.
-When a button expires or the platform reports an action failure, copy the ID
-shown in the card and send the equivalent text command.
+buttons for approvals. Ask questions are sent as text; reply with normal text,
+option numbers, or `/answer <id> <answer>`. When a button expires or the
+platform reports an action failure, copy the ID shown in the card and send the
+equivalent text command.
 
-QQ does not support QR-code scanning for connection setup. You must
-configure the App ID and App Secret manually.
+QQ does not support QR-code scanning for connection setup. You must configure
+the App ID and App Secret manually. The adapter reads only the configured
+`app_secret_env` value; it does not fall back to an unrelated `QQ_SECRET`
+environment variable. QQ and WeChat HTTP calls use bounded clients so a stalled
+provider request cannot block the gateway indefinitely.
 
 ## Run the bot headlessly
 
@@ -114,7 +147,8 @@ runtime itself can also run as a long-lived headless gateway:
 
 ```sh
 reasonix bot doctor
-reasonix bot start --channels feishu,lark,weixin --dir /path/to/project
+reasonix bot doctor --deep
+reasonix bot start --channels qq,feishu,lark,weixin --dir /path/to/project
 ```
 
 Use `--channels` to choose which configured IM inputs to accept. `feishu` and
@@ -135,17 +169,96 @@ The headless gateway uses the same config records as the desktop app:
 - `workspace_root`, `model`, and `tool_approval_mode` can be set per
   connection. This lets different IM channels route to different local projects
   or approval postures.
+- `access` can also be set per connection with `enabled`, `allow_all`,
+  `pairing_enabled`, `users`, `groups`, `admins`, and `approvers`. When a
+  connection has active access settings, they are checked before the legacy
+  global `[bot.allowlist]`.
+- `[[bot.routes]]` adds finer routing by connection, platform, chat type, chat
+  ID, user ID, or thread ID. Empty match fields are wildcards; the first matching
+  route wins and can override `workspace_root`, `model`, and
+  `tool_approval_mode`.
 - `session_mappings` are filled from inbound messages with the remote chat ID
   and scope. The desktop UI can open the matching conversation once the mapping
   also has a local `session_id` target, such as a saved `path:` session target
   from a desktop-managed bot runtime or a manually configured mapping.
+- The bot's project/session index is intentionally bounded to configured
+  `workspace_root` values, route workspaces, active bot sessions, and saved
+  `session_mappings`. Commands such as `/use project` and `/attach session`
+  can only jump to those indexed targets; arbitrary local directories are not
+  accepted from IM text.
 
-Access control is still mandatory. Either enable an allowlist under
-`[bot.allowlist]` with at least one relevant platform user ID, or set
-`allow_all = true` deliberately. Group IDs are optional additional scoping for
-group chats; they do not replace the required user allowlist. Remote users go
-through the same controller, permission policy, tool approval mode, and sandbox
-rules as local desktop or CLI turns.
+Access control is still mandatory. New desktop-created bots should normally set
+access inside that bot's own detail panel, which saves to `[[bot.connections]]`
+or `[bot.qq].access`. The legacy global `[bot.allowlist]` remains a fallback for
+older configs and for connections without active per-bot access. You can
+deliberately set `allow_all = true`, or enable `pairing_enabled` for a single
+bot / `[bot.pairing]` globally so an unknown DM sender receives a one-time
+pairing code. That code must be approved locally with
+`reasonix bot pairing approve <code>` before the sender can drive the bot; when
+the request is tied to a connection, approval adds the sender to that
+connection's access list. Users listed in `admins` / `approvers` or the legacy
+`*_admins` / `*_approvers` also receive base bot admission, so they do not need
+to be duplicated in `users` / `*_users`. Group chats are not opened by DM
+pairing or role admission; group IDs remain an additional narrowing layer.
+Use these commands to manage pending requests:
+
+```sh
+reasonix bot pairing list
+reasonix bot pairing approve CODE
+reasonix bot pairing reject CODE
+```
+
+If `qq_admins`, `feishu_admins`, `weixin_admins`, or the matching
+`*_approvers` lists are configured, `/yolo` and `/mode` are admin-only while
+`/projects`, `/use project`, `/sessions`, `/attach session`, and `/search all`
+are also admin-only. `/approve` and `/deny` require an approver or admin. When
+no role lists are set, existing allowlisted users keep the previous command
+behavior for compatibility. Remote users go through the same controller,
+permission policy, tool approval mode, and sandbox rules as local desktop or CLI
+turns.
+
+```toml
+[bot.allowlist]
+enabled = true
+feishu_users = ["ou_member"]
+feishu_admins = ["ou_admin"]
+feishu_approvers = ["ou_approver"]
+```
+
+`ignore_self_messages = true` is enabled by default. The gateway remembers the
+platform `message_id` values it just sent and ignores matching echo events. If a
+platform does not echo the same message ID reliably, configure the bot's own user
+IDs under `[bot.self_user_ids]` as a second layer of loop protection. `/status`
+also includes the current queue mode and adapter health, such as
+`feishu-lark=running` or `weixin-weixin=degraded`.
+
+The optional `[bot.control]` section exposes a local loopback HTTP API and is
+disabled by default. When enabled, `token_env` must point to an environment
+variable and every request must include `Authorization: Bearer <token>`. The
+server only binds to `localhost`, `127.0.0.1`, or `::1`. Current endpoints are
+`GET /status` for session and adapter health snapshots, `GET /metrics` for
+Prometheus text metrics, and `POST /send` for sending text or media through a
+configured connection.
+
+Example:
+
+```sh
+export REASONIX_BOT_CONTROL_TOKEN="change-me"
+
+curl -H "Authorization: Bearer $REASONIX_BOT_CONTROL_TOKEN" \
+  http://127.0.0.1:37913/status
+
+curl -X POST http://127.0.0.1:37913/send \
+  -H "Authorization: Bearer $REASONIX_BOT_CONTROL_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "connection_id": "feishu-lark",
+    "domain": "lark",
+    "chat_id": "oc_xxx",
+    "chat_type": "dm",
+    "text": "hello from local control API"
+  }'
+```
 
 ## Usage flow
 
@@ -153,7 +266,7 @@ rules as local desktop or CLI turns.
 sequenceDiagram
   participant U as "User"
   participant IM as "Feishu / Lark / WeChat / QQ"
-  participant R as "Reasonix desktop"
+  participant R as "Reasonix desktop or bot start"
   participant T as "Local tools and model"
 
   U->>IM: "Send a request"
@@ -195,8 +308,8 @@ without exposing real account IDs, local paths, or private chat content.
 | --- | --- | --- | --- | --- |
 | Feishu | Scan to create a PersonalAgent | Interactive card buttons, or commands | Interactive card buttons, or commands | Feishu workspaces, DMs, and groups |
 | Lark | Scan to create a PersonalAgent | Interactive card buttons, or commands | Interactive card buttons, or commands | International Lark workspaces |
-| WeChat | Scan with WeChat | Reply `1` / `2`, or commands | Single-choice questions can use a number, or commands | Lightweight personal/mobile testing |
-| QQ | Manual setup (App ID + App Secret) | Inline keyboard buttons, numeric replies, or commands | Single-choice questions can use a number, or commands | QQ groups, DMs, and official QQ Bot platform |
+| WeChat | Scan with WeChat | Reply `1` / `2`, or commands | Reply with normal text, option numbers, or commands | Lightweight personal/mobile testing |
+| QQ | Manual setup (App ID + App Secret) | Inline keyboard buttons, numeric replies, or commands | Reply with normal text, option numbers, or commands | QQ groups, DMs, and official QQ Bot platform |
 
 Feishu and Lark card buttons are converted into commands such as
 `/approve <id>`, `/deny <id>`, or `/answer <id> <option>`. QQ approval buttons
@@ -210,7 +323,7 @@ These commands work in Feishu, Lark, WeChat, and QQ.
 | Command | Purpose | Example |
 | --- | --- | --- |
 | `/help` | Show available commands | `/help` |
-| `/status` | Show active tasks, retained sessions, and tool approval mode | `/status` |
+| `/status` | Show active tasks, queue state, tool approval mode, and adapter health | `/status` |
 | `/stop` | Stop the current task | `/stop` |
 | `/new` | Start a fresh session | `/new` |
 | `/reset` | Reset the current session | `/reset` |
@@ -225,14 +338,63 @@ These commands work in Feishu, Lark, WeChat, and QQ.
 | `/mode yolo` | Switch to YOLO | `/mode yolo` |
 | `/mode ask` | Switch to Ask mode | `/mode ask` |
 | `/mode auto` | Switch to Auto mode | `/mode auto` |
+| `/queue status` | Show the current queue mode | `/queue status` |
+| `/queue steer` | Treat mid-run messages as guidance for the current task | `/queue steer` |
+| `/queue followup` | Queue mid-run messages as later turns | `/queue followup` |
+| `/queue collect` | Merge queued messages into one later turn | `/queue collect` |
+| `/queue interrupt` | Cancel the current task and keep the newest message | `/queue interrupt` |
+| `/projects [query]` | List indexed project workspaces | `/projects reasonix` |
+| `/use project <id\|name>` | Route this remote session to an indexed project | `/use project p1` |
+| `/use project default` | Clear the project override and return to configured routing | `/use project default` |
+| `/sessions search <query>` | Search indexed desktop/bot sessions | `/sessions search release bug` |
+| `/attach session <id\|query>` | Continue this remote session from an indexed `path:` transcript | `/attach session s1` |
+| `/search all <query>` | Search file contents across indexed project roots | `/search all TODO` |
 
 Shortcut replies:
 
 - When an approval is pending, reply `1` to approve and `2` to deny.
-- When a single-choice Ask question is pending, reply with the option number.
+- When an Ask question is pending, reply with any normal non-slash text. Option
+  numbers still work for choice questions.
+- Slash commands such as `/stop`, `/mode`, or `/answer ...` are not captured as
+  Ask shortcut replies.
 - If there is no pending operation, `1` / `2` are treated as normal text or
   produce guidance.
 
+The default queue mode is `steer`: when the same session is already running, a
+new message is injected as mid-turn guidance instead of waiting for the whole
+turn to finish. `queue_cap` and `queue_drop` bound backlog growth in config.
+`reasonix bot doctor --deep` reports queue, pairing, and role diagnostics.
+
+Queue modes:
+
+- `steer`: mid-run messages become guidance for the current turn when possible.
+- `followup`: mid-run messages are queued as later turns.
+- `collect`: queued messages are merged into one later turn.
+- `interrupt`: the active turn is canceled and the newest message is kept as the
+  next turn.
+
+Project and session navigation:
+
+- `/projects [query]` lists workspaces from configured bot routes, connection
+  workspaces, active bot sessions, and saved session mappings.
+- `/use project <id|name>` pins the current remote session to one indexed
+  project. `/use project default` clears the override.
+- `/sessions search <query>` searches indexed desktop and bot session metadata.
+- `/attach session <id|query>` continues the remote session from an indexed
+  `path:` transcript.
+- `/search all <query>` searches file contents across indexed project roots.
+  Reasonix uses `rg` when available and falls back to a bounded Go scanner.
+
+These navigation commands never accept arbitrary paths typed from IM. They only
+jump to indexed targets and, when role lists are configured, require an admin.
+
+When an adapter supplies media URLs, the gateway downloads those files into the
+current workspace's `.reasonix/attachments` directory and passes them to
+Reasonix as `@.reasonix/attachments/...` references. If an attachment cannot be
+saved, the bot sends a short warning and continues with the available text. The
+built-in Feishu, Weixin, and QQ adapters currently focus on text events; ordinary
+IM attachment extraction can be added at the adapter layer.
+
 ## Approvals and YOLO
 
 Reasonix bots use the same permission system as the desktop app. Ask mode is the
@@ -299,11 +461,11 @@ You may need to bind again if:
 | Symptom | What to check |
 | --- | --- |
 | QR code says the link expired | Generate a new QR code in Settings; QR codes expire (Feishu, Lark, WeChat only — QQ uses manual setup and has no QR code). |
-| Connected but no reply | Make sure the Reasonix desktop app is running, the bot connection is enabled, and the sender ID is allowlisted or access is open. |
+| Connected but no reply | Make sure the desktop bot runtime or `reasonix bot start` process is running, the bot connection is enabled, and the sender ID is allowlisted, paired, or access is open. |
 | Feishu or Lark button action fails | Send the text command from the card, such as `/approve <id>` or `/deny <id>`. |
 | QQ button action fails | Same as Feishu/Lark — send the text command from the card, such as `/approve <id>` or `/deny <id>`. |
-| WeChat reply `1` does nothing | Numeric shortcuts only work when an approval or single-choice Ask is pending; use the full command if needed. |
-| QQ reply `1` does nothing | Same as WeChat — numeric shortcuts only work when an approval or single-choice Ask is pending; use the full command if needed. |
+| WeChat reply `1` does nothing | Numeric shortcuts only work when an approval or Ask is pending; use the full command if needed. |
+| QQ reply `1` does nothing | Same as WeChat — numeric shortcuts only work when an approval or Ask is pending; use the full command if needed. |
 | Need to confirm the current mode | Send `/status` or `/yolo status`. |
 | Need a fresh context | Send `/new` or `/reset`. |
 | Need to stop the current task | Send `/stop`. |

docs/BOT_GUIDE.zh-CN.md

@@ -6,12 +6,13 @@
  · 
 <a href="./GUIDE.zh-CN.md">通用指南</a>
 
-> 面向桌面端用户。本文说明如何连接飞书、Lark、微信和 QQ 机器人，如何在 IM
-> 里使用 Reasonix，以及审批、问答、YOLO 和常用命令的交互方式。
+> 面向桌面端和 CLI 用户。本文说明如何连接飞书、Lark、微信和 QQ 机器人，
+> 如何在 IM 里使用 Reasonix，以及审批、问答、YOLO 和常用命令的交互方式。
 
 ## 目录
 
 - [能做什么](#能做什么)
+- [在哪里运行](#在哪里运行)
 - [连接四个渠道](#连接四个渠道)
 - [无界面运行 Bot](#无界面运行-bot)
 - [使用流程](#使用流程)
@@ -23,8 +24,9 @@
 
 ## 能做什么
 
-连接机器人后，你可以在飞书、Lark 或微信里给 Reasonix 发消息，让桌面端
-Reasonix 在本机执行同一套模型、工具、权限与沙盒逻辑。
+连接机器人后，你可以在飞书、Lark、微信或 QQ 里给 Reasonix 发消息，让桌面端
+Reasonix 或 `reasonix bot start` 进程在本机执行同一套模型、工具、权限与
+沙盒逻辑。
 
 典型场景：
 
@@ -34,6 +36,24 @@ Reasonix 在本机执行同一套模型、工具、权限与沙盒逻辑。
 - 对临时测试任务开启 YOLO，跳过普通工具审批。
 - 打开桌面端对应 IM 会话，继续查看上下文、成本、tokens 和工具轨迹。
 
+## 在哪里运行
+
+Bot gateway 是一套共享的 Go runtime。核心行为在 Windows、macOS 和 Linux
+上都生效；实际差异主要来自各 IM 平台的凭据、网络、回调或 WebSocket 配置，
+以及本机保存的账号状态。
+
+目前有两个入口：
+
+- **桌面端 runtime**：在 **设置 -> 机器人** 中配置。桌面端会启动 gateway，
+  在应用内维护状态，持久化每个连接的工具审批模式变化，并允许打开匹配的
+  本地 IM 会话。
+- **CLI runtime**：执行 `reasonix bot start` 启动无界面长期进程。它复用
+  与桌面端相同的配置、白名单、路由、队列设置、配对存储、适配器和
+  项目/会话索引。
+
+普通 `reasonix run` 不会自动启动 IM 网关。只有桌面端 bot runtime 正在运行，
+或存在一个存活的 `reasonix bot start` 进程时，远端 IM bot 能力才会生效。
+
 ## 连接四个渠道
 
 打开桌面端 Reasonix，进入 **设置 -> 机器人**。在 **添加 IM Bot** 区域选择
@@ -73,7 +93,12 @@ flowchart LR
 5. 给 Lark Bot 发送消息。
 
 飞书和 Lark 使用同一套能力，但作为两个独立连接保存。你可以给它们设置不同
-模型、工作目录或工具审批模式。
+模型、工作目录或工具审批模式。Bot 文本回复会以独立 Interactive Card JSON
+2.0 markdown 发送，避免飞书/Lark 平台级引用前缀，同时保留 CommonMark
+格式；如果卡片超过平台限制，Reasonix 会自动降级为纯文本。
+
+Webhook 模式需要配置 verification token。传入事件会 fail-closed 校验：如果
+配置中的 token 为空或缺失，调用方会被拒绝，不会静默开放 webhook。
 
 ### 微信
 
@@ -83,7 +108,8 @@ flowchart LR
 4. 等待页面显示已连接。
 5. 给微信 Bot 发送消息。
 
-微信没有交互卡片按钮，因此审批和问答主要通过文字命令完成。
+微信没有交互卡片按钮，因此审批通过数字或文字命令完成。Ask 问题可以直接
+回复普通文本、选项编号，或使用 `/answer <id> <回答>`。
 
 ### QQ
 
@@ -94,11 +120,13 @@ flowchart LR
 5. 给 QQ Bot 发送消息。
 
 QQ Bot 使用官方 QQ Bot 平台 API。它支持内联键盘按钮来完成审批。
-Ask 问答会以文字形式发送；单选题可以直接回复选项编号，也可以用
-`/answer <id> <选项>`。当按钮过期或平台提示操作失败时，可以直接
-复制卡片里的 ID，用文字命令继续。
+Ask 问答会以文字形式发送；可以直接回复普通文本、选项编号，或使用
+`/answer <id> <回答>`。当按钮过期或平台提示操作失败时，可以直接复制
+卡片里的 ID，用文字命令继续。
 
-QQ 不支持扫码连接，必须手动配置 App ID 和 App Secret。
+QQ 不支持扫码连接，必须手动配置 App ID 和 App Secret。适配器只读取配置的
+`app_secret_env`，不会再回退读取无关的 `QQ_SECRET` 环境变量。QQ 和微信的
+HTTP 调用使用带超时的 client，避免平台请求卡住后无限阻塞 gateway。
 
 ## 无界面运行 Bot
 
@@ -107,7 +135,8 @@ QQ 不支持扫码连接，必须手动配置 App ID 和 App Secret。
 
 ```sh
 reasonix bot doctor
-reasonix bot start --channels feishu,lark,weixin --dir /path/to/project
+reasonix bot doctor --deep
+reasonix bot start --channels qq,feishu,lark,weixin --dir /path/to/project
 ```
 
 `--channels` 用来选择接受哪些已配置的 IM 输入。`feishu` 和 `lark` 会选择对应
@@ -124,15 +153,88 @@ Bot。`--dir` 用来把远端消息绑定到某个项目工作区，`--model`
   密钥仍保存在环境变量或 Reasonix 用户凭据中。
 - `workspace_root`、`model` 和 `tool_approval_mode` 可以按连接单独设置，
   因此不同 IM 渠道可以路由到不同本地项目或审批模式。
+- `access` 也可以按连接单独设置，包括 `enabled`、`allow_all`、
+  `pairing_enabled`、`users`、`groups`、`admins` 和 `approvers`。当某个连接
+  有启用中的 access 配置时，会优先检查该连接自己的访问控制，再退回旧的全局
+  `[bot.allowlist]`。
+- `[[bot.routes]]` 可以继续按远端连接、平台、会话类型、chat ID、用户 ID
+  或 thread ID 做更细粒度路由；空匹配字段表示通配，按配置顺序第一个命中。
+  命中后可覆盖 `workspace_root`、`model` 和 `tool_approval_mode`。
 - `session_mappings` 会根据收到的远端消息自动填充远端 chat ID 和作用域。
   只有当该映射同时具备本地 `session_id` 目标时，桌面端才能打开对应会话；
   例如桌面端托管的 Bot runtime 保存了 `path:` 会话目标，或用户手动配置了
   映射目标。
+- Bot 的项目/会话索引只来自已配置的 `workspace_root`、route workspace、
+  当前活跃 bot 会话，以及已保存的 `session_mappings`。`/use project` 和
+  `/attach session` 只能跳到这些已索引目标；IM 文本里临时输入的任意本地
+  目录不会被接受。
 
-访问控制仍然是必需项。你需要在 `[bot.allowlist]` 下为对应平台至少配置一个
-用户 ID，或者有意设置 `allow_all = true`。群 ID 是群聊里的额外收窄条件，
-不能替代用户白名单。远端用户进入的是同一个 Reasonix controller、权限策略、
-工具审批模式和沙盒边界，和本地桌面端或 CLI 回合一致。
+访问控制仍然是必需项。桌面端新建的 Bot 通常建议在该 Bot 自己的详情面板里
+设置谁可以使用，这会保存到 `[[bot.connections]]` 或 `[bot.qq].access`。
+旧的全局 `[bot.allowlist]` 仍然作为兼容兜底，适用于旧配置或没有启用单
+Bot access 的连接。你可以有意设置 `allow_all = true`，也可以为单个 Bot
+启用 `pairing_enabled`，或全局启用 `[bot.pairing]`，让未知私聊用户先收到
+一次性配对码。配对码需要在本机执行 `reasonix bot pairing approve <code>`
+后才会放行；如果该请求带有连接 ID，批准后会把发送者加入对应 Bot 自己的
+access 名单。列在 `admins` / `approvers` 或旧的 `*_admins` /
+`*_approvers` 里的用户也会获得基础 bot 准入，不需要再重复写进 `users` /
+`*_users`。群聊不会因为私聊配对或角色准入自动开放，群 ID 仍是额外收窄
+条件。常用配对管理命令：
+
+```sh
+reasonix bot pairing list
+reasonix bot pairing approve CODE
+reasonix bot pairing reject CODE
+```
+
+如果配置了 `qq_admins`、`feishu_admins`、`weixin_admins` 或对应
+`*_approvers`，`/yolo`、`/mode` 等运行模式命令只允许 admin 使用；
+`/projects`、`/use project`、`/sessions`、`/attach session` 和
+`/search all` 也只允许 admin 使用。`/approve` 和 `/deny` 只允许 approver
+或 admin 使用。没有配置角色列表时，为兼容旧配置，已允许的用户保持原有
+命令能力。远端用户进入的是同一个 Reasonix controller、权限策略、工具
+审批模式和沙盒边界，和本地桌面端或 CLI 回合一致。
+
+```toml
+[bot.allowlist]
+enabled = true
+feishu_users = ["ou_member"]
+feishu_admins = ["ou_admin"]
+feishu_approvers = ["ou_approver"]
+```
+
+默认启用 `ignore_self_messages = true`。网关会记录刚发出的平台
+`message_id`，并忽略平台回传的同 ID 消息；如果某个平台不会稳定回传同一
+消息 ID，可以在 `[bot.self_user_ids]` 里配置 bot 自己的用户 ID 作为第二层
+回声防护。`/status` 会显示当前会话队列模式和各连接的健康状态，例如
+`feishu-lark=running` 或 `weixin-weixin=degraded`。
+
+可选的 `[bot.control]` 提供本机 loopback HTTP API，默认关闭。启用后必须
+设置 `token_env` 对应的环境变量，所有请求都需要
+`Authorization: Bearer <token>`；地址只能绑定 `localhost`、`127.0.0.1`
+或 `::1`。当前接口包括 `GET /status`（会话与连接健康快照）和
+`GET /metrics`（Prometheus 文本指标）以及 `POST /send`（向指定连接发送
+文本或媒体消息）。
+
+示例：
+
+```sh
+export REASONIX_BOT_CONTROL_TOKEN="change-me"
+
+curl -H "Authorization: Bearer $REASONIX_BOT_CONTROL_TOKEN" \
+  http://127.0.0.1:37913/status
+
+curl -X POST http://127.0.0.1:37913/send \
+  -H "Authorization: Bearer $REASONIX_BOT_CONTROL_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "connection_id": "feishu-lark",
+    "domain": "lark",
+    "chat_id": "oc_xxx",
+    "chat_type": "dm",
+    "text": "hello from local control API"
+  }'
+```
 
 ## 使用流程
 
@@ -140,7 +242,7 @@ Bot。`--dir` 用来把远端消息绑定到某个项目工作区，`--model`
 sequenceDiagram
   participant U as "用户"
   participant IM as "飞书 / Lark / 微信 / QQ"
-  participant R as "Reasonix 桌面端"
+  participant R as "Reasonix 桌面端或 bot start"
   participant T as "本机工具与模型"
 
   U->>IM: "发送需求"
@@ -180,8 +282,8 @@ sequenceDiagram
 | --- | --- | --- | --- | --- |
 | 飞书 | 扫码创建 PersonalAgent | 交互卡片按钮，也可用命令 | 交互卡片按钮，也可用命令 | 国内飞书工作流、群聊或个人助手 |
 | Lark | 扫码创建 PersonalAgent | 交互卡片按钮，也可用命令 | 交互卡片按钮，也可用命令 | 国际版 Lark 工作流 |
-| 微信 | 微信扫码登录 | 回复 `1` / `2` 或命令 | 单选可直接回复编号，也可用命令 | 微信个人测试、轻量移动触发 |
-| QQ | 手动配置（App ID + App Secret） | 内联键盘按钮、数字回复或命令 | 单选可直接回复编号，也可用命令 | QQ 群聊、个人会话和官方 QQ Bot 平台 |
+| 微信 | 微信扫码登录 | 回复 `1` / `2` 或命令 | 回复普通文本、选项编号或命令 | 微信个人测试、轻量移动触发 |
+| QQ | 手动配置（App ID + App Secret） | 内联键盘按钮、数字回复或命令 | 回复普通文本、选项编号或命令 | QQ 群聊、个人会话和官方 QQ Bot 平台 |
 
 飞书和 Lark 的卡片按钮会在后台转换为命令，例如 `/approve <id>`、
 `/deny <id>` 或 `/answer <id> <选项>`。QQ 的审批按钮也是如此。
@@ -195,7 +297,7 @@ sequenceDiagram
 | 命令 | 作用 | 示例 |
 | --- | --- | --- |
 | `/help` | 查看可用命令 | `/help` |
-| `/status` | 查看活跃任务、保留会话数和工具审批模式 | `/status` |
+| `/status` | 查看活跃任务、队列、工具审批模式和连接健康 | `/status` |
 | `/stop` | 停止当前任务 | `/stop` |
 | `/new` | 开始新会话 | `/new` |
 | `/reset` | 重置当前会话 | `/reset` |
@@ -210,13 +312,59 @@ sequenceDiagram
 | `/mode yolo` | 切换到 YOLO | `/mode yolo` |
 | `/mode ask` | 切换到询问模式 | `/mode ask` |
 | `/mode auto` | 切换到自动模式 | `/mode auto` |
+| `/queue status` | 查看当前队列模式 | `/queue status` |
+| `/queue steer` | 运行中消息作为当前任务补充 | `/queue steer` |
+| `/queue followup` | 运行中消息排队为后续回合 | `/queue followup` |
+| `/queue collect` | 合并排队消息为一个后续回合 | `/queue collect` |
+| `/queue interrupt` | 取消当前任务并处理最新消息 | `/queue interrupt` |
+| `/projects [关键词]` | 查看已索引项目工作区 | `/projects reasonix` |
+| `/use project <id\|名称>` | 将当前远端会话路由到已索引项目 | `/use project p1` |
+| `/use project default` | 清除项目覆盖，恢复配置路由 | `/use project default` |
+| `/sessions search <关键词>` | 搜索已索引桌面/bot 会话 | `/sessions search 发布 bug` |
+| `/attach session <id\|关键词>` | 从已索引 `path:` transcript 继续当前远端会话 | `/attach session s1` |
+| `/search all <关键词>` | 跨已索引项目检索文件内容 | `/search all TODO` |
 
 快捷回复：
 
 - 有待审批操作时，回复 `1` 表示批准，回复 `2` 表示拒绝。
-- 有单选 Ask 问题时，可以直接回复选项编号。
+- 有待处理 Ask 问题时，可以直接回复任意普通非 slash 文本；选择题仍然可以
+  直接回复选项编号。
+- `/stop`、`/mode`、`/answer ...` 等 slash 命令不会被 Ask 快捷回复截获。
 - 如果没有待处理操作，`1` / `2` 会被当作普通消息或收到提示。
 
+默认队列模式是 `steer`：同一会话正在运行时，新消息会作为当前任务的
+mid-turn guidance 注入，而不是等完整回合结束。`queue_cap` 和 `queue_drop`
+可以在配置里限制排队堆积；`reasonix bot doctor --deep` 会显示当前队列、
+配对和角色诊断信息。
+
+队列模式：
+
+- `steer`：运行中的消息会尽量作为当前回合的补充指导。
+- `followup`：运行中的消息排队成为后续回合。
+- `collect`：把排队消息合并成一个后续回合。
+- `interrupt`：取消当前回合，并保留最新消息作为下一回合。
+
+项目与会话跳转：
+
+- `/projects [关键词]` 会列出来自 bot route、连接工作区、活跃 bot 会话和
+  已保存 session mapping 的工作区。
+- `/use project <id|名称>` 会把当前远端会话固定到某个已索引项目；
+  `/use project default` 会清除覆盖。
+- `/sessions search <关键词>` 搜索已索引桌面端和 bot 会话元数据。
+- `/attach session <id|关键词>` 从已索引 `path:` transcript 继续当前远端
+  会话。
+- `/search all <关键词>` 跨已索引项目根目录检索文件内容；有 `rg` 时优先用
+  `rg`，否则使用带边界的 Go fallback scanner。
+
+这些跳转命令不会接受 IM 中临时输入的任意本地路径，只能跳到索引内目标；
+配置了角色列表时，这些命令需要 admin 权限。
+
+当适配器提供媒体 URL 时，gateway 会把文件下载到当前工作区的
+`.reasonix/attachments`，并以 `@.reasonix/attachments/...` 形式传给
+Reasonix。保存失败的附件会在 IM 中提示，文本内容仍会继续处理。内置
+Feishu、Weixin、QQ 适配器当前仍以文本事件为主，普通 IM 附件抽取可以继续在
+适配器层补齐。
+
 ## 审批与 YOLO
 
 Reasonix 的机器人沿用桌面端权限系统。默认是询问模式：写文件、执行命令等
@@ -277,11 +425,11 @@ YOLO 的边界很重要：
 | 现象 | 可以检查 |
 | --- | --- |
 | 扫码提示链接失效 | 回到设置页重新生成二维码；二维码有有效期（飞书、Lark、微信；QQ 不使用扫码，请检查手动配置）。 |
-| 已连接但没有回复 | 确认桌面端 Reasonix 正在运行，Bot 连接已开启，用户 ID 在白名单内或允许所有人。 |
+| 已连接但没有回复 | 确认桌面端 bot runtime 或 `reasonix bot start` 进程正在运行，Bot 连接已开启，用户 ID 在白名单内、已配对或允许所有人。 |
 | 飞书或 Lark 按钮提示失败 | 直接发送卡片里的命令，例如 `/approve <id>` 或 `/deny <id>`。 |
 | QQ 按钮提示失败 | 与飞书/Lark 相同 —— 直接发送卡片里的命令，例如 `/approve <id>` 或 `/deny <id>`。 |
-| 微信回复 `1` 没反应 | 只有存在待审批或单选 Ask 时数字快捷回复才生效；也可以使用完整命令。 |
-| QQ 回复 `1` 没反应 | 与微信相同 —— 只有存在待审批或单选 Ask 时数字快捷回复才生效；也可以使用完整命令。 |
+| 微信回复 `1` 没反应 | 只有存在待审批或 Ask 时数字快捷回复才生效；也可以使用完整命令。 |
+| QQ 回复 `1` 没反应 | 与微信相同 —— 只有存在待审批或 Ask 时数字快捷回复才生效；也可以使用完整命令。 |
 | 想确认当前模式 | 发送 `/status` 或 `/yolo status`。 |
 | 想重新开始上下文 | 发送 `/new` 或 `/reset`。 |
 | 想停止当前任务 | 发送 `/stop`。 |

docs/CONFIG_PATHS.md

@@ -117,6 +117,11 @@ complete chat request URL, set `chat_url`; Reasonix will use it directly and wil
 not append `/chat/completions`. If model discovery needs a separate address, set
 `models_url`.
 
+If a gateway requires vendor-specific top-level request body fields, set
+`extra_body`, for example `extra_body = { enable_thinking = true }`. These values
+are merged into the OpenAI-compatible chat JSON request body without allowing
+core fields such as `model`, `messages`, `tools`, or `stream` to be overridden.
+
 ## Global `.env`
 
 `<Reasonix home>/.env` is the single runtime source for provider API keys saved

docs/DESKTOP_HOOKS.zh-CN.md

@@ -117,7 +117,7 @@ Hooks 让 Reasonix 在会话、用户输入、工具调用、模型返回、压
 | `UserPromptSubmit` | 用户输入提交后、本轮模型调用前 | 是 | 无特殊作用 |
 | `Stop` | 一轮对话结束后 | 否 | 无特殊作用 |
 | `PostLLMCall` | 模型流式返回完成后，reasoning 入库前 | 否 | exit 0 且 stdout 非空时，用 stdout 替换展示的 reasoning |
-| `SessionStart` | 会话第一次变为活跃，或 `/new`、清空后新会话开始 | 否 | 无特殊作用 |
+| `SessionStart` | 会话第一次变为活跃，或 `/new`、清空后新会话开始 | 否 | stdout 会作为下一轮模型上下文注入 |
 | `SessionEnd` | 会话关闭、切换、`/new`、清空或控制器释放时 | 否 | 无特殊作用 |
 | `SubagentStop` | 前台 `task` 子代理完成后 | 否 | 无特殊作用 |
 | `Notification` | 需要用户注意时，例如等待工具审批 | 否 | 无特殊作用 |
@@ -166,9 +166,37 @@ stdout 和 stderr 会被捕获、去掉首尾空白，并限制单路输出最
 特殊 stdout 行为：
 
 - `PostLLMCall`：exit 0 且 stdout 非空时，stdout 会替换用户看到的 reasoning。若 provider 的 reasoning 带签名，Reasonix 会保留原始 signed reasoning 用于后续请求，同时仍展示 hook 转换后的文本。
+- `SessionStart`：exit 0 且 stdout 非空时，stdout 会作为一次性 `<hook-context event="SessionStart">` 注入下一轮真实用户输入。纯文本 stdout 会原样作为上下文；也可以输出 Claude Code / Codex 兼容 JSON：
+
+  ```json
+  {
+    "hookSpecificOutput": {
+      "hookEventName": "SessionStart",
+      "additionalContext": "Load the workspace conventions before editing."
+    }
+  }
+  ```
+
+  `hookEventName` 必须与当前事件一致。该上下文不会写入 system prompt、工具 schema 或项目记忆；它只影响下一轮模型请求。单个 hook 上下文最多保留约 10000 字符，总量最多约 20000 字符，超出会截断并标记。
 - `PreCompact`：所有非空 stdout 会按换行拼接，作为本次压缩摘要的额外指导。
 - 其它事件：stdout 只在非通过结果中作为提示文本使用，不会自动进入模型上下文。
 
+## 示例：SessionStart 注入启动上下文
+
+```json
+{
+  "hooks": {
+    "SessionStart": [
+      {
+        "command": "printf '%s\\n' '{\"hookSpecificOutput\":{\"hookEventName\":\"SessionStart\",\"additionalContext\":\"Before coding, check the available skills and follow matching workflows.\"}}'"
+      }
+    ]
+  }
+}
+```
+
+这适合把插件或工作流的 bootstrap 说明带入会话。比如 Superpowers 不需要内置到 Reasonix；可以让它自己的 `hooks/session-start-codex` 在 `SessionStart` 输出 `additionalContext`，或让插件根目录 `CLAUDE.md` 被插件包兼容层直接作为 `SessionStart` 上下文读取，Reasonix 会在下一轮把这段说明注入模型上下文。插件包兼容层也会读取 `.claude/settings.json` 里的 command hooks，并按同名事件映射到 Reasonix hooks。Reasonix 默认允许 `max_subagent_depth = 2`，因此 Superpowers 的父会话或第一层 workflow subagent 可以再派发 reviewer/implementer subagent；第二层不会继续获得递归委派工具。若要恢复旧的单层边界，设 `agent.max_subagent_depth = 1`。这会改变子代理可见工具面，可能影响子代理请求的 prompt cache，但不会把 Superpowers 写进 Reasonix 的稳定 system prompt。
+
 ## 示例：阻止危险 bash 命令
 
 `.reasonix/settings.json`：

docs/GUIDE.md

@@ -17,6 +17,7 @@
 - [Configuration paths](./CONFIG_PATHS.md)
 - [Reasoning language](./REASONING_LANGUAGE.md)
 - [Custom OpenAI-compatible providers](#custom-openai-compatible-providers)
+- [Desktop hooks](#desktop-hooks)
 - [Keyboard shortcuts](#keyboard-shortcuts)
 - [Permissions & sandbox](#permissions--sandbox)
 - [Plugins (MCP)](#plugins-mcp)
@@ -60,9 +61,11 @@ planner_max_steps = 0            # user/global only; planner read-only tool-call
 reasoning_language = "auto"      # visible reasoning text: auto|zh|en
 # plan_mode_allowed_tools = ["custom_reader"]   # extra read-only custom tools only;
 #                                                # does not unlock blocked tools or unsafe bash
+# plan_mode_read_only_commands = ["gh issue view", "gh pr diff"]   # extra read-only shell prefixes for planning
 # planner_model = "deepseek-pro"      # optional low-frequency planner
 # subagent_model = "deepseek-pro"     # optional default for runAs=subagent skills
 # subagent_models = { review = "deepseek-pro", security_review = "deepseek-pro" }
+# max_subagent_depth = 2              # nested delegation depth; set 1 for the old single-layer boundary
 auto_plan = "off"                  # user-level only; off|on; off keeps plan mode manual
 # auto_plan_classifier = "deepseek-flash"   # optional; only borderline tasks call it
 tool_result_snip_ratio = 0.6       # shorten stale tool output before summary compaction
@@ -125,6 +128,18 @@ compatibility escape valve. It never unlocks known blocked plan-mode tools such
 as `bash`, `task`, writers, installers, or memory mutation tools, and it never
 bypasses bash's plan-mode safety checks.
 
+Use `[agent].plan_mode_read_only_commands` when plan-mode research needs a
+specific shell command that Reasonix cannot classify but you know is read-only,
+such as `gh issue view` or an internal query CLI. Entries are concrete command
+prefixes, not tool names: `["gh issue view"]` permits `gh issue view 4572`, while
+`bash`, `sh`, and other shell interpreters are ignored. Shell operators,
+redirection, command substitution, background execution, and unsafe built-in
+command flags remain blocked while planning. In interactive plan mode, Reasonix
+can also ask you to trust a concrete unknown query prefix the first time it is
+needed; the persistent choice writes the same
+`[agent].plan_mode_read_only_commands` entry. Auto/YOLO approval never answers
+this trust prompt.
+
 ### Environment variables
 
 Most day-to-day settings belong in `config.toml` or the global Reasonix `.env`
@@ -203,7 +218,29 @@ user-global `default_model`.
 
 In the desktop app, open **Settings -> Model -> Access -> Add model service ->
 Custom provider** for proxies, aggregators, or self-hosted services that speak
-the OpenAI-compatible chat API.
+the OpenAI-compatible chat API or Anthropic-compatible Messages API.
+
+For common providers, choose **Add model service -> Recommended preset** instead.
+Reasonix can prefill editable custom-provider entries for Kimi CN, Kimi Global,
+Kimi Coding Plan, MiMo API, MiMo Anthropic, MiMo Token Plan CN/SGP/AMS and their
+Anthropic-compatible variants, MiniMax CN/Global API, MiniMax CN/Global
+Anthropic, GLM CN, Z.AI Global, GLM/Z.AI Coding Plan OpenAI-compatible and
+Anthropic-compatible endpoints, OpenCode Go, OpenCode Go Anthropic, OpenCode Zen
+Anthropic, Qwen/DashScope CN/Global, Qwen Coding Plan CN/Global
+OpenAI-compatible and Anthropic-compatible endpoints, StepFun OpenAI-compatible
+and Anthropic-compatible endpoints, NovitaAI, GMI Cloud, Vercel AI Gateway,
+HuggingFace Router, NVIDIA NIM, KiloCode, and Ollama Cloud. Plan names describe
+the access/payment route; they include CN/Global only when the provider exposes
+distinct regional endpoints. Kimi Coding Plan is therefore a dedicated plan
+endpoint, while Kimi direct API is split into CN and Global. The preset path
+usually needs only the provider API key: the key value is stored in Reasonix home
+`.env`, while `config.toml` stores the endpoint, model list, key
+environment-variable name, context window, vision model metadata, proxy bypass
+for China-only endpoints, MiniMax `reasoning_split`, GLM/MiniMax thinking
+heuristics, Anthropic-compatible Bearer auth where needed, Ollama Cloud
+max-effort support, and OpenCode Go per-model reasoning overrides. After adding
+a preset, open its provider card if you need to change models, headers,
+endpoint, or compatibility settings.
 
 Fill **API address** with the provider endpoint that should receive the standard
 chat path. In this mode Reasonix previews and sends chat requests to:
@@ -219,13 +256,90 @@ preview under the field shows the exact request URL that will be used.
 
 Model discovery uses the API address to try likely model-list URLs such as
 `/models` and `/v1/models`. If the gateway requires a separate model-list
-endpoint, open **Advanced settings** and set `models_url`, for example
+endpoint, open **Compatibility settings** and set `models_url`, for example
 `https://gateway.example.com/v1/models`. If discovery is not available, fill the
 model list manually.
 
 **Full URL** still uses the OpenAI-compatible chat request body. It does not
 switch the request schema to the OpenAI Responses API.
 
+### Compatibility settings
+
+The **Compatibility settings (usually leave unchanged)** section is for gateways
+whose authentication, model-list endpoint, or reasoning/thinking request shape
+differs from the normal OpenAI-compatible defaults. Leave these fields at their
+defaults unless the provider documentation or a proxy error tells you otherwise.
+For Anthropic-compatible services, such as some coding-plan endpoints, choose
+**Anthropic-compatible** as the connection protocol before saving.
+
+| Field | What it controls | When to change it |
+| --- | --- | --- |
+| `api_key_env` | The environment-variable name used for this provider's API key. Desktop-saved key values are stored in Reasonix home `.env` under this name; the TOML config stores only the name. | Change it when several providers need distinct keys, or leave it blank for a service that does not require an API key. |
+| `models_url` | The URL used only for model discovery. Chat requests still use the API address or Full URL above. | Set it when `/models` or `/v1/models` is not where the gateway exposes its model list. |
+| Extra request headers | Static HTTP headers, one `Header: value` per line. | Use for gateways such as OpenRouter that require `HTTP-Referer`, `X-Title`, or similar site headers. Keep bearer/API keys in the key field instead of duplicating them here. |
+| Extra request body | A JSON object merged into the top-level chat request body. | Use only for provider-specific flags such as `{"enable_thinking": true}`. Reasonix still owns core fields such as `model`, `messages`, `tools`, `stream`, and `thinking`, and null values are rejected. |
+| Authorization: Bearer | For Anthropic-compatible providers, sends the saved API key as `Authorization: Bearer <key>` instead of `x-api-key`. | Enable it only when the gateway documents Bearer auth, such as MiniMax Global or Vercel AI Gateway. |
+| Model capability mode | Which reasoning request protocol Reasonix should use for this provider. | Keep **Auto-detect** unless the gateway is misdetected or the model docs require a specific reasoning format. |
+| Thinking override | Provider-specific override for `thinking.type`. | Keep **Auto** unless the backend documents `enabled`, `disabled`, or `adaptive`. Unsupported values can make some OpenAI-compatible gateways reject the request. |
+| Balance URL | Optional endpoint for wallet/balance lookup. | Set it when the provider exposes a balance endpoint and you want the desktop status bar to show it. |
+| Context window | The maximum number of tokens this provider keeps in context. `0` means provider default. | Set it when the model's real context size differs from Reasonix's default or built-in metadata. |
+
+Model capability mode options:
+
+| Option | Effect |
+| --- | --- |
+| Auto-detect (recommended) | Reasonix chooses the request shape from model capability metadata and endpoint detection. |
+| DeepSeek thinking | Uses DeepSeek-style thinking control, including `thinking.type` and DeepSeek-supported reasoning depth. |
+| OpenAI reasoning | Uses the standard OpenAI-compatible `reasoning_effort` levels. |
+| Plain chat | Sends no reasoning or thinking control fields. Use this for text-only proxies that reject reasoning parameters. |
+
+Thinking override options:
+
+| Option | Effect |
+| --- | --- |
+| Auto (provider default) | Does not write an explicit provider-level `thinking` override. Reasonix uses the provider/model default behavior. |
+| Enabled | Sends `thinking.type = "enabled"` for compatible providers. |
+| Disabled | Sends `thinking.type = "disabled"` for compatible providers. On DeepSeek-style providers this also avoids sending a reasoning depth hint. |
+| Adaptive (self-adjusting) | Sends or preserves `thinking.type = "adaptive"` only for providers that document adaptive thinking, such as MiniMax-M3-style endpoints. |
+
+Some OpenAI-compatible gateways require non-standard top-level request body
+fields. Add them with `extra_body` on the provider entry:
+
+```toml
+[[providers]]
+name        = "spark"
+kind        = "openai"
+base_url    = "https://maas-coding-api.cn-huabei-1.xf-yun.com/v2"
+models      = ["xopglm52"]
+api_key_env = "SPARK_API_KEY"
+extra_body  = { enable_thinking = true }
+```
+
+`extra_body` is merged into the chat JSON request body. Reasonix keeps core
+fields such as `model`, `messages`, `tools`, `stream`, and `thinking` under its
+own control.
+
+## Desktop hooks
+
+Desktop hooks run local commands at lifecycle events such as `SessionStart`,
+`UserPromptSubmit`, `PreToolUse`, and `PreCompact`. A successful `SessionStart`
+hook may write plain text to stdout, or return JSON with
+`hookSpecificOutput.additionalContext`; Reasonix injects that text once into the
+next real user turn as `<hook-context event="SessionStart">...</hook-context>`.
+This is intended for plugin or workflow bootstrap context, including
+Superpowers-style startup instructions, without baking that workflow into
+Reasonix's system prompt.
+
+Plugin packages can provide this startup context through
+`hooks/session-start-codex` or a plugin-root `CLAUDE.md`. Claude-style
+`.claude/settings.json` command hooks are also mapped to matching Reasonix hook
+events.
+
+The injected hook context is dynamic current-turn context. It does not change
+the stable system prompt, memory prefix, or tool schema, though dynamic content
+can still reduce cache reuse for that turn. The detailed desktop hook schema and
+trust model are documented in [the Chinese desktop hooks guide](./DESKTOP_HOOKS.zh-CN.md).
+
 ## Keyboard shortcuts
 
 Shortcuts are documented by client because users usually look for the keys that
@@ -302,7 +416,8 @@ Chat and transcript shortcuts:
 | `Ctrl+L` or `/cls` | Clears only the visible transcript | The LLM context, session file, tools, memory, and plugins stay loaded. Use `/clear` when you want to discard the conversation context. |
 | `Esc` | Backs out of the current action | It un-sends a just-submitted turn before any reply, cancels a running turn, or clears non-empty input. |
 | Double `Esc` on an empty idle composer | Opens the rewind picker | Same entry point as `/rewind`. |
-| Transcript text selection | Copies transcript text | The full-screen TUI enables mouse reporting, so drag in the transcript to select text in-app, then press `Ctrl+C`, `Super+C`, `Meta+C`, or right-click the active selection to copy it. |
+| Transcript text selection | Copies transcript text | The full-screen TUI enables mouse reporting, so drag in the transcript to select text in-app; releasing the mouse copies it automatically, and `Ctrl+C`/`Super+C`/`Meta+C` or right-clicking the active selection copy it again. |
+| `/mouse` | Toggles in-app mouse capture | Off hands the mouse back to your terminal, restoring its native click-drag selection and right-click context menu, at the cost of in-app drag-select, the transcript scrollbar, and wheel-scroll. Set `REASONIX_DISABLE_MOUSE=1` to start every session with it off. |
 | `Ctrl+C` | Copies, cancels, clears, or quits | Copies an active transcript selection first. Otherwise it cancels a running turn, clears non-empty input, or quits on a second empty-composer press. |
 | `Ctrl+D` | Quits the TUI | Immediate quit. |
 | `Ctrl+V`, `Ctrl+Shift+V`, `Meta+V`, or `Super+V` | Pastes clipboard content | The CLI tries an image first, then falls back to text or file references. |
@@ -363,13 +478,14 @@ edits stay in the project), resolving symlinks and `..` so a link can't tunnel
 out. `forbid_read` optionally hides sensitive directories from the agent's
 read/list/search tools; use absolute paths or `${HOME}` / `${VAR}` references,
 not `~`, because config expansion is environment-variable based. `bash` is
-itself jailed on macOS by default (`[sandbox] bash`, Seatbelt): commands may
-write only those same roots (plus temp and toolchain caches), cannot read
-configured `forbid_read` roots while the OS sandbox is active, and reach the
-network only when `[sandbox] network` is set. Other platforms fall back to
-running unconfined when no OS sandbox is available (see
+itself jailed by default when an OS sandbox is available (`[sandbox] bash`,
+Seatbelt on macOS and bubblewrap on Linux): commands may write only those same
+roots (plus temp and toolchain caches), cannot read configured `forbid_read`
+roots while the OS sandbox is active, and reach the network only when
+`[sandbox] network` is set. When no OS sandbox is available, `bash = "enforce"`
+refuses bash execution instead of running unconfined (see
 [`SPEC.md` §9](./SPEC.md#9-roadmap-not-in-current-scope) for the escape-prompt and
-Linux support still to come).
+broader OS support still to come).
 
 ## Plugins (MCP)
 
@@ -472,8 +588,10 @@ archives, and saved facts on demand instead of injecting that dynamic state into
 the stable system prompt. `/forget <name>` archives a saved fact rather than
 deleting it permanently; the CLI/TUI and desktop memory panel can show those
 archived files for traceability, but they are not searched as active memory.
-Agent-initiated `remember` and `forget` calls always ask for fresh approval and
-show a compact preview of the saved or archived memory before they run.
+Agent-initiated `remember` and `forget` calls always ask for fresh human approval
+and show a compact preview of the saved or archived memory before they run.
+Guardian review cannot answer for the user; non-interactive runs refuse these
+tools instead of auto-approving them.
 Retrieval keeps the top BM25 result while trimming weak common-word matches, and
 0-result responses suggest narrower, more distinctive follow-up searches.
 Memory v5 is enabled by default across the CLI/TUI, `reasonix serve`, and the
@@ -611,11 +729,22 @@ Subagent skills inherit the executor model by default. Set `subagent_model` to
 run them on another configured model, or use `subagent_models` to override only
 specific skills such as `review` or `security_review`.
 
+Subagents may delegate one more layer by default: the root session is depth 0,
+first-layer subagents are depth 1, and the maximum `max_subagent_depth = 2`
+means a depth-1 workflow can dispatch a depth-2 reviewer or implementer. Depth-2
+subagents do not receive recursive agent/skill tools. Set
+`agent.max_subagent_depth = 1` to restore the old single-layer boundary. This is
+intended for workflows such as Superpowers where a workflow skill may dispatch a
+reviewer subagent, while still avoiding unbounded recursion and background
+fanout.
+
 Use `read_only_task` when planning needs isolated, deeper research without
 granting write-capable delegation. Use `read_only_skill` when the same need is
 best expressed through an existing skill. Both run ephemeral read-only
 subagents with only read-only research tools plus safe foreground bash, return
-only the final answer, and do not create resumable subagent transcripts. In
+only the final answer, and do not create resumable subagent transcripts.
+Read-only nested delegation may be available until `max_subagent_depth` is
+reached, but writer-capable `task` / `run_skill` remain unavailable. In
 token economy mode, connect only this narrow surface with
 `connect_tool_source(source="read_only_skill")`; the full `skills` source still
 enables writer-capable skill tools and remains blocked in plan mode.

docs/GUIDE.zh-CN.md

@@ -37,6 +37,8 @@ Reasonix 全局 `<Reasonix home>/.env`。项目 `.env`、home `.env`、继承的
 
 桌面端和 CLI 端的可见思考语言设置，见 [思考语言](./REASONING_LANGUAGE.zh-CN.md)。
 桌面端 Hooks 的 JSON 配置、事件 key 和 payload 字段，见 [桌面端 Hooks](./DESKTOP_HOOKS.zh-CN.md)。
+`SessionStart` hook 可通过 stdout 或 `hookSpecificOutput.additionalContext` 把插件/工作流 bootstrap 内容一次性注入下一轮真实用户输入上下文，而不是写入稳定 system prompt。
+插件包可通过 `hooks/session-start-codex` 或插件根目录 `CLAUDE.md` 提供该启动上下文；Claude 风格 `.claude/settings.json` command hooks 也会按同名事件映射到 Reasonix hooks。
 
 ```toml
 default_model = "deepseek-flash"   # 执行器；设 [agent].planner_model 可加规划器
@@ -52,9 +54,11 @@ planner_max_steps = 0            # 仅用户/全局；规划器只读工具调
 reasoning_language = "auto"      # 可见思考过程语言：auto|zh|en
 # plan_mode_allowed_tools = ["custom_reader"]   # 仅声明额外只读自定义工具；
 #                                                # 不能解锁被计划模式阻断的工具或 unsafe bash
+# plan_mode_read_only_commands = ["gh issue view", "gh pr diff"]   # 计划模式额外只读 shell 前缀
 # planner_model = "deepseek-pro"      # 可选的低频规划器
 # subagent_model = "deepseek-pro"     # runAs=subagent skill 的默认模型
 # subagent_models = { review = "deepseek-pro", security_review = "deepseek-pro" }
+# max_subagent_depth = 2              # 子代理嵌套委派深度；设为 1 可恢复旧的单层边界
 auto_plan = "off"                  # 仅用户级生效；off|on；off 表示计划模式仅手动开启
 # auto_plan_classifier = "deepseek-flash"   # 可选；只在边界任务上调用
 tool_result_snip_ratio = 0.6       # 在摘要 compaction 前先缩短旧工具输出
@@ -113,6 +117,14 @@ planner / read-only research 可用的可信只读工具。优先使用 MCP 只
 再在 plugin 上写 `trusted_read_only_tools`，`plan_mode_allowed_tools` 保留为兼容逃生阀。它不再解锁 `bash`、`task`、
 写文件工具、安装器、记忆变更工具等计划模式已知阻断项，也不会绕过 bash 在计划模式下的安全检查。
 
+当计划阶段需要运行 Reasonix 尚不能自动分类、但你确认只读的 shell 查询命令时，使用
+`[agent].plan_mode_read_only_commands`，例如 `gh issue view`、`gh pr diff` 或内部只读查询 CLI。
+这里声明的是具体命令前缀，不是工具名：`["gh issue view"]` 会允许 `gh issue view 4572`，
+但 `bash`、`sh` 等 shell 解释器前缀会被忽略。shell 操作符、重定向、命令替换、后台进程、
+以及内置命令里的写能力参数在计划模式下仍会被阻断。交互式计划模式第一次需要某个未知查询前缀时，
+Reasonix 也可以提示你确认是否信任它为只读；选择持久信任会写入同一个
+`[agent].plan_mode_read_only_commands` 配置。Auto/YOLO 审批不会替用户回答这个信任提示。
+
 ### 环境变量
 
 多数日常设置应写在 `config.toml` 或前文提到的 Reasonix 全局 `.env` 中。下面这些变量是进程级高级开关；
@@ -178,7 +190,25 @@ Goal、由 `todo_write` 工具驱动的实时 Todo 面板，以及已配置 prov
 ## 自定义 OpenAI-compatible provider
 
 在桌面端打开 **设置 -> 模型 -> 接入 -> 添加模型服务 -> 自定义供应商**，用于接入代理、
-聚合平台或自建 OpenAI-compatible chat API 服务。
+聚合平台或自建 OpenAI-compatible chat API / Anthropic-compatible Messages API 服务。
+
+常用服务优先使用 **添加模型服务 -> 推荐预设**。Reasonix 可以预填可编辑的自定义 provider：
+Kimi CN、Kimi Global、Kimi Coding Plan、MiMo API、MiMo Anthropic、MiMo Token Plan
+CN/SGP/AMS 及其 Anthropic-compatible 变体、MiniMax CN/Global API、MiniMax
+CN/Global Anthropic、GLM CN、Z.AI Global、GLM/Z.AI Coding Plan 的
+OpenAI-compatible 与 Anthropic-compatible 端点、OpenCode Go、OpenCode Go
+Anthropic、OpenCode Zen Anthropic、Qwen/DashScope CN/Global、Qwen Coding Plan
+CN/Global 的 OpenAI-compatible 与 Anthropic-compatible 端点、StepFun
+OpenAI-compatible 与 Anthropic-compatible 端点、NovitaAI、GMI Cloud、Vercel AI
+Gateway、HuggingFace Router、NVIDIA NIM、KiloCode 和 Ollama Cloud。Plan 表示
+访问/付费形态；只有服务商确实提供不同区域端点时，预设名才同时带 CN/Global。
+因此 Kimi Coding Plan 是独立 plan 端点，Kimi 直连 API 才拆成 CN 和 Global。
+预设路径通常只需要填写服务商 API Key：真实 key 会写入 Reasonix home `.env`，
+`config.toml` 只保存端点、模型列表、key 环境变量名、上下文窗口、视觉模型元数据、
+中国区端点直连、MiniMax `reasoning_split`、GLM/MiniMax thinking heuristic、
+Anthropic-compatible 网关需要的 Bearer 认证、Ollama Cloud max-effort 支持，
+以及 OpenCode Go 的每模型 reasoning 覆盖。添加后仍然可以打开 provider 卡片，
+继续修改模型、请求头、端点或兼容设置。
 
 **API 地址** 填写服务端点。默认模式下，Reasonix 会预览并把聊天请求发送到：
 
@@ -191,12 +221,49 @@ Goal、由 `todo_write` 工具驱动的实时 Todo 面板，以及已配置 prov
 输入框下方的预览就是最终请求地址。
 
 模型发现会基于 API 地址尝试 `/models`、`/v1/models` 等候选地址。如果网关要求单独的
-模型列表端点，在 **高级设置** 中填写 `models_url`，例如
+模型列表端点，在 **兼容设置** 中填写 `models_url`，例如
 `https://gateway.example.com/v1/models`。如果接口不支持模型发现，也可以手动填写模型列表。
 
 **完整 URL** 仍使用 OpenAI-compatible chat 请求体；它不会切换成 OpenAI Responses API
 的请求 schema。
 
+### 兼容设置
+
+**兼容设置（通常不用改）** 用于处理认证变量、模型发现地址、请求头、以及 reasoning/thinking
+请求格式和普通 OpenAI-compatible 默认行为不一致的网关。除非服务商文档明确要求，或代理报错说明
+不兼容，否则保持默认值即可。Kimi Coding Plan、MiniMax CN/Global Anthropic 这类 Anthropic-compatible 服务，
+保存前在基础区域把接入协议切到 **Anthropic-compatible**。
+
+| 字段 | 作用 | 什么时候改 |
+| --- | --- | --- |
+| `api_key_env` | 该 provider 使用的 API key 环境变量名。桌面端保存的真实 key 会写入 Reasonix home `.env` 的同名变量；TOML 配置里只保存变量名。 | 多个 provider 需要不同 key 时改名；服务不需要 API key 时可以留空。 |
+| `models_url` | 只用于自动发现模型列表的 URL。聊天请求仍使用上方的 API 地址或完整 URL。 | `/models` 或 `/v1/models` 不是该网关模型列表地址时填写。 |
+| 额外请求头 | 静态 HTTP header，一行一个 `Header: value`。 | OpenRouter 等网关要求 `HTTP-Referer`、`X-Title` 或类似站点来源 header 时使用。API key 仍放在上方密钥字段，不要重复写到这里。 |
+| 额外请求体 | 合并到聊天请求体顶层的 JSON 对象。 | 仅用于服务商专用开关，例如 `{"enable_thinking": true}`。`model`、`messages`、`tools`、`stream`、`thinking` 等核心字段仍由 Reasonix 控制，且不接受 `null` 值。 |
+| Authorization: Bearer | 对 Anthropic-compatible provider，把已保存的 API key 用 `Authorization: Bearer <key>` 发送，而不是 `x-api-key`。 | MiniMax Global、Vercel AI Gateway 等网关文档明确要求 Bearer 认证时开启。 |
+| 模型能力模式 | 指定 Reasonix 对该 provider 使用哪种 reasoning 请求协议。 | 默认用“自动识别”。只有网关被误判，或模型文档要求特定 reasoning 格式时再切换。 |
+| Thinking 覆盖 | provider 专用的 `thinking.type` 覆盖项。 | 默认用 Auto。只有后端文档明确支持 `enabled`、`disabled` 或 `adaptive` 时再手动指定；不支持的值可能让中转站拒绝请求。 |
+| 余额查询 URL | 可选的钱包余额查询接口。 | 服务商提供余额接口，且希望桌面端状态栏显示余额时填写。 |
+| 上下文窗口 | 该 provider 可保留的最大上下文 token 数。`0` 表示使用模型服务默认值。 | 模型实际上下文大小和 Reasonix 默认值或内置元数据不一致时填写。 |
+
+模型能力模式选项：
+
+| 选项 | 作用 |
+| --- | --- |
+| 自动识别（推荐） | Reasonix 根据模型能力元数据和端点自动选择请求格式。 |
+| DeepSeek 思考 | 使用 DeepSeek 风格的 thinking 控制，包括 `thinking.type` 和 DeepSeek 支持的推理深度。 |
+| OpenAI reasoning | 使用标准 OpenAI-compatible 的 `reasoning_effort` 档位。 |
+| 普通聊天（不发送思考参数） | 不发送 reasoning 或 thinking 控制字段。适合会拒绝 reasoning 参数的普通文本代理。 |
+
+Thinking 覆盖选项：
+
+| 选项 | 作用 |
+| --- | --- |
+| Auto（使用服务默认） | 不写 provider 级 `thinking` 覆盖，让 Reasonix 使用 provider/model 默认行为。 |
+| Enabled（开启） | 对兼容 provider 发送 `thinking.type = "enabled"`。 |
+| Disabled（关闭） | 对兼容 provider 发送 `thinking.type = "disabled"`。DeepSeek 风格 provider 下还会避免继续发送推理深度提示。 |
+| Adaptive（自适应） | 仅在服务文档明确支持 adaptive thinking 时使用，例如 MiniMax-M3 风格端点；语义是发送或保留 `thinking.type = "adaptive"`。 |
+
 ## 快捷键
 
 这里按使用端来写，因为用户通常是先知道“我现在在桌面端/CLI”，再找对应按键。
@@ -409,7 +476,8 @@ headers = { Authorization = "Bearer ${STRIPE_KEY}" }
 compaction archive 和已保存事实；这些动态内容不会被塞进稳定的 system prompt 前缀。
 `/forget <name>` 会把已保存事实归档而不是永久删除；CLI/TUI 和桌面记忆面板能显示归档文件用于追溯，
 但它们不会作为 active memory 被检索。检索会保留 BM25 最强命中，同时裁掉弱的泛词命中；
-agent 发起的 `remember` 和 `forget` 每次都会要求新的人工确认，并在执行前展示将保存或归档的记忆摘要。
+agent 发起的 `remember` 和 `forget` 每次都会要求新的人工确认，并在执行前展示将保存或归档的记忆摘要；
+Guardian 审查不能代替用户批准，非交互运行会拒绝这类工具而不是自动批准。
 0 结果会提示 agent 改用更少、更有区分度的词继续查。
 Memory v5 在 CLI/TUI、`reasonix serve` 和桌面端默认开启，因为这些入口共用同一套本地
 controller。它会把本地、按项目隔离的执行轨迹和编译器状态写在 Reasonix home 下，并且只有
@@ -512,10 +580,17 @@ Planner 会看到已加载的 `REASONIX.md` / `AGENTS.md` 记忆，并拿到一
 Subagent skills 默认继承执行器模型。设置 `subagent_model` 可让它们统一走另一个已配置
 模型；设置 `subagent_models` 则只覆盖 `review`、`security_review` 等指定 skill。
 
+Subagent 默认允许再委派一层：根会话是 depth 0，第一层 subagent 是 depth 1，
+`max_subagent_depth = 2` 表示 depth 1 的 workflow 可以再派 depth 2 的 reviewer
+或 implementer；depth 2 不再拿到递归 agent/skill 工具。设
+`agent.max_subagent_depth = 1` 可恢复旧的单层边界。这主要用于 Superpowers 这类
+workflow skill 派发 reviewer subagent 的场景，同时避免无限递归和后台 fanout。
+
 当计划阶段需要隔离上下文做更深的调研时，用 `read_only_task`，而不是放开可写的
 `task`。如果这类调研更适合复用已有 skill，用 `read_only_skill`。两者都会启动
 ephemeral 只读 subagent，只暴露只读研究工具和安全前台 bash，只返回最终答案，不创建
-可续接的 subagent transcript。在 token economy 模式下，只用
+可续接的 subagent transcript。只读嵌套委派会在 `max_subagent_depth` 内可用，但
+可写的 `task` / `run_skill` 仍不可用。在 token economy 模式下，只用
 `connect_tool_source(source="read_only_skill")` 连接这条窄入口；完整的 `skills`
 source 仍会启用可写 skill 工具，plan mode 下继续阻断。
 

docs/MIGRATING.md

@@ -103,14 +103,23 @@ and DeepSeek prefix-cache–oriented design.
   external tools**: `[agent].plan_mode_allowed_tools` now only declares extra
   read-only custom/external tools. It no longer unlocks known blocked plan-mode
   tools such as `bash`, `task`, writers, installers, or memory mutation tools, and
-  unsafe bash commands still remain blocked. An MCP/plugin tool whose read-only
-  status comes from the server's untrusted `readOnlyHint` is confirmed the first
-  time an interactive plan-mode run needs it; choose the persistent option to
-  write the plugin-level `trusted_read_only_tools` raw-name list. Auto/YOLO tool
-  approval does not answer this trust prompt, although a session or persistent
-  trust choice prevents repeat prompts for the same MCP tool. Non-interactive
-  runs still fail closed, so pre-seed `trusted_read_only_tools` or declare a
-  concrete `mcp__<server>__<tool>` when no user can approve. In the desktop MCP
+  unsafe bash commands still remain blocked. To migrate old
+  `plan_mode_allowed_tools = ["bash", ...]` configs, move concrete read-only
+  shell prefixes such as `gh issue view` or internal query CLIs to
+  `[agent].plan_mode_read_only_commands`; do not declare shell interpreters or
+  writer-capable commands there. Interactive plan-mode runs can also ask you to
+  trust a concrete unknown query prefix once, and the persistent choice writes
+  the same `plan_mode_read_only_commands` entry. Auto/YOLO tool approval does
+  not answer this bash trust prompt. Use `read_only_task` / `read_only_skill`
+  instead of trying to unlock `task` / `run_skill` while planning. An MCP/plugin tool
+  whose read-only status comes from the server's untrusted `readOnlyHint` is
+  confirmed the first time an interactive plan-mode run needs it; choose the
+  persistent option to write the plugin-level `trusted_read_only_tools` raw-name
+  list. Auto/YOLO tool approval does not answer this trust prompt, although a
+  session or persistent trust choice prevents repeat prompts for the same MCP
+  tool. Non-interactive runs still fail closed, so pre-seed
+  `trusted_read_only_tools` or declare a concrete `mcp__<server>__<tool>` when no
+  user can approve. In the desktop MCP
   panel, expand a server and use **Pre-trust read-only** for currently listed
   `readOnlyHint` tools, per-tool **Pre-trust** for audited readers, or
   **Untrust** to remove a tool; those actions write the same

docs/PLUGIN_PACKAGES.md

@@ -0,0 +1,262 @@
+# Reasonix Plugin Packages
+
+Reasonix plugin packages bundle skills, hooks, and MCP servers behind one
+installable unit.
+
+## CLI Mode
+
+Use `reasonix plugin` when installing or managing plugin packages from a
+terminal. Plugin packages are installed globally under the Reasonix home
+directory.
+
+### Install From CLI
+
+`install` accepts one source:
+
+- A GitHub repository, such as `git:github.com/obra/superpowers` or
+  `https://github.com/obra/superpowers`.
+- A GitHub branch or subdirectory URL, such as
+  `https://github.com/owner/repo/tree/main/path/to/plugin`.
+- A local directory that contains `reasonix-plugin.json` or
+  `.codex-plugin/plugin.json`.
+
+Preview the install plan without writing files:
+
+```bash
+reasonix plugin install git:github.com/obra/superpowers --dry-run
+```
+
+Install a plugin after reviewing the plan:
+
+```bash
+reasonix plugin install git:github.com/obra/superpowers --yes
+```
+
+Install with an explicit name or replace an installed plugin with the same name:
+
+```bash
+reasonix plugin install git:github.com/obra/superpowers --name superpowers --replace --yes
+```
+
+Use a local directory in developer mode:
+
+```bash
+reasonix plugin install /path/to/plugin --link --replace --yes
+```
+
+CLI install flags:
+
+- `--dry-run` plans and validates the install without writing files.
+- `--yes` is required for any install that writes files.
+- `--replace` allows the source to replace an installed plugin with the same
+  name.
+- `--name <name>` or `--name=<name>` overrides the name from the plugin
+  manifest for this install.
+- `--link` links a local plugin directory instead of copying it into Reasonix's
+  plugin storage. Moving or deleting that directory breaks the linked plugin.
+
+Running `reasonix plugin install <source>` without `--dry-run` or `--yes`
+refuses to write files and prints a reminder to rerun with one of those flags.
+Install and remove commands print the structured JSON response from the same
+install-source backend used by the desktop UI.
+
+Installed plugin state is stored in:
+
+```text
+~/.reasonix/plugin-packages.json
+~/.reasonix/plugins/<name>/
+```
+
+### Manage From CLI
+
+List installed plugins:
+
+```bash
+reasonix plugin list
+```
+
+Show one plugin's metadata, root, source, and exported capability counts:
+
+```bash
+reasonix plugin show superpowers
+```
+
+`show` also prints the concrete capability inventory when available:
+
+- **skills** include suggested `/<skill>` invocations and descriptions.
+- **hooks** list lifecycle events, matchers, and commands or context files.
+- **mcpServers** list server names, transports, and launch targets.
+
+Check that the manifest and skill roots are readable:
+
+```bash
+reasonix plugin doctor superpowers
+```
+
+Enable or disable a plugin without uninstalling it:
+
+```bash
+reasonix plugin disable superpowers
+reasonix plugin enable superpowers
+```
+
+Remove a plugin:
+
+```bash
+reasonix plugin remove superpowers --yes
+```
+
+`remove` also accepts `uninstall` as an alias. It requires `--yes` because it
+writes state and removes copied plugin content. For linked local plugins, the
+external source directory is left in place.
+
+### Use Installed Plugins From CLI
+
+Installed plugins do not open a separate chat surface. When a plugin is enabled,
+Reasonix loads its capabilities into normal interactive sessions:
+
+- Run `/plugins` inside an interactive session to list installed plugin
+  packages. Run `/plugins show <name>` to inspect a plugin's exported skills,
+  hooks, MCP servers, and usage hints without leaving the chat.
+- **Skills** appear in `/skills`. Invoke a skill with `/<skill> [args]`, or ask
+  naturally and let the agent choose a matching skill by description.
+- **Hooks** run automatically at their configured lifecycle events, such as
+  `SessionStart`, `UserPromptSubmit`, `PreToolUse`, or `PostToolUse`.
+- **MCP servers** join the normal MCP/tool flow. Ask for the task you want done;
+  Reasonix can call the plugin's tools when they are relevant.
+
+After installing, enabling, disabling, or updating a plugin from a separate
+terminal while a session is already running, start a new `reasonix` session or
+reopen `/skills` to verify the current session sees the expected skills.
+
+## Desktop Settings
+
+Open **Settings -> Plugins** to install and manage plugin packages without using
+the CLI.
+
+### Install Plugins
+
+The installer has two modes:
+
+- **Local folder**: click **Choose plugin folder** and select a plugin directory
+  on disk. The selected path is shown next to the button.
+- **Git repository**: enter a Git source such as
+  `git:github.com/obra/superpowers`. **Install name (optional)** can override
+  the plugin manifest name for this install or overwrite.
+
+Use the action buttons after choosing the source and options:
+
+- **Preview** validates the source and shows the planned install actions without
+  writing files.
+- **Install plugin** installs the selected source using the current options.
+- **Refresh plugins** reloads the installed-plugin list from disk and config.
+
+Installer options:
+
+- **Overwrite same-name plugin** allows the current source to replace an
+  installed plugin with the same name. Leave it off when duplicate-name installs
+  should fail instead of replacing existing content.
+- **Developer mode: link source folder** appears for **Local folder** installs.
+  It links the selected directory instead of copying it into Reasonix's plugin
+  storage. Use it while developing or debugging a plugin. Moving or deleting the
+  selected directory will break the linked plugin.
+
+Preview is the safest first step for a new Git source or local plugin directory.
+
+### Manage Installed Plugins
+
+The installed-plugin list shows each plugin package and its exported skills,
+hooks, and MCP servers. Use **Refresh plugins** after editing plugin files or
+changing config outside the app.
+
+Expand a plugin row to manage it:
+
+- Enable or disable the plugin.
+- Read **How to use** for the plugin's exported skills, hooks, and MCP servers.
+- **Update** pulls or refreshes an installed plugin when an update source is
+  available.
+- **Doctor** checks the plugin manifest and reports warnings or diagnostics.
+- **Remove plugin** uninstalls the package after confirmation.
+
+### Use Installed Plugins From Desktop
+
+The desktop settings page uses the same runtime model as the CLI:
+
+- Expand an installed plugin to see its **How to use** section.
+- In any desktop session, type `/plugins` to list installed plugins, or
+  `/plugins show <name>` to see the same usage details from the chat surface.
+- Skills are shown with suggested direct commands such as `/plan`; they are also
+  discoverable from `/skills` in a session.
+- Hooks and MCP servers are listed for transparency. They do not need a manual
+  "run" button: enabled hooks trigger automatically, and MCP tools are available
+  through ordinary tool use.
+- If a currently open session does not reflect a plugin change, refresh the
+  plugin list and open a new session.
+
+## Native Manifest
+
+Reasonix plugins can declare `reasonix-plugin.json` at the plugin root:
+
+```json
+{
+  "name": "example",
+  "version": "1.0.0",
+  "description": "Example plugin",
+  "skills": "skills",
+  "hooks": {
+    "SessionStart": [
+      {
+        "command": "hooks/session-start",
+        "description": "Load startup context"
+      }
+    ]
+  },
+  "mcpServers": {
+    "helper": {
+      "command": "bin/helper"
+    }
+  }
+}
+```
+
+Relative paths are resolved inside the plugin root. Reasonix does not run
+third-party install scripts during plugin installation.
+
+## Codex Compatibility
+
+Reasonix also reads Codex plugin manifests at `.codex-plugin/plugin.json`.
+For packages such as Superpowers and Claude-style skill packs, Reasonix maps:
+
+- `skills` to Reasonix skill roots.
+- `hooks/session-start-codex` to the Reasonix `SessionStart` hook when present.
+- A plugin-root `CLAUDE.md` file to a built-in `SessionStart` context hook. The
+  file is read directly by Reasonix, without spawning a shell command.
+- `.claude/settings.json` command hooks to Reasonix hook events when the event
+  names match. Claude's `matcher` field maps to Reasonix `match`; hook commands
+  run as shell commands with the plugin root as `cwd`; Claude `timeout` values
+  are interpreted as seconds.
+
+Unsupported Claude hook item types are skipped with a warning. Reasonix does not
+run third-party install scripts or implement marketplace-specific install
+protocols.
+
+Plugin hooks receive these environment variables:
+
+- `REASONIX_PLUGIN_ROOT`
+- `REASONIX_PLUGIN_NAME`
+- `REASONIX_PLUGIN_VERSION`
+- `REASONIX_HOME`
+- `REASONIX_WORKSPACE_ROOT`
+- `CLAUDE_PROJECT_DIR`
+
+## Desktop Backend Methods
+
+Desktop exposes plugin package operations through Wails methods:
+
+- `Plugins`
+- `PlanPluginInstall`
+- `InstallPlugin`
+- `RemovePlugin`
+- `SetPluginEnabled`
+- `UpdatePlugin`
+- `PluginDoctor`

docs/PLUGIN_PACKAGES.zh-CN.md

@@ -0,0 +1,242 @@
+# Reasonix 插件包
+
+Reasonix 插件包把 skills、hooks 和 MCP servers 组织成一个可安装单元。
+
+## CLI 模式
+
+在终端里使用 `reasonix plugin` 安装和管理插件包。插件包当前按全局范围安装，
+写入 Reasonix home 目录。
+
+### 通过 CLI 安装
+
+`install` 接收一个来源：
+
+- GitHub 仓库，例如 `git:github.com/obra/superpowers` 或
+  `https://github.com/obra/superpowers`。
+- GitHub 分支或子目录 URL，例如
+  `https://github.com/owner/repo/tree/main/path/to/plugin`。
+- 本地目录，目录内需要包含 `reasonix-plugin.json` 或
+  `.codex-plugin/plugin.json`。
+
+只预览安装计划，不写文件：
+
+```bash
+reasonix plugin install git:github.com/obra/superpowers --dry-run
+```
+
+确认计划后安装：
+
+```bash
+reasonix plugin install git:github.com/obra/superpowers --yes
+```
+
+指定安装名称，或覆盖已安装的同名插件：
+
+```bash
+reasonix plugin install git:github.com/obra/superpowers --name superpowers --replace --yes
+```
+
+以开发模式使用本地目录：
+
+```bash
+reasonix plugin install /path/to/plugin --link --replace --yes
+```
+
+CLI 安装参数：
+
+- `--dry-run` 只规划和校验安装，不写文件。
+- `--yes` 用于确认执行会写文件的安装。
+- `--replace` 允许当前来源替换已安装的同名插件。
+- `--name <name>` 或 `--name=<name>` 覆盖插件 manifest 里的名称，
+  作为本次安装名称。
+- `--link` 链接本地插件目录，而不是复制到 Reasonix 的插件存储目录。
+  移动或删除该目录会导致这个链接插件失效。
+
+如果运行 `reasonix plugin install <source>` 时既没有 `--dry-run`，
+也没有 `--yes`，CLI 会拒绝写文件，并提示使用其中一个参数重新运行。
+安装和移除命令会输出结构化 JSON，来源于桌面端同一套 install-source 后端。
+
+插件状态和内容写入：
+
+```text
+~/.reasonix/plugin-packages.json
+~/.reasonix/plugins/<name>/
+```
+
+### 通过 CLI 管理
+
+列出已安装插件：
+
+```bash
+reasonix plugin list
+```
+
+查看某个插件的元数据、根目录、来源以及导出的能力数量：
+
+```bash
+reasonix plugin show superpowers
+```
+
+如果能读取到能力明细，`show` 也会输出具体清单：
+
+- **skills** 会展示建议的 `/<skill>` 调用方式和描述。
+- **hooks** 会展示生命周期事件、matcher、命令或上下文文件。
+- **mcpServers** 会展示服务器名称、传输方式和启动目标。
+
+检查 manifest 和 skill roots 是否可读：
+
+```bash
+reasonix plugin doctor superpowers
+```
+
+在不卸载的情况下启用或禁用插件：
+
+```bash
+reasonix plugin disable superpowers
+reasonix plugin enable superpowers
+```
+
+移除插件：
+
+```bash
+reasonix plugin remove superpowers --yes
+```
+
+`remove` 也可以写成 `uninstall`。它需要 `--yes`，
+因为会写入状态并删除复制安装的插件内容。如果是链接模式安装的本地插件，
+外部源目录会保留。
+
+### 在 CLI 中使用已安装插件
+
+已安装插件不会打开一个独立聊天界面。插件启用后，Reasonix 会把它的能力加载到普通交互会话里：
+
+- 在交互会话里运行 `/plugins` 可以列出已安装插件包。
+  运行 `/plugins show <name>` 可以在不离开聊天的情况下查看该插件导出的
+  skills、hooks、MCP servers 和使用提示。
+- **Skills** 会出现在 `/skills` 中。可以用 `/<skill> [args]` 直接调用，
+  也可以自然描述任务，让 agent 按 description 选择匹配的 skill。
+- **Hooks** 会在配置的生命周期事件里自动运行，例如 `SessionStart`、
+  `UserPromptSubmit`、`PreToolUse` 或 `PostToolUse`。
+- **MCP servers** 会进入正常 MCP/工具流程。用户只需要描述任务，
+  Reasonix 会在相关时调用插件提供的工具。
+
+如果是在另一个终端里安装、启用、禁用或更新插件，而当前已有 `reasonix` 会话正在运行，
+建议开启新会话，或重新打开 `/skills` 确认当前会话能看到预期技能。
+
+## 桌面端设置
+
+打开 **设置 -> 插件**，可以不用 CLI 直接安装和管理插件包。
+
+### 安装插件
+
+安装区有两种模式：
+
+- **本地目录**：点击 **选择插件目录**，从磁盘选择一个插件目录。
+  选中路径会显示在按钮右侧。
+- **Git 仓库**：填写 Git 来源，例如 `git:github.com/obra/superpowers`。
+  **安装名称（可选）** 可覆盖插件 manifest 声明的名称，用于本次安装或覆盖。
+
+选择来源和选项后，再使用操作按钮：
+
+- **预检** 校验来源并展示计划安装动作，不写入文件。
+- **安装插件** 按当前来源和选项执行安装。
+- **刷新插件** 从磁盘和配置重新读取已安装插件列表。
+
+安装选项：
+
+- **覆盖同名插件** 允许当前来源替换已安装的同名插件。关闭时，同名安装会失败，
+  而不是覆盖已有内容。
+- **开发模式：链接源目录** 只在 **本地目录** 模式出现。它不会复制插件，
+  而是直接链接所选目录；适合开发或调试插件。移动或删除该目录会导致这个链接插件失效。
+
+对新的 Git 来源或本地插件目录，建议先点 **预检**。
+
+### 管理已安装插件
+
+已安装插件列表会展示每个插件包以及它导出的 skills、hooks 和 MCP servers。
+通过应用外编辑插件文件或配置后，可点 **刷新插件** 重新读取。
+
+展开插件行后可以：
+
+- 启用或禁用插件。
+- 查看 **使用方法**，了解该插件导出的 skills、hooks 和 MCP servers。
+- 使用 **更新** 拉取或刷新具备更新来源的插件。
+- 使用 **诊断** 检查插件 manifest，并查看警告或诊断信息。
+- 使用 **移除插件**，确认后卸载该插件包。
+
+### 在桌面端使用已安装插件
+
+桌面端设置页和 CLI 使用同一套运行模型：
+
+- 展开已安装插件，可以看到 **使用方法** 区域。
+- 在任意桌面会话里输入 `/plugins` 可以列出已安装插件；
+  输入 `/plugins show <name>` 可以直接从聊天界面查看同一套使用详情。
+- Skills 会展示建议的直接命令，例如 `/plan`；在会话中也可以通过 `/skills` 浏览。
+- Hooks 和 MCP servers 作为透明能力清单展示。它们不需要单独的“运行”按钮：
+  启用的 hooks 会自动触发，MCP 工具会通过普通工具调用流程可用。
+- 如果当前打开的会话没有反映插件变更，刷新插件列表并开启新会话。
+
+## 原生 Manifest
+
+Reasonix 原生插件在根目录声明 `reasonix-plugin.json`：
+
+```json
+{
+  "name": "example",
+  "version": "1.0.0",
+  "description": "Example plugin",
+  "skills": "skills",
+  "hooks": {
+    "SessionStart": [
+      {
+        "command": "hooks/session-start",
+        "description": "Load startup context"
+      }
+    ]
+  },
+  "mcpServers": {
+    "helper": {
+      "command": "bin/helper"
+    }
+  }
+}
+```
+
+相对路径都按插件根目录解析。Reasonix 安装插件时不会执行第三方安装脚本。
+
+## Codex 兼容
+
+Reasonix 也会读取 `.codex-plugin/plugin.json`。对于 Superpowers 和 Claude 风格
+skill 包，Reasonix 会映射：
+
+- `skills` 到 Reasonix skill root。
+- 如果存在 `hooks/session-start-codex`，映射为 Reasonix `SessionStart` hook。
+- 插件根目录的 `CLAUDE.md` 会映射为内置的 `SessionStart` 上下文 hook。
+  Reasonix 会直接读取该文件，不通过 shell 命令。
+- `.claude/settings.json` 里的 command hooks 会按同名事件映射到 Reasonix hooks。
+  Claude 的 `matcher` 字段会映射到 Reasonix `match`；hook 命令以插件根目录作为
+  `cwd` 执行；Claude `timeout` 按秒解析。
+
+不支持的 Claude hook item type 会跳过并产生 warning。Reasonix 不会执行第三方安装脚本，
+也不会实现 marketplace 专用安装协议。
+
+插件 hook 会收到这些环境变量：
+
+- `REASONIX_PLUGIN_ROOT`
+- `REASONIX_PLUGIN_NAME`
+- `REASONIX_PLUGIN_VERSION`
+- `REASONIX_HOME`
+- `REASONIX_WORKSPACE_ROOT`
+- `CLAUDE_PROJECT_DIR`
+
+## 桌面端后端方法
+
+Desktop 通过 Wails 方法暴露插件包操作：
+
+- `Plugins`
+- `PlanPluginInstall`
+- `InstallPlugin`
+- `RemovePlugin`
+- `SetPluginEnabled`
+- `UpdatePlugin`
+- `PluginDoctor`

docs/REASONING_LANGUAGE.md

@@ -19,7 +19,9 @@ explicit without changing the stable system prompt or tool definitions.
 
 The setting is intentionally small:
 
-- `auto` follows the conversation language and injects no extra instruction.
+- `auto` anchors visible reasoning to Chinese when the raw user prompt is
+  clearly Chinese, ignoring injected reference context such as `@file` contents;
+  English and ambiguous turns add no extra instruction.
 - `zh` asks visible reasoning to prefer Simplified Chinese.
 - `en` asks visible reasoning to prefer English.
 
@@ -89,11 +91,15 @@ argument.
 
 ## Cache Behavior
 
-`auto` is the cache-friendliest choice. It injects nothing and relies on the
-existing stable language policy.
+`auto` is still cache-friendly. When the raw user prompt clearly looks Chinese,
+Reasonix adds the same small transient `<reasoning-language>` block for that
+turn; English and ambiguous turns inject nothing and rely on the existing stable
+language policy. Injected reference context such as `@file` contents is ignored
+for this auto decision.
 
-When set to `zh` or `en`, Reasonix adds a small transient
-`<reasoning-language>` block to the user turn. It does not change:
+When set to `zh` or `en`, Reasonix always adds a small transient
+`<reasoning-language>` block to the user turn. In all modes, this does not
+change:
 
 - the system prompt
 - tool schema bytes or ordering

docs/REASONING_LANGUAGE.zh-CN.md

@@ -14,7 +14,7 @@
 
 取值只有三种：
 
-- `auto`：跟随当前对话语言，不额外注入语言指令。
+- `auto`：用户原始提问明显为中文时锚定中文，并忽略 `@file` 等注入的引用内容；英文或不确定时不额外注入语言指令。
 - `zh`：可见思考过程优先使用简体中文。
 - `en`：可见思考过程优先使用英文。
 
@@ -79,9 +79,9 @@ reasoning_language = "auto" # auto|zh|en
 
 ## 缓存影响
 
-`auto` 对缓存最友好。它不会注入任何额外内容，只复用已有的稳定语言策略。
+`auto` 仍然对缓存友好。用户原始提问明显是中文时，Reasonix 会为这一轮加入同样很小的 `<reasoning-language>` 临时 block；英文或信号不明确时不注入，只复用已有的稳定语言策略。`@file` 等注入的引用内容不会参与这个自动判断。
 
-当设置为 `zh` 或 `en` 时，Reasonix 会把一个很小的 `<reasoning-language>` 临时 block 放进本次 user turn。它不会改变：
+当设置为 `zh` 或 `en` 时，Reasonix 总是会把一个很小的 `<reasoning-language>` 临时 block 放进本次 user turn。所有模式下，它都不会改变：
 
 - system prompt
 - 工具 schema 的字节或顺序

docs/REASONING_PROVIDERS.md

@@ -0,0 +1,57 @@
+# Reasoning controls by provider
+
+Reasonix exposes a single `/effort` knob (and the per-provider `effort` /
+`thinking` config fields), but OpenAI-compatible backends disagree on *how*
+chain-of-thought is requested on the wire. The `openai` provider adapts the
+request shape per backend; this table is the reference for which protocol each
+known backend uses and which parameters it honours or ignores.
+
+## Auto-detected backends
+
+These are recognised by base URL (see `internal/provider/openai/host.go`) and
+get a tailored request shape automatically — no extra config needed.
+
+| Provider | Base URL | Reasoning control | `/effort` levels | Notes |
+|----------|----------|-------------------|------------------|-------|
+| DeepSeek | `api.deepseek.com`, `*.deepseek.com` | `thinking.type` + `reasoning_effort` (depth) | `auto`, `disabled`, `high`, `max` | Thinking on by default; `disabled` turns it off via `thinking.type=disabled`. |
+| MiniMax M3 | `api.minimaxi.com`, `*.minimaxi.com` | `thinking.type` (`adaptive`\|`disabled`) | `auto`, `adaptive`, `disabled` | No depth scale; `reasoning_effort` is omitted. |
+| Zhipu GLM | `open.bigmodel.cn` / `*.bigmodel.cn`, `api.z.ai` / `*.z.ai` | `thinking.type` (`enabled`\|`disabled`) | `auto`, `enabled`, `disabled` | **`reasoning_effort` is silently ignored** by the endpoint, so reasoning is driven purely through `thinking.type`. |
+
+## Everything else (standard `reasoning_effort`)
+
+Any other OpenAI-compatible backend falls through to the standard
+`reasoning_effort` scale (`low`\|`medium`\|`high`). Surveyed popular providers
+that need **no special handling** because they already follow this convention:
+
+Qwen (`dashscope.aliyuncs.com`), Moonshot/Kimi (`api.moonshot.cn`), Yi
+(`api.01.ai`), SiliconFlow (`api.siliconflow.cn`), Stepfun (`api.stepfun.com`),
+Groq (`api.groq.com`), Together (`api.together.xyz`), OpenRouter
+(`openrouter.ai`), Perplexity (`api.perplexity.ai`), xAI (`api.x.ai`).
+
+For a backend that uses a binary `thinking.type` toggle but is **not**
+auto-detected, set the vendor-agnostic `thinking` field on the provider entry:
+
+```toml
+[[providers]]
+name        = "my-glm-proxy"
+kind        = "openai"
+base_url    = "https://my-gateway.example.com/v1"
+model       = "glm-4.6"
+api_key_env = "MY_API_KEY"
+thinking    = "disabled"   # enabled | disabled — emits thinking.type
+```
+
+## Troubleshooting
+
+If a model keeps thinking when you asked it not to (or vice versa):
+
+1. Check the table above — a backend may **ignore** the parameter you set
+   (e.g. Zhipu ignores `reasoning_effort`; use `thinking`/`/effort` instead).
+2. If the backend isn't auto-detected, set the explicit `thinking` field.
+3. If the backend uses a non-OpenAI protocol entirely (e.g. Baidu Wenxin), the
+   `openai` kind cannot drive its thinking mode — that needs a dedicated
+   provider kind.
+
+Distinguishing "provider ignores the field" from a Reasonix bug starts here:
+the request shape Reasonix emits is fixed per the table, so a mismatch between
+the table and observed behaviour is the provider's, not Reasonix's.

docs/RELEASING.md

@@ -18,7 +18,7 @@ provides the pre-release buffer instead of a long-lived branch.
 | Surface | Stable | Pre-release buffer |
 |---|---|---|
 | npm | `latest` (0.x), `next` (1.x) | `canary` (`npm i reasonix@canary`) |
-| Desktop | R2 `latest/` pointer | R2 `canary/` pointer (R2-only — never on the GitHub releases page) |
+| Desktop | R2 `latest/` pointer + release gateway | R2 `canary/` pointer + release gateway proxy (never on the GitHub releases page) |
 
 A canary build is isolated: it **never** moves `latest` / `next` / desktop `latest/`.
 Testers opt in explicitly. (Desktop builds carry `-X main.channel=canary`; npm versions
@@ -68,8 +68,13 @@ the `release` environment deployment.
 - Canary version numbers use the workflow `run_number`, so the desktop and CLI canary
   numbers differ (e.g. `canary.11` vs `canary.2`). Only monotonicity per channel matters.
 - A stable `-rc` tag (e.g. `npm-v1.4.0-rc.1`) still ships under `next`, not `canary`.
-- Desktop in-app updates use R2 first. Stable has a GitHub release fallback; canary is
-  R2-only and never appears on the GitHub releases page.
+- Desktop in-app updates use R2 first, then the `crash.reasonix.io` desktop release
+  gateway. The gateway resolves the `desktop-v*` release line directly and never uses
+  GitHub's repository-wide `/releases/latest`, because plain `v*` tags are the CLI
+  release line. Stable CLI releases also carry a compatibility `latest.json` asset so
+  older desktop builds that still use GitHub `latest` do not 404.
+- Canary uses R2 plus the same gateway proxy for the `canary/` pointer; it never
+  appears on the GitHub releases page.
 - Windows and Linux apply downloaded, minisign-verified artifacts in place. macOS
   applies in-app only for Developer ID signed and notarized builds; ad-hoc/local
   builds fall back to the download page.

docs/SESSION_MEMORY_RETRIEVAL.md

@@ -169,8 +169,11 @@ time.
 The controller treats these tools like plan approval:
 
 - Auto approval and YOLO/full-access mode do not bypass them.
+- Guardian/safety review cannot allow them on the user's behalf.
 - Session grants and persistent allow rules are not created for them.
 - Pending memory approvals are not drained when the user toggles auto approval.
+- Non-interactive headless runs and sub-agents refuse them instead of treating
+  `Ask` as autonomous allow.
 
 The approval subject is generated from the tool arguments before the
 `ApprovalRequest` event is emitted:

docs/SPEC.md

@@ -226,8 +226,11 @@ Long tasks eventually fill the model's context window. Reasonix manages this wit
   history when exact original wording or tool output matters.
 - Agent-initiated `remember` and `forget` calls require a fresh human approval
   each time, even when tool auto-approval or YOLO/full-access mode is enabled.
-  The approval request includes a compact preview of the memory being saved or
-  archived, while external notification hooks only receive the tool name.
+  Guardian/safety review cannot answer these prompts on the user's behalf. In
+  non-interactive headless runs or sub-agents, these tools are refused rather
+  than auto-approved. The approval request includes a compact preview of the
+  memory being saved or archived, while external notification hooks only receive
+  the tool name.
   User-initiated memory edits in the local UI are already explicit user actions.
   See [`SESSION_MEMORY_RETRIEVAL.md`](SESSION_MEMORY_RETRIEVAL.md) for the
   detailed implementation contract.
@@ -313,8 +316,16 @@ func (p Policy) Decide(toolName string, readOnly bool, args json.RawMessage) Dec
   decision: `auto`, `yolo`, and the approved-plan execution window do not answer
   it, but an explicit session grant still prevents repeat prompts for the same
   tool. Non-interactive sessions and declined approvals remain fail-closed.
-  Writers, installers, memory mutation, process
-  control, and `complete_step` (read-only yet post-approval only, so it
+  Bash is gated separately: built-in read-only commands and concrete prefixes
+  declared in `[agent].plan_mode_read_only_commands` may run. Interactive
+  controllers may also ask once before running an unknown query-shaped prefix
+  and may remember a persistent approval as the same
+  `plan_mode_read_only_commands` entry. This bash trust prompt is also a fresh
+  user decision: `auto`, `yolo`, and the approved-plan execution window do not
+  answer it, while explicit session/persistent trust prevents repeat prompts for
+  that prefix. Shell operators, background execution, shell interpreters, and
+  unsafe arguments stay blocked while planning. Writers, installers, memory
+  mutation, process control, and `complete_step` (read-only yet post-approval only, so it
   self-reports plan-unsafe) are refused; the enforced invariant is
   PlanSafe ⇒ ReadOnly. An untrusted read-only MCP/plugin tool is therefore
   blocked until the user approves or pre-trusts it, and it is excluded from
@@ -520,6 +531,7 @@ memory_compiler = { enabled = true, verbosity = "observe" }   # user/global only
 reasoning_language = "auto"       # visible reasoning text: auto|zh|en
 # plan_mode_allowed_tools = ["custom_reader"]   # extra read-only declarations for custom tools;
 #                                                # cannot unlock known blocked tools or unsafe bash
+# plan_mode_read_only_commands = ["gh issue view", "gh pr diff"]   # extra read-only shell prefixes for plan mode
 # planner_model = "deepseek-pro"   # optional: two-model collaboration (low-frequency planner)
 # subagent_model = "deepseek-pro"   # optional default for runAs=subagent skills
 # subagent_models = { review = "deepseek-pro", security_review = "deepseek-pro" }
@@ -632,12 +644,14 @@ by default (root = cwd), so edits stay in the project while the agent can still
 update its own global config. `forbid_read` lists directories the agent should
 not read, list, or search; entries support `${VAR}` / `${VAR:-default}` expansion
 and should be absolute, or use `${HOME}` for home-relative secrets such as
-`${HOME}/.ssh`. `bash` is itself jailed on macOS by default
-(`[sandbox] bash = "enforce"`, Seatbelt): each command runs under sandbox-exec
-allowed to write only the same roots (+ temp and toolchain caches), denied reads
-under `forbid_read`, and allowed to reach the network only when `network = true`.
-Unsupported platforms fall back to running unconfined when no OS sandbox is
-available. The escape-prompt and Linux support are Phase 1's remainder (§9).
+`${HOME}/.ssh`. `bash` is itself jailed by default when an OS sandbox is
+available (`[sandbox] bash = "enforce"`: Seatbelt on macOS, bubblewrap on
+Linux): each command is allowed to write only the same roots (+ temp and
+toolchain caches), denied reads under `forbid_read`, and allowed to reach the
+network only when `network = true`.
+When no OS sandbox is available, `bash = "enforce"` refuses bash execution
+instead of running unconfined. The escape-prompt and broader OS support are Phase
+1's remainder (§9).
 
 ## 6. Error Handling
 
@@ -664,13 +678,15 @@ available. The escape-prompt and Linux support are Phase 1's remainder (§9).
 ## 9. Roadmap (not in current scope)
 
 - Sandbox Phase 1: an OS-level jail for `bash` so commands — not just the
-  file-writer built-ins (Phase 0) — are confined to the workspace. **macOS
-  (Seatbelt via `sandbox-exec`) ships, on by default** (see §5). Remaining: (a)
-  the escape-prompt — detect a sandbox-denied failure and offer to re-run the
-  command unconfined via the permission gate (in `reasonix run`, the command just
-  fails and the model adapts), which completes the "allow inside the box, prompt
-  at its edge" model; (b) Linux (bubblewrap / landlock). Shells out to OS tooling
-  so the binary stays dependency-free; Windows is out of scope. With this in
+  file-writer built-ins (Phase 0) — are confined to the workspace. **Seatbelt on
+  macOS and bubblewrap on Linux ship, on by default when available** (see §5).
+  Remaining: (a)
+  the escape-prompt — detect sandbox-unavailable or sandbox-denied failures and
+  offer an explicit, permission-gated unconfined rerun (in `reasonix run`, the
+  command just fails and the model adapts), which completes the "allow inside the
+  box, prompt at its edge" model; (b) broader OS sandbox support beyond current
+  Seatbelt/bubblewrap hosts. Shells out to OS tooling so the binary stays
+  dependency-free. With this in
   place, "always allow" rule persistence becomes optional rather than load-bearing.
 - MCP long tail (deferred deliberately — no consumer / no foundation yet): OAuth
   2.0 + `headersHelper` auth for remote servers; the remaining `.mcp.json` scopes

docs/TOOL_APPROVAL_MODES.zh-CN.md

@@ -49,7 +49,7 @@ Reasonix 桌面端输入框下方的“询问 / 自动 / Yolo”控制的是工
 - 显式 `deny` 规则仍然阻止对应工具或命令。
 - 显式 `ask` 规则仍然会弹出审批。
 - 计划模式的“开始执行”确认仍然需要你选择。
-- `remember` / `forget` 等记忆写入或删除操作仍需要新鲜审批。
+- `remember` / `forget` 等记忆写入或删除操作仍需要新鲜人工审批，Guardian/安全审查不能代替用户批准。
 - ask 问题仍然等待你回答，不会由自动模式代选。
 
 ### 建议使用场景
@@ -86,7 +86,7 @@ Yolo 模式用于最大化连续执行。开启后，普通工具审批会被跳
 
 - Yolo 只跳过工具权限审批，不会跳过计划确认和 ask 问题。
 - Yolo 不会绕过后端的 `deny` 规则、系统权限、工作区限制或沙箱限制。
-- 记忆写入/删除、计划确认等强制新鲜审批仍然会等待你处理。
+- 记忆写入/删除、计划确认等强制新鲜审批仍然会等待你处理；非交互运行和子代理会拒绝这类工具，而不是自动批准。
 - 不建议在生产环境、敏感目录、密钥文件、发布流程或需求不清楚的任务里开启。
 
 ## 与协作方式的组合

docs/index.html

@@ -207,17 +207,17 @@ reasonix"><span data-en>Copy</span><span data-zh>复制</span></button><pre><spa
   </div>
   <div class="panel" data-panel="desk">
     <div class="dls">
-      <a class="dl" href="https://pub-147fb53b9c1e4bbf891a257968619ea7.r2.dev/latest/Reasonix-darwin-arm64.zip" download>
+      <a class="dl" href="https://dl.reasonix.io/latest/Reasonix-darwin-arm64.zip" download>
         <span class="os">macOS</span><span class="fn">Reasonix-darwin-arm64.zip</span><span class="ar" data-en>universal · Intel + Apple Silicon</span><span class="ar" data-zh>通用 · Intel + Apple Silicon</span><span class="go">↓</span>
       </a>
-      <a class="dl" href="https://pub-147fb53b9c1e4bbf891a257968619ea7.r2.dev/latest/Reasonix-windows-amd64-installer.exe" download>
+      <a class="dl" href="https://dl.reasonix.io/latest/Reasonix-windows-amd64-installer.exe" download>
         <span class="os">Windows</span><span class="fn">Reasonix-windows-amd64-installer.exe</span><span class="ar" data-en>installer · x64</span><span class="ar" data-zh>安装器 · x64</span><span class="go">↓</span>
       </a>
-      <a class="dl" href="https://pub-147fb53b9c1e4bbf891a257968619ea7.r2.dev/latest/Reasonix-linux-amd64.tar.gz" download>
+      <a class="dl" href="https://dl.reasonix.io/latest/Reasonix-linux-amd64.tar.gz" download>
         <span class="os">Linux</span><span class="fn">Reasonix-linux-amd64.tar.gz</span><span class="ar" data-en>x64</span><span class="ar" data-zh>x64</span><span class="go">↓</span>
       </a>
     </div>
-    <p class="undertext"><span data-en>Served from the R2 CDN (fast worldwide). Or browse <a href="https://github.com/esengine/DeepSeek-Reasonix/releases/latest">all releases on GitHub</a>.</span><span data-zh>经 R2 CDN 分发(全球加速)。或在 <a href="https://github.com/esengine/DeepSeek-Reasonix/releases/latest">GitHub releases</a> 浏览全部。</span></p>
+    <p class="undertext"><span data-en>Served from the R2 CDN (fast worldwide). Or browse <a href="https://github.com/esengine/DeepSeek-Reasonix/releases">all releases on GitHub</a>.</span><span data-zh>经 R2 CDN 分发(全球加速)。或在 <a href="https://github.com/esengine/DeepSeek-Reasonix/releases">GitHub releases</a> 浏览全部。</span></p>
   </div>
 </section>
 
@@ -258,7 +258,7 @@ reasonix"><span data-en>Copy</span><span data-zh>复制</span></button><pre><spa
       <a href="https://github.com/esengine/DeepSeek-Reasonix">GitHub</a>
       <a href="https://www.npmjs.com/package/reasonix">npm</a>
       <a href="https://platform.deepseek.com" target="_blank" rel="noreferrer">DeepSeek</a>
-      <a href="https://github.com/esengine/DeepSeek-Reasonix/releases/latest"><span data-en>Releases</span><span data-zh>下载</span></a>
+      <a href="https://github.com/esengine/DeepSeek-Reasonix/releases"><span data-en>Releases</span><span data-zh>下载</span></a>
     </div>
   </div>
 </footer>

go.mod

@@ -8,13 +8,23 @@ require (
 	charm.land/bubbles/v2 v2.1.0
 	charm.land/bubbletea/v2 v2.0.7
 	charm.land/lipgloss/v2 v2.0.4
+	git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3
 	github.com/BurntSushi/toml v1.6.0
 	github.com/alecthomas/chroma/v2 v2.27.0
 	github.com/atotto/clipboard v0.1.4
+	github.com/aymanbagabas/go-udiff v0.4.1
+	github.com/bmatcuk/doublestar/v4 v4.10.0
 	github.com/charmbracelet/x/ansi v0.11.7
+	github.com/joho/godotenv v1.5.1
 	github.com/larksuite/oapi-sdk-go/v3 v3.9.5
 	github.com/mattn/go-runewidth v0.0.24
+	github.com/rivo/uniseg v0.4.7
 	github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06
+	github.com/tree-sitter/go-tree-sitter v0.25.0
+	github.com/tree-sitter/tree-sitter-javascript v0.25.0
+	github.com/tree-sitter/tree-sitter-python v0.25.0
+	github.com/tree-sitter/tree-sitter-rust v0.24.2
+	github.com/tree-sitter/tree-sitter-typescript v0.23.2
 	github.com/yuin/goldmark v1.8.2
 	github.com/zalando/go-keyring v0.2.8
 	go.uber.org/goleak v1.3.0
@@ -25,6 +35,8 @@ require (
 	golang.org/x/sys v0.46.0
 	golang.org/x/term v0.44.0
 	golang.org/x/text v0.38.0
+	gopkg.in/yaml.v3 v3.0.1
+	mvdan.cc/sh/v3 v3.13.1
 )
 
 require (
@@ -37,12 +49,13 @@ require (
 	github.com/clipperhouse/uax29/v2 v2.7.0 // indirect
 	github.com/danieljoos/wincred v1.2.3 // indirect
 	github.com/dlclark/regexp2/v2 v2.2.1 // indirect
+	github.com/go-ole/go-ole v1.3.0 // indirect
 	github.com/godbus/dbus/v5 v5.2.2 // indirect
 	github.com/gogo/protobuf v1.3.2 // indirect
 	github.com/gorilla/websocket v1.5.0 // indirect
 	github.com/lucasb-eyer/go-colorful v1.4.0 // indirect
+	github.com/mattn/go-pointer v0.0.1 // indirect
 	github.com/muesli/cancelreader v0.2.2 // indirect
-	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e // indirect
 	golang.org/x/sync v0.21.0 // indirect
 )

go.sum

@@ -4,6 +4,8 @@ charm.land/bubbletea/v2 v2.0.7 h1:7qw2tTAVar7m7klOPBYfTB0mniv/RuexsYwMRNxSeL0=
 charm.land/bubbletea/v2 v2.0.7/go.mod h1:DGW2q8gvzHnOpMpZTORs0aySVHCox5C+2Svk0fci1qs=
 charm.land/lipgloss/v2 v2.0.4 h1:lcPeVtcp23SNra7lHy8iYE4UC2aIipVQ47sbGyyxR5Q=
 charm.land/lipgloss/v2 v2.0.4/go.mod h1:0653x8epbZSzdDfO/XPS1a/uYPOBeSsCssOpJOqDzik=
+git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3 h1:N3IGoHHp9pb6mj1cbXbuaSXV/UMKwmbKLf53nQmtqMA=
+git.sr.ht/~jackmordaunt/go-toast/v2 v2.0.3/go.mod h1:QtOLZGz8olr4qH2vWK0QH0w0O4T9fEIjMuWpKUsH7nc=
 github.com/BurntSushi/toml v1.6.0 h1:dRaEfpa2VI55EwlIW72hMRHdWouJeRF7TPYhI+AUQjk=
 github.com/BurntSushi/toml v1.6.0/go.mod h1:ukJfTF/6rtPPRCnwkur4qwRxa8vTRFBF0uk2lLoLwho=
 github.com/MakeNowJust/heredoc v1.0.0 h1:cXCdzVdstXyiTqTvfqk9SDHpKNjxuom+DOlyEeQ4pzQ=
@@ -18,6 +20,8 @@ github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z
 github.com/atotto/clipboard v0.1.4/go.mod h1:ZY9tmq7sm5xIbd9bOK4onWV4S6X0u6GY7Vn0Yu86PYI=
 github.com/aymanbagabas/go-udiff v0.4.1 h1:OEIrQ8maEeDBXQDoGCbbTTXYJMYRCRO1fnodZ12Gv5o=
 github.com/aymanbagabas/go-udiff v0.4.1/go.mod h1:0L9PGwj20lrtmEMeyw4WKJ/TMyDtvAoK9bf2u/mNo3w=
+github.com/bmatcuk/doublestar/v4 v4.10.0 h1:zU9WiOla1YA122oLM6i4EXvGW62DvKZVxIe6TYWexEs=
+github.com/bmatcuk/doublestar/v4 v4.10.0/go.mod h1:xBQ8jztBU6kakFMg+8WGxn0c6z1fTSPVIjEY1Wr7jzc=
 github.com/charmbracelet/colorprofile v0.4.3 h1:QPa1IWkYI+AOB+fE+mg/5/4HRMZcaXex9t5KX76i20Q=
 github.com/charmbracelet/colorprofile v0.4.3/go.mod h1:/zT4BhpD5aGFpqQQqw7a+VtHCzu+zrQtt1zhMt9mR4Q=
 github.com/charmbracelet/ultraviolet v0.0.0-20260601155805-6cf7526a1b3f h1:vKsPSlO4g4jKfJ9enESgNZ45BkbHngTIq3UxNOzic74=
@@ -43,20 +47,34 @@ github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/dlclark/regexp2/v2 v2.2.1 h1:mf4KkFUj0gJuarK8P+LgiS+Lit7m9N1yAwEfPbee7R0=
 github.com/dlclark/regexp2/v2 v2.2.1/go.mod h1:avUrQvPaLz2DrFNHJF0taWAFFX2C1GMSSoeiqFjcBmU=
+github.com/go-ole/go-ole v1.3.0 h1:Dt6ye7+vXGIKZ7Xtk4s6/xVdGDQynvom7xCFEdWr6uE=
+github.com/go-ole/go-ole v1.3.0/go.mod h1:5LS6F96DhAwUc7C+1HLexzMXY1xGRSryjyPPKW6zv78=
+github.com/go-quicktest/qt v1.101.0 h1:O1K29Txy5P2OK0dGo59b7b0LR6wKfIhttaAhHUyn7eI=
+github.com/go-quicktest/qt v1.101.0/go.mod h1:14Bz/f7NwaXPtdYEgzsx46kqSxVwTbzVZsDC26tQJow=
 github.com/godbus/dbus/v5 v5.2.2 h1:TUR3TgtSVDmjiXOgAAyaZbYmIeP3DPkld3jgKGV8mXQ=
 github.com/godbus/dbus/v5 v5.2.2/go.mod h1:3AAv2+hPq5rdnr5txxxRwiGjPXamgoIHgz9FPBfOp3c=
 github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q=
 github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q=
+github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
+github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
 github.com/gorilla/websocket v1.5.0 h1:PPwGk2jz7EePpoHN/+ClbZu8SPxiqlu12wZP/3sWmnc=
 github.com/gorilla/websocket v1.5.0/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
 github.com/hexops/gotextdiff v1.0.3 h1:gitA9+qJrrTCsiCl7+kh75nPqQt1cx4ZkudSTLoUqJM=
 github.com/hexops/gotextdiff v1.0.3/go.mod h1:pSWU5MAI3yDq+fZBTazCSJysOMbxWL1BSow5/V2vxeg=
+github.com/joho/godotenv v1.5.1 h1:7eLL/+HRGLY0ldzfGMeQkb7vMd0as4CfYvUVzLqw0N0=
+github.com/joho/godotenv v1.5.1/go.mod h1:f4LDr5Voq0i2e/R5DDNOoa2zzDfwtkZa6DnEwAbqwq4=
 github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI2bnpBCr8=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
+github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
+github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/larksuite/oapi-sdk-go/v3 v3.9.5 h1:2axP/8ME2YZNdgvX+10lWeRiAKvXx5SjVKRyePmPE3Q=
 github.com/larksuite/oapi-sdk-go/v3 v3.9.5/go.mod h1:ZEplY+kwuIrj/nqw5uSCINNATcH3KdxSN7y+UxYY5fI=
 github.com/lucasb-eyer/go-colorful v1.4.0 h1:UtrWVfLdarDgc44HcS7pYloGHJUjHV/4FwW4TvVgFr4=
 github.com/lucasb-eyer/go-colorful v1.4.0/go.mod h1:R4dSotOR9KMtayYi1e77YzuveK+i7ruzyGqttikkLy0=
+github.com/mattn/go-pointer v0.0.1 h1:n+XhsuGeVO6MEAp7xyEukFINEa+Quek5psIR/ylA6o0=
+github.com/mattn/go-pointer v0.0.1/go.mod h1:2zXcozF6qYGgmsG+SeTZz3oAbFLdD3OWqnUbNvJZAlc=
 github.com/mattn/go-runewidth v0.0.24 h1:cpokDiIn0MGnhdHwuWnJBITySJ20QyNGnY2kR/ay2DU=
 github.com/mattn/go-runewidth v0.0.24/go.mod h1:XBkDxAl56ILZc9knddidhrOlY5R/pDhgLpndooCuJAs=
 github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELUXHmA=
@@ -65,6 +83,8 @@ github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZb
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
 github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
+github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
+github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
 github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06 h1:OkMGxebDjyw0ULyrTYWeN0UNCCkmCWfjPnIA2W6oviI=
 github.com/sabhiram/go-gitignore v0.0.0-20210923224102-525f6e181f06/go.mod h1:+ePHsJ1keEjQtpvf9HHw0f4ZeJ0TLRsxhunSI2hYJSs=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
@@ -73,6 +93,34 @@ github.com/stretchr/objx v0.5.2/go.mod h1:FRsXN1f5AsAjCGJKqEizvkpNtU+EGNCLh3NxZ/
 github.com/stretchr/testify v1.6.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
 github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
+github.com/tree-sitter/go-tree-sitter v0.25.0 h1:sx6kcg8raRFCvc9BnXglke6axya12krCJF5xJ2sftRU=
+github.com/tree-sitter/go-tree-sitter v0.25.0/go.mod h1:r77ig7BikoZhHrrsjAnv8RqGti5rtSyvDHPzgTPsUuU=
+github.com/tree-sitter/tree-sitter-c v0.23.4 h1:nBPH3FV07DzAD7p0GfNvXM+Y7pNIoPenQWBpvM++t4c=
+github.com/tree-sitter/tree-sitter-c v0.23.4/go.mod h1:MkI5dOiIpeN94LNjeCp8ljXN/953JCwAby4bClMr6bw=
+github.com/tree-sitter/tree-sitter-cpp v0.23.4 h1:LaWZsiqQKvR65yHgKmnaqA+uz6tlDJTJFCyFIeZU/8w=
+github.com/tree-sitter/tree-sitter-cpp v0.23.4/go.mod h1:doqNW64BriC7WBCQ1klf0KmJpdEvfxyXtoEybnBo6v8=
+github.com/tree-sitter/tree-sitter-embedded-template v0.23.2 h1:nFkkH6Sbe56EXLmZBqHHcamTpmz3TId97I16EnGy4rg=
+github.com/tree-sitter/tree-sitter-embedded-template v0.23.2/go.mod h1:HNPOhN0qF3hWluYLdxWs5WbzP/iE4aaRVPMsdxuzIaQ=
+github.com/tree-sitter/tree-sitter-go v0.23.4 h1:yt5KMGnTHS+86pJmLIAZMWxukr8W7Ae1STPvQUuNROA=
+github.com/tree-sitter/tree-sitter-go v0.23.4/go.mod h1:Jrx8QqYN0v7npv1fJRH1AznddllYiCMUChtVjxPK040=
+github.com/tree-sitter/tree-sitter-html v0.23.2 h1:1UYDV+Yd05GGRhVnTcbP58GkKLSHHZwVaN+lBZV11Lc=
+github.com/tree-sitter/tree-sitter-html v0.23.2/go.mod h1:gpUv/dG3Xl/eebqgeYeFMt+JLOY9cgFinb/Nw08a9og=
+github.com/tree-sitter/tree-sitter-java v0.23.5 h1:J9YeMGMwXYlKSP3K4Us8CitC6hjtMjqpeOf2GGo6tig=
+github.com/tree-sitter/tree-sitter-java v0.23.5/go.mod h1:NRKlI8+EznxA7t1Yt3xtraPk1Wzqh3GAIC46wxvc320=
+github.com/tree-sitter/tree-sitter-javascript v0.25.0 h1:ZkWETb66/w8cc13yhfnNuHOLDQWl3BnKlH6f9AdR88c=
+github.com/tree-sitter/tree-sitter-javascript v0.25.0/go.mod h1:lmGD1EJdCA+v0S1u2fFgepMg/opzSg/4pgFym2FPGAs=
+github.com/tree-sitter/tree-sitter-json v0.24.8 h1:tV5rMkihgtiOe14a9LHfDY5kzTl5GNUYe6carZBn0fQ=
+github.com/tree-sitter/tree-sitter-json v0.24.8/go.mod h1:F351KK0KGvCaYbZ5zxwx/gWWvZhIDl0eMtn+1r+gQbo=
+github.com/tree-sitter/tree-sitter-php v0.23.11 h1:iHewsLNDmznh8kgGyfWfujsZxIz1YGbSd2ZTEM0ZiP8=
+github.com/tree-sitter/tree-sitter-php v0.23.11/go.mod h1:T/kbfi+UcCywQfUNAJnGTN/fMSUjnwPXA8k4yoIks74=
+github.com/tree-sitter/tree-sitter-python v0.25.0 h1:O6XD9v8U1LOcRc3cNj9nM7XufrtEBezE6VrpRrHZDf0=
+github.com/tree-sitter/tree-sitter-python v0.25.0/go.mod h1:cpdthSy/Yoa28aJFBscFHlGiU+cnSiSh1kuDVtI8YeM=
+github.com/tree-sitter/tree-sitter-ruby v0.23.1 h1:T/NKHUA+iVbHM440hFx+lzVOzS4dV6z8Qw8ai+72bYo=
+github.com/tree-sitter/tree-sitter-ruby v0.23.1/go.mod h1:kUS4kCCQloFcdX6sdpr8p6r2rogbM6ZjTox5ZOQy8cA=
+github.com/tree-sitter/tree-sitter-rust v0.24.2 h1:NL4nF67ib21RMzzfvkmXlVwe45vvhW10DVyO+D0z/W0=
+github.com/tree-sitter/tree-sitter-rust v0.24.2/go.mod h1:hfeGWic9BAfgTrc7Xf6FaOAguCFJRo3RBbs7QJ6D7MI=
+github.com/tree-sitter/tree-sitter-typescript v0.23.2 h1:/Odvphn18PniVixb9e97X0DbNVsU6Qocv9mfkyzdXwU=
+github.com/tree-sitter/tree-sitter-typescript v0.23.2/go.mod h1:zjzMXT/Ulffel2xfOcAkQQkiAkmgnbtPGlFQw/5X4xA=
 github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e h1:JVG44RsyaB9T2KIHavMF/ppJZNG9ZpyihvCd0w101no=
 github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e/go.mod h1:RbqR21r5mrJuqunuUZ/Dhy/avygyECGrLceyNeo4LiM=
 github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
@@ -110,6 +158,7 @@ golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.46.0 h1:noSf2Fq6F8DBgS+LysIkx7rIExoNHJsxOAtPp4rthXw=
 golang.org/x/sys v0.46.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 golang.org/x/term v0.44.0 h1:0rLvDRCtNj0gZkyIXhCyOb2OAzEhLVqc4B+hrsBhrmc=
@@ -127,6 +176,10 @@ golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8T
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/check.v1 v1.0.0-20180628173108-788fd7840127 h1:qIbj1fsPNlZgppZ+VLlY7N33q108Sa+fhmuc+sWQYwY=
+gopkg.in/check.v1 v1.0.0-20180628173108-788fd7840127/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+mvdan.cc/sh/v3 v3.13.1 h1:DP3TfgZhDkT7lerUdnp6PTGKyxxzz6T+cOlY/xEvfWk=
+mvdan.cc/sh/v3 v3.13.1/go.mod h1:lXJ8SexMvEVcHCoDvAGLZgFJ9Wsm2sulmoNEXGhYZD0=

internal/acp/dispatch.go

@@ -251,8 +251,6 @@ func (s *updateSink) requestPermission(ctx context.Context, a event.Approval) {
 				allow = true
 			case OptAllowAlways:
 				allow, session = true, true
-			case OptAllowPersistent:
-				allow, session, persist = true, true, true
 			}
 		}
 	}
@@ -331,18 +329,16 @@ func (s *updateSink) requestAskQuestion(ctx context.Context, askID string, q eve
 	return label, ok
 }
 
-func approvalOptionNames(tool, subject string) (session, persistent string) {
+func approvalSessionOptionName(tool, subject string) string {
 	sessionRule := permission.SessionGrantRuleForScope(tool, subject)
-	persistentRule := permission.RememberRuleForScope(tool, subject)
-	return "Allow " + sessionRule + " for this session", "Always allow " + persistentRule + " (save to config)"
+	return "Allow " + sessionRule + " for this session"
 }
 
 func approvalOptions(tool, subject string) []PermissionOption {
-	allowSessionName, allowPersistentName := approvalOptionNames(tool, subject)
+	allowSessionName := approvalSessionOptionName(tool, subject)
 	options := []PermissionOption{
 		{OptionID: string(OptAllowOnce), Name: "Allow", Kind: OptAllowOnce},
 		{OptionID: string(OptAllowAlways), Name: allowSessionName, Kind: OptAllowAlways},
-		{OptionID: string(OptAllowPersistent), Name: allowPersistentName, Kind: OptAllowPersistent},
 		{OptionID: string(OptRejectOnce), Name: "Reject", Kind: OptRejectOnce},
 	}
 	return options

internal/acp/dispatch_test.go

@@ -189,6 +189,31 @@ type approveCall struct {
 	persist bool
 }
 
+func invalidACPv1PermissionOptionKind(options []PermissionOption) (PermissionOption, bool) {
+	// ACP v1 schema only accepts these four PermissionOptionKind values. ACP hosts
+	// own cross-session persistence, so Reasonix-specific persistent approvals must
+	// not appear in session/request_permission options.
+	valid := map[PermissionOptionKind]bool{
+		OptAllowOnce:    true,
+		OptAllowAlways:  true,
+		OptRejectOnce:   true,
+		OptRejectAlways: true,
+	}
+	for _, opt := range options {
+		if !valid[opt.Kind] {
+			return opt, true
+		}
+	}
+	return PermissionOption{}, false
+}
+
+func assertACPv1PermissionOptionKinds(t *testing.T, options []PermissionOption) {
+	t.Helper()
+	if opt, ok := invalidACPv1PermissionOptionKind(options); ok {
+		t.Fatalf("permission option %q uses non-ACP-v1 kind %q", opt.OptionID, opt.Kind)
+	}
+}
+
 func TestUpdateSinkApprovalAllowAlways(t *testing.T) {
 	fn := &fakeNotifier{onReq: func(method string, params any) (json.RawMessage, error) {
 		if method != "session/request_permission" {
@@ -208,6 +233,7 @@ func TestUpdateSinkApprovalAllowAlways(t *testing.T) {
 		if p.ToolCall.ToolCallID != "gate-9" {
 			t.Errorf("toolCallId = %q, want gate-9", p.ToolCall.ToolCallID)
 		}
+		assertACPv1PermissionOptionKinds(t, p.Options)
 		res, _ := json.Marshal(PermissionRequestResult{
 			Outcome: PermissionOutcome{Outcome: "selected", OptionID: string(OptAllowAlways)},
 		})
@@ -236,22 +262,30 @@ func TestUpdateSinkApprovalBashPrefix(t *testing.T) {
 		if err := json.Unmarshal(raw, &p); err != nil {
 			t.Fatalf("permission params: %v", err)
 		}
-		// Bash with a safe prefix now uses the same standard options as any
-		// other tool (allow_always / allow_persistent); the old prefix-specific
-		// OptAllowPrefix/OptPersistPrefix are gone.
-		var hasSession, hasPersistent bool
+		// ACP permission options stay within the official spec kinds, and ACP
+		// mode leaves cross-session persistence to the host.
+		assertACPv1PermissionOptionKinds(t, p.Options)
+		var hasOnce, hasSession, hasReject bool
 		for _, opt := range p.Options {
-			hasSession = hasSession || opt.OptionID == string(OptAllowAlways)
-			hasPersistent = hasPersistent || opt.OptionID == string(OptAllowPersistent)
+			switch opt.OptionID {
+			case string(OptAllowOnce):
+				hasOnce = opt.Kind == OptAllowOnce
+			case string(OptAllowAlways):
+				hasSession = opt.Kind == OptAllowAlways
+			case string(OptRejectOnce):
+				hasReject = opt.Kind == OptRejectOnce
+			default:
+				t.Fatalf("unexpected ACP permission option %+v in %+v", opt, p.Options)
+			}
 		}
-		if !hasSession || !hasPersistent {
-			t.Fatalf("options = %+v, want standard session and persistent choices", p.Options)
+		if !hasOnce || !hasSession || !hasReject {
+			t.Fatalf("options = %+v, want allow once, session, reject", p.Options)
 		}
-		if len(p.Options) != 4 {
-			t.Fatalf("options = %+v, want allow once, session, persistent, reject", p.Options)
+		if len(p.Options) != 3 {
+			t.Fatalf("options = %+v, want allow once, session, reject", p.Options)
 		}
 		res, _ := json.Marshal(PermissionRequestResult{
-			Outcome: PermissionOutcome{Outcome: "selected", OptionID: string(OptAllowPersistent)},
+			Outcome: PermissionOutcome{Outcome: "selected", OptionID: string(OptAllowAlways)},
 		})
 		return res, nil
 	}}
@@ -263,7 +297,7 @@ func TestUpdateSinkApprovalBashPrefix(t *testing.T) {
 
 	select {
 	case c := <-got:
-		want := approveCall{id: "10", allow: true, session: true, persist: true}
+		want := approveCall{id: "10", allow: true, session: true, persist: false}
 		if c != want {
 			t.Errorf("approve = %+v, want %+v", c, want)
 		}
@@ -328,6 +362,7 @@ func TestUpdateSinkAskRequestUsesPermissionChoices(t *testing.T) {
 		if len(p.Options) != 3 {
 			t.Fatalf("options = %+v, want two answers plus cancel", p.Options)
 		}
+		assertACPv1PermissionOptionKinds(t, p.Options)
 		if p.Options[0].Name != "Tests - Run the suite" || p.Options[0].Kind != OptAllowOnce {
 			t.Fatalf("first option = %+v", p.Options[0])
 		}

internal/acp/e2e_test.go

@@ -504,6 +504,10 @@ func TestE2EApprovalRoundTrip(t *testing.T) {
 		var pr PermissionRequestParams
 		json.Unmarshal(req.Params, &pr)
 		reqSeen <- pr
+		if _, ok := invalidACPv1PermissionOptionKind(pr.Options); ok {
+			client.replyError(req.ID, ErrInvalidParams, "Invalid params")
+			return
+		}
 		client.reply(req.ID, PermissionRequestResult{
 			Outcome: PermissionOutcome{Outcome: "selected", OptionID: string(OptAllowOnce)},
 		})
@@ -522,6 +526,7 @@ func TestE2EApprovalRoundTrip(t *testing.T) {
 		if !strings.Contains(pr.ToolCall.Title, "writeit") {
 			t.Errorf("permission title = %q, want it to mention writeit", pr.ToolCall.Title)
 		}
+		assertACPv1PermissionOptionKinds(t, pr.Options)
 	case <-time.After(2 * time.Second):
 		t.Fatal("no permission request was raised")
 	}

internal/acp/protocol.go

@@ -490,15 +490,16 @@ type SessionCancelParams struct {
 
 // --- session/request_permission (agent → client request) ---
 
-// PermissionOptionKind classifies an option for host UI styling. Matches main.
+// PermissionOptionKind classifies an option for host UI styling. It is an ACP v1
+// wire enum, so host-visible permission choices must stay within the official
+// protocol values.
 type PermissionOptionKind string
 
 const (
-	OptAllowOnce       PermissionOptionKind = "allow_once"
-	OptAllowAlways     PermissionOptionKind = "allow_always"
-	OptAllowPersistent PermissionOptionKind = "allow_persistent"
-	OptRejectOnce      PermissionOptionKind = "reject_once"
-	OptRejectAlways    PermissionOptionKind = "reject_always"
+	OptAllowOnce    PermissionOptionKind = "allow_once"
+	OptAllowAlways  PermissionOptionKind = "allow_always"
+	OptRejectOnce   PermissionOptionKind = "reject_once"
+	OptRejectAlways PermissionOptionKind = "reject_always"
 )
 
 // PermissionOption is one choice offered to the user for a permission request.

internal/acp/server_test.go

@@ -331,6 +331,10 @@ func (c *rpcClient) reply(id *json.RawMessage, result any) {
 	c.send(map[string]any{"jsonrpc": "2.0", "id": id, "result": result})
 }
 
+func (c *rpcClient) replyError(id *json.RawMessage, code int, message string) {
+	c.send(map[string]any{"jsonrpc": "2.0", "id": id, "error": rpcError{Code: code, Message: message}})
+}
+
 func startServer(t *testing.T, factory Factory) (*rpcClient, func()) {
 	t.Helper()
 	inR, inW := io.Pipe()

internal/agent/agent.go

@@ -11,6 +11,8 @@ import (
 	"time"
 	"unicode/utf8"
 
+	"mvdan.cc/sh/v3/syntax"
+
 	"reasonix/internal/diff"
 	"reasonix/internal/event"
 	"reasonix/internal/evidence"
@@ -21,6 +23,7 @@ import (
 	"reasonix/internal/nilutil"
 	"reasonix/internal/planmode"
 	"reasonix/internal/provider"
+	"reasonix/internal/shellparse"
 	"reasonix/internal/tool"
 )
 
@@ -58,6 +61,7 @@ type Asker interface {
 // callContextKey carries the executing tool call's identity into Execute.
 type callContextKey struct{}
 type parentSessionContextKey struct{}
+type subagentDepthContextKey struct{}
 type userImagesContextKey struct{}
 
 // callContext is the per-call context a tool can read. parentID is the call being
@@ -108,6 +112,24 @@ func ParentSession(ctx context.Context) string {
 	return strings.TrimSpace(parentSession)
 }
 
+// WithSubagentDepth carries the current subagent depth through nested tool calls.
+// The root agent runs at depth 0; each spawned subagent increments by one.
+func WithSubagentDepth(ctx context.Context, depth int) context.Context {
+	if depth < 0 {
+		depth = 0
+	}
+	return context.WithValue(ctx, subagentDepthContextKey{}, depth)
+}
+
+// SubagentDepth returns the current subagent depth carried by a turn context.
+func SubagentDepth(ctx context.Context) int {
+	depth, _ := ctx.Value(subagentDepthContextKey{}).(int)
+	if depth < 0 {
+		return 0
+	}
+	return depth
+}
+
 // WithUserImages carries the data URLs of images the user attached to this turn,
 // resolved by the controller (which owns attachments) since the agent must not
 // depend on it. Run embeds them on the user message; the provider sends them only
@@ -132,19 +154,25 @@ type Gate interface {
 	Check(ctx context.Context, toolName string, args json.RawMessage, readOnly bool) (allow bool, reason string, err error)
 }
 
-// PlanModeReadOnlyTrustRequest describes an external read-only hint that plan
-// mode will not trust without a user decision. ToolName is the provider-visible
-// name; ServerName and RawToolName are the MCP identifiers persisted in config.
+const PlanModeReadOnlyCommandApprovalTool = "plan_mode_read_only_command"
+
+// PlanModeReadOnlyTrustRequest describes a read-only claim that plan mode will
+// not trust without a user decision. For MCP, ServerName and RawToolName are the
+// identifiers persisted in config. For bash, Command is the concrete attempted
+// command and Prefix is the command prefix to trust for planning.
 type PlanModeReadOnlyTrustRequest struct {
 	ToolName    string
 	ServerName  string
 	RawToolName string
+	Command     string
+	Prefix      string
 	Args        json.RawMessage
 }
 
-// PlanModeReadOnlyTrustGate optionally confirms an MCP server's self-reported
-// read-only hint at execution time. It is separate from Gate because the
-// plan-mode check runs before ordinary permission policy.
+// PlanModeReadOnlyTrustGate optionally confirms MCP read-only hints and
+// user-approved bash read-only command prefixes at execution time. It is
+// separate from Gate because the plan-mode check runs before ordinary permission
+// policy.
 type PlanModeReadOnlyTrustGate interface {
 	CheckPlanModeReadOnlyTrust(ctx context.Context, req PlanModeReadOnlyTrustRequest) (allow bool, reason string, err error)
 }
@@ -154,6 +182,17 @@ const (
 	MemoryCompilerVerbosityCompact = "compact"
 )
 
+const DefaultMaxSubagentDepth = 2
+
+// NormalizeMaxSubagentDepth applies the public config contract: values below 1
+// preserve the old single-delegation boundary.
+func NormalizeMaxSubagentDepth(depth int) int {
+	if depth < 1 {
+		return 1
+	}
+	return depth
+}
+
 // ToolHooks fires user-configured shell hooks around each tool call. PreToolUse
 // runs before the call and may block it (block=true; message is the reason fed
 // back to the model); PostToolUse runs after and only surfaces output to the
@@ -314,7 +353,13 @@ type Agent struct {
 	// planModeAllowedTools declares extra custom tools that the centralized
 	// plan-mode policy may treat as read-only. Known blocked tools still lose.
 	// Populated from Options.PlanModeAllowedTools during construction.
-	planModeAllowedTools []string
+	planModeAllowedTools     []string
+	planModeReadOnlyCommands []string
+
+	// subagentDepth tracks the current agent's nesting depth. maxSubagentDepth
+	// caps delegation; when reached, recursive agent/skill tools are excluded.
+	subagentDepth    int
+	maxSubagentDepth int
 
 	// Context management: when a turn's prompt nears contextWindow, the older
 	// middle of the session is summarized away, keeping a token-bounded recent
@@ -533,7 +578,8 @@ func (a *Agent) SetGate(g Gate) {
 }
 
 // SetPlanModeReadOnlyTrustGate installs the optional confirmation path for MCP
-// tools whose read-only flag comes from an external readOnlyHint.
+// tools whose read-only flag comes from an external readOnlyHint and bash
+// commands the user may trust as plan-mode read-only.
 func (a *Agent) SetPlanModeReadOnlyTrustGate(g PlanModeReadOnlyTrustGate) {
 	if nilutil.IsNil(g) {
 		g = nil
@@ -746,6 +792,15 @@ type Options struct {
 	// PlanModeAllowedTools names extra custom tools the plan-mode policy may treat
 	// as read-only. It cannot unlock known blocked tools or unsafe bash commands.
 	PlanModeAllowedTools []string
+	// PlanModeReadOnlyCommands names concrete shell command prefixes that plan mode
+	// may treat as read-only. Shell operators, background execution, and shell
+	// interpreter prefixes remain blocked.
+	PlanModeReadOnlyCommands []string
+
+	// SubagentDepth is the current nesting depth for this agent. Root sessions are
+	// depth 0; child subagents are depth 1. MaxSubagentDepth caps delegation.
+	SubagentDepth    int
+	MaxSubagentDepth int
 
 	// MemoryCompiler enables Memory v5 execution trace writeback and cache-safe
 	// execution-contract compilation.
@@ -801,33 +856,46 @@ func New(prov provider.Provider, tools *tool.Registry, session *Session, opts Op
 	if strings.TrimSpace(maxStepsKey) == "" {
 		maxStepsKey = "agent.max_steps"
 	}
+	maxSubagentDepth := opts.MaxSubagentDepth
+	if maxSubagentDepth == 0 {
+		maxSubagentDepth = DefaultMaxSubagentDepth
+	} else {
+		maxSubagentDepth = NormalizeMaxSubagentDepth(maxSubagentDepth)
+	}
+	subagentDepth := opts.SubagentDepth
+	if subagentDepth < 0 {
+		subagentDepth = 0
+	}
 	a := &Agent{
-		prov:                    prov,
-		tools:                   tools,
-		session:                 session,
-		maxSteps:                opts.MaxSteps,
-		maxStepsKey:             maxStepsKey,
-		temperature:             opts.Temperature,
-		pricing:                 opts.Pricing,
-		usageSource:             usageSourceOrDefault(opts.UsageSource, event.UsageSourceExecutor),
-		sink:                    sink,
-		gate:                    gate,
-		planModeReadOnlyTrust:   planModeReadOnlyTrust,
-		hooks:                   hooks,
-		jobs:                    opts.Jobs,
-		evidence:                evidence.NewLedger(),
-		projectChecks:           append([]instruction.VerifyCheck(nil), opts.ProjectChecks...),
-		contextWindow:           opts.ContextWindow,
-		softCompactRatio:        opts.SoftCompactRatio,
-		toolResultSnipRatio:     opts.ToolResultSnipRatio,
-		compactRatio:            opts.CompactRatio,
-		compactForceRatio:       opts.CompactForceRatio,
-		recentKeep:              opts.RecentKeep,
-		archiveDir:              opts.ArchiveDir,
-		keepPolicy:              opts.KeepPolicy,
-		planModeAllowedTools:    append([]string(nil), opts.PlanModeAllowedTools...),
-		memoryCompiler:          opts.MemoryCompiler,
-		memoryCompilerVerbosity: normalizeMemoryCompilerVerbosity(opts.MemoryCompilerVerbosity),
+		prov:                     prov,
+		tools:                    tools,
+		session:                  session,
+		maxSteps:                 opts.MaxSteps,
+		maxStepsKey:              maxStepsKey,
+		temperature:              opts.Temperature,
+		pricing:                  opts.Pricing,
+		usageSource:              usageSourceOrDefault(opts.UsageSource, event.UsageSourceExecutor),
+		sink:                     sink,
+		gate:                     gate,
+		planModeReadOnlyTrust:    planModeReadOnlyTrust,
+		hooks:                    hooks,
+		jobs:                     opts.Jobs,
+		evidence:                 evidence.NewLedger(),
+		projectChecks:            append([]instruction.VerifyCheck(nil), opts.ProjectChecks...),
+		contextWindow:            opts.ContextWindow,
+		softCompactRatio:         opts.SoftCompactRatio,
+		toolResultSnipRatio:      opts.ToolResultSnipRatio,
+		compactRatio:             opts.CompactRatio,
+		compactForceRatio:        opts.CompactForceRatio,
+		recentKeep:               opts.RecentKeep,
+		archiveDir:               opts.ArchiveDir,
+		keepPolicy:               opts.KeepPolicy,
+		planModeAllowedTools:     append([]string(nil), opts.PlanModeAllowedTools...),
+		planModeReadOnlyCommands: append([]string(nil), opts.PlanModeReadOnlyCommands...),
+		subagentDepth:            subagentDepth,
+		maxSubagentDepth:         maxSubagentDepth,
+		memoryCompiler:           opts.MemoryCompiler,
+		memoryCompilerVerbosity:  normalizeMemoryCompilerVerbosity(opts.MemoryCompilerVerbosity),
 	}
 	// 初始化分类器
 	if opts.UseMemoryCompilerLLMClassification && prov != nil {
@@ -1561,7 +1629,7 @@ func (a *Agent) stream(ctx context.Context, turn int) (string, string, string, [
 	ch, err := a.prov.Stream(ctx, provider.Request{
 		Messages:    a.session.Messages,
 		Tools:       a.tools.Schemas(),
-		Temperature: a.temperature,
+		Temperature: provider.OptionalTemperature(a.temperature),
 	})
 	if err != nil {
 		return "", "", "", nil, nil, false, false, err
@@ -2022,6 +2090,7 @@ func (a *Agent) executeOne(ctx context.Context, call provider.ToolCall) toolOutc
 			errMsg:  "blocked: fresh read required",
 		}
 	}
+	planModeTrustedReadOnly := false
 	if a.planMode.Load() {
 		// Translate the tool's optional plan-mode self-report into the policy's
 		// tri-state. Mirrors the t.(tool.Previewer) assertion precedent below.
@@ -2039,10 +2108,18 @@ func (a *Agent) executeOne(ctx context.Context, call provider.ToolCall) toolOutc
 		if u, ok := t.(tool.PlanModeUntrustedReadOnly); ok {
 			untrusted = u.PlanModeUntrustedReadOnly()
 		}
-		if blocked, msg := a.planModeBlocked(call.Name, t.ReadOnly(), untrusted, safety, json.RawMessage(call.Arguments)); blocked {
+		if decision := a.planModeDecision(call.Name, t.ReadOnly(), untrusted, safety, json.RawMessage(call.Arguments)); decision.Blocked {
 			trustAllowed := false
-			if t.ReadOnly() && untrusted && safety != planmode.PlanSafetyUnsafe {
-				if allow, outcome, handled := a.checkPlanModeReadOnlyTrust(ctx, call, t); handled {
+			if decision.ReadOnlyCommandTrust != nil {
+				if allow, outcome, handled := a.checkPlanModeBashReadOnlyTrust(ctx, call, decision.ReadOnlyCommandTrust); handled {
+					if !allow {
+						return outcome
+					}
+					trustAllowed = true
+					planModeTrustedReadOnly = true
+				}
+			} else if t.ReadOnly() && untrusted && safety != planmode.PlanSafetyUnsafe {
+				if allow, outcome, handled := a.checkPlanModeMCPReadOnlyTrust(ctx, call, t); handled {
 					if !allow {
 						return outcome
 					}
@@ -2051,7 +2128,7 @@ func (a *Agent) executeOne(ctx context.Context, call provider.ToolCall) toolOutc
 			}
 			if !trustAllowed {
 				return toolOutcome{
-					output:  msg,
+					output:  decision.Message,
 					blocked: true,
 					errMsg:  "blocked: plan mode is read-only",
 				}
@@ -2059,7 +2136,8 @@ func (a *Agent) executeOne(ctx context.Context, call provider.ToolCall) toolOutc
 		}
 	}
 	if a.gate != nil {
-		allow, reason, err := a.gate.Check(ctx, call.Name, json.RawMessage(call.Arguments), t.ReadOnly())
+		readOnly := t.ReadOnly() || planModeTrustedReadOnly
+		allow, reason, err := a.gate.Check(ctx, call.Name, json.RawMessage(call.Arguments), readOnly)
 		if err != nil {
 			return toolOutcome{
 				output:  fmt.Sprintf("blocked: %s (%v)", reason, err),
@@ -2101,6 +2179,7 @@ func (a *Agent) executeOne(ctx context.Context, call provider.ToolCall) toolOutc
 		}
 	}
 	cctx := withCallContext(ctx, call.ID, a.sink, a.asker, a.planMode.Load())
+	cctx = WithSubagentDepth(cctx, a.subagentDepth)
 	if a.evidence != nil {
 		cctx = evidence.WithLedger(cctx, a.evidence)
 		cctx = evidence.WithSessionMessages(cctx, a.session.Snapshot())
@@ -2172,7 +2251,7 @@ func (a *Agent) executeOne(ctx context.Context, call provider.ToolCall) toolOutc
 	return toolOutcome{output: body, truncated: truncMsg != "", truncMsg: truncMsg}
 }
 
-func (a *Agent) checkPlanModeReadOnlyTrust(ctx context.Context, call provider.ToolCall, t tool.Tool) (bool, toolOutcome, bool) {
+func (a *Agent) checkPlanModeMCPReadOnlyTrust(ctx context.Context, call provider.ToolCall, t tool.Tool) (bool, toolOutcome, bool) {
 	if a.planModeReadOnlyTrust == nil {
 		return false, toolOutcome{}, false
 	}
@@ -2207,6 +2286,37 @@ func (a *Agent) checkPlanModeReadOnlyTrust(ctx context.Context, call provider.To
 	return true, toolOutcome{}, true
 }
 
+func (a *Agent) checkPlanModeBashReadOnlyTrust(ctx context.Context, call provider.ToolCall, trust *planmode.ReadOnlyCommandTrust) (bool, toolOutcome, bool) {
+	if a.planModeReadOnlyTrust == nil || trust == nil || strings.TrimSpace(trust.Prefix) == "" {
+		return false, toolOutcome{}, false
+	}
+	req := PlanModeReadOnlyTrustRequest{
+		ToolName: PlanModeReadOnlyCommandApprovalTool,
+		Command:  trust.Command,
+		Prefix:   trust.Prefix,
+		Args:     json.RawMessage(call.Arguments),
+	}
+	allow, reason, err := a.planModeReadOnlyTrust.CheckPlanModeReadOnlyTrust(ctx, req)
+	if err != nil {
+		return false, toolOutcome{
+			output:  fmt.Sprintf("blocked: plan-mode read-only command trust approval aborted (%v)", err),
+			blocked: true,
+			errMsg:  fmt.Sprintf("blocked: %v", err),
+		}, true
+	}
+	if !allow {
+		if strings.TrimSpace(reason) == "" {
+			reason = "the user declined to trust this bash command as read-only for plan mode — do not retry it; continue planning with other trusted read-only tools."
+		}
+		return false, toolOutcome{
+			output:  "blocked: " + reason,
+			blocked: true,
+			errMsg:  "blocked by plan-mode bash read-only trust",
+		}, true
+	}
+	return true, toolOutcome{}, true
+}
+
 func planModeMCPTrustTarget(toolName string, t tool.Tool) (server, rawTool string, ok bool) {
 	if meta, metaOK := t.(tool.MCPMetadata); metaOK {
 		server = strings.TrimSpace(meta.MCPServerName())
@@ -2220,14 +2330,21 @@ func planModeMCPTrustTarget(toolName string, t tool.Tool) (server, rawTool strin
 }
 
 func (a *Agent) planModeBlocked(toolName string, readOnly, untrusted bool, safety planmode.PlanSafety, args json.RawMessage) (blocked bool, message string) {
-	decision := planmode.Policy{AllowedTools: a.planModeAllowedTools}.Decide(planmode.Call{
+	decision := a.planModeDecision(toolName, readOnly, untrusted, safety, args)
+	return decision.Blocked, decision.Message
+}
+
+func (a *Agent) planModeDecision(toolName string, readOnly, untrusted bool, safety planmode.PlanSafety, args json.RawMessage) planmode.Decision {
+	return planmode.Policy{
+		AllowedTools:     a.planModeAllowedTools,
+		ReadOnlyCommands: a.planModeReadOnlyCommands,
+	}.Decide(planmode.Call{
 		Name:      toolName,
 		ReadOnly:  readOnly,
 		Untrusted: untrusted,
 		Safety:    safety,
 		Args:      args,
 	})
-	return decision.Blocked, decision.Message
 }
 
 func planModeBashBlocked(args json.RawMessage) (bool, string) {
@@ -2327,6 +2444,9 @@ func canonicalToolArgs(raw string) string {
 }
 
 func normalizeShellCommand(command string) string {
+	if fields, malformed := shellparse.StaticFields(command); malformed == "" && len(fields) > 0 {
+		return strings.Join(fields, " ")
+	}
 	return strings.Join(strings.Fields(command), " ")
 }
 
@@ -2362,6 +2482,78 @@ func shellPythonOpenWrites(lower string) bool {
 }
 
 func hasShellWriteRedirect(command string) bool {
+	file, err := shellparse.ParseBash(command)
+	if err == nil {
+		hasWrite := false
+		syntax.Walk(file, func(node syntax.Node) bool {
+			redir, ok := node.(*syntax.Redirect)
+			if !ok {
+				return true
+			}
+			if bashRedirectWritesFile(command, redir) {
+				hasWrite = true
+				return false
+			}
+			return true
+		})
+		return hasWrite
+	}
+	return hasShellWriteRedirectFallback(command)
+}
+
+func bashRedirectWritesFile(source string, redir *syntax.Redirect) bool {
+	if redir == nil {
+		return false
+	}
+	switch redir.Op {
+	case syntax.RdrOut, syntax.AppOut, syntax.RdrClob, syntax.AppClob,
+		syntax.RdrAll, syntax.RdrAllClob, syntax.AppAll, syntax.AppAllClob,
+		syntax.RdrInOut:
+		return !redirectWordIsNullSink(source, redir.Word)
+	default:
+		return false
+	}
+}
+
+func redirectWordIsNullSink(source string, word *syntax.Word) bool {
+	if word == nil {
+		return false
+	}
+	if value, ok := shellparse.StaticWord(word); ok {
+		if isNullSinkWord(strings.TrimSpace(value)) {
+			return true
+		}
+	}
+	value := strings.TrimSpace(redirectWordSource(source, word))
+	if isNullSinkWord(value) {
+		return true
+	}
+	if len(value) >= 2 && ((value[0] == '\'' && value[len(value)-1] == '\'') || (value[0] == '"' && value[len(value)-1] == '"')) {
+		return isNullSinkWord(value[1 : len(value)-1])
+	}
+	return false
+}
+
+func isNullSinkWord(value string) bool {
+	if value == "/dev/null" {
+		return true
+	}
+	return strings.EqualFold(value, "$null") || strings.EqualFold(value, "nul")
+}
+
+func redirectWordSource(source string, word *syntax.Word) string {
+	if word == nil || !word.Pos().IsValid() || !word.End().IsValid() {
+		return ""
+	}
+	start := int(word.Pos().Offset())
+	end := int(word.End().Offset())
+	if start < 0 || end < start || end > len(source) {
+		return ""
+	}
+	return source[start:end]
+}
+
+func hasShellWriteRedirectFallback(command string) bool {
 	var quote rune
 	var prev rune
 	for _, r := range command {

internal/agent/branch.go

@@ -28,11 +28,18 @@ type BranchMeta struct {
 	WorkspaceRoot    string    `json:"workspace_root,omitempty"`
 	TopicID          string    `json:"topic_id,omitempty"`
 	TopicTitle       string    `json:"topic_title,omitempty"`
+	CustomTitle      string    `json:"custom_title,omitempty"`
 	Model            string    `json:"model,omitempty"`
 	TokenMode        string    `json:"token_mode,omitempty"`
 	Mode             string    `json:"mode,omitempty"`
 	ToolApprovalMode string    `json:"tool_approval_mode,omitempty"`
 	Goal             string    `json:"goal,omitempty"`
+	Recovered        bool      `json:"recovered,omitempty"`
+	RecoveryReason   string    `json:"recovery_reason,omitempty"`
+	RecoveryDigest   string    `json:"recovery_digest,omitempty"`
+	Revision         int64     `json:"revision,omitempty"`
+	ContentDigest    string    `json:"content_digest,omitempty"`
+	WriterID         string    `json:"writer_id,omitempty"`
 	// SchemaVersion records the BranchMeta version that last wrote the listing
 	// fields (Turns/Preview) FROM the session's content. It is stamped only by the
 	// writers that actually derive those counts — Controller.snapshot's
@@ -147,6 +154,9 @@ func saveBranchMeta(sessionPath string, m BranchMeta, touchUpdated bool) error {
 	if touchUpdated || m.UpdatedAt.IsZero() {
 		m.UpdatedAt = now
 	}
+	if existing, ok, err := LoadBranchMeta(sessionPath); err == nil && ok {
+		preserveBranchMetaPersistence(&m, existing)
+	}
 	if err := os.MkdirAll(filepath.Dir(metaPath), 0o755); err != nil {
 		return err
 	}
@@ -176,6 +186,26 @@ func saveBranchMeta(sessionPath string, m BranchMeta, touchUpdated bool) error {
 	return nil
 }
 
+func preserveBranchMetaPersistence(next *BranchMeta, existing BranchMeta) {
+	if next == nil {
+		return
+	}
+	if existing.Revision > next.Revision {
+		next.Revision = existing.Revision
+		next.ContentDigest = existing.ContentDigest
+		next.WriterID = existing.WriterID
+		return
+	}
+	if existing.Revision == next.Revision {
+		if strings.TrimSpace(next.ContentDigest) == "" {
+			next.ContentDigest = existing.ContentDigest
+		}
+		if strings.TrimSpace(next.WriterID) == "" {
+			next.WriterID = existing.WriterID
+		}
+	}
+}
+
 func EnsureBranchMeta(sessionPath string) (BranchMeta, error) {
 	if sessionPath == "" {
 		return BranchMeta{}, fmt.Errorf("empty session path")
@@ -288,9 +318,10 @@ func ListBranches(dir string) ([]BranchInfo, error) {
 	return out, nil
 }
 
-// RenameSession updates the TopicTitle in the session's .jsonl.meta sidecar
-// file. If no meta file exists yet, one is created. This is used by the
-// /rename CLI command and desktop UI to give sessions human-readable names.
+// RenameSession updates the user-chosen display title in the session's
+// .jsonl.meta sidecar file. If no meta file exists yet, one is created. The
+// topic title remains a separate grouping label, so explicit session names do
+// not fight topic auto-titling.
 func RenameSession(sessionPath string, title string) error {
 	if sessionPath == "" {
 		return fmt.Errorf("empty session path")
@@ -299,8 +330,8 @@ func RenameSession(sessionPath string, title string) error {
 	if err != nil {
 		return err
 	}
-	m.TopicTitle = title
-	return SaveBranchMeta(sessionPath, m)
+	m.CustomTitle = strings.TrimSpace(title)
+	return SaveBranchMetaPreserveUpdated(sessionPath, m)
 }
 
 // LoadSessionModel reads the canonical provider/model ref saved beside a

internal/agent/compact.go

@@ -623,7 +623,7 @@ func (a *Agent) summarize(ctx context.Context, region []provider.Message, instru
 			{Role: provider.RoleSystem, Content: sys},
 			{Role: provider.RoleUser, Content: renderTranscript(region)},
 		},
-		Temperature: a.temperature,
+		Temperature: provider.OptionalTemperature(a.temperature),
 	})
 	if err != nil {
 		return "", err

internal/agent/coordinator.go

@@ -281,7 +281,7 @@ func (c *Coordinator) plan(ctx context.Context, input string) (string, error) {
 
 	ch, err := c.planner.Stream(ctx, provider.Request{
 		Messages:    c.plannerSess.Messages,
-		Temperature: c.temperature,
+		Temperature: provider.OptionalTemperature(c.temperature),
 	})
 	if err != nil {
 		return "", err

internal/agent/empty_final_test.go

@@ -58,7 +58,7 @@ func TestRunPrefixesReasoningLanguageOnSyntheticRetry(t *testing.T) {
 	}
 	for i, req := range prov.requests {
 		got := lastUser(req)
-		if !strings.HasPrefix(got, "<reasoning-language>") || !strings.Contains(got, "Simplified Chinese") {
+		if !strings.HasPrefix(got, "<reasoning-language>") || !strings.Contains(got, "简体中文") {
 			t.Fatalf("request %d last user = %q, want reasoning-language prefix", i, got)
 		}
 	}

internal/agent/parallel_tasks.go

@@ -1,23 +1,21 @@
 package agent
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
-	"os"
-	"path/filepath"
 	"strings"
 	"sync"
 
 	"reasonix/internal/event"
-	"reasonix/internal/jobs"
 	"reasonix/internal/provider"
 	"reasonix/internal/tool"
 )
 
-// ParallelTasksTool dispatches multiple sub-agent tasks concurrently and
-// collects all results. Each sub-task runs as a foreground sub-agent in its
+// ParallelTasksTool dispatches multiple read-only sub-agent tasks concurrently
+// and collects all results. Each sub-task runs as a foreground sub-agent in its
 // own goroutine, emitting nested events so the frontend renders independent
 // cards for each sub-task.
 type ParallelTasksTool struct {
@@ -34,7 +32,7 @@ func NewParallelTasksTool(taskTool *TaskTool, reg *tool.Registry) *ParallelTasks
 func (p *ParallelTasksTool) Name() string { return "parallel_tasks" }
 
 func (p *ParallelTasksTool) Description() string {
-	return "Dispatch multiple sub-agent tasks concurrently and collect their results. Each task runs in its own sub-agent in parallel. Blocks until all complete."
+	return "Dispatch multiple read-only sub-agent tasks concurrently and collect their results. Each task runs in its own read-only sub-agent in parallel. Blocks until all complete."
 }
 
 func (p *ParallelTasksTool) Schema() json.RawMessage {
@@ -62,7 +60,9 @@ func (p *ParallelTasksTool) Schema() json.RawMessage {
 }`)
 }
 
-func (p *ParallelTasksTool) ReadOnly() bool { return false }
+func (p *ParallelTasksTool) ReadOnly() bool { return true }
+
+func (p *ParallelTasksTool) PlanModeSafe() bool { return true }
 
 type parallelTaskItem struct {
 	Prompt      string   `json:"prompt"`
@@ -71,7 +71,6 @@ type parallelTaskItem struct {
 	MaxSteps    int      `json:"max_steps"`
 	Model       string   `json:"model"`
 	Effort      string   `json:"effort"`
-	DependsOn   []int    `json:"depends_on"`
 }
 
 type parallelTaskStatus string
@@ -88,7 +87,9 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 	var params struct {
 		Tasks []parallelTaskItem `json:"tasks"`
 	}
-	if err := json.Unmarshal(args, &params); err != nil {
+	dec := json.NewDecoder(bytes.NewReader(args))
+	dec.DisallowUnknownFields()
+	if err := dec.Decode(&params); err != nil {
 		return "", fmt.Errorf("invalid args: %w", err)
 	}
 	if len(params.Tasks) == 0 {
@@ -100,11 +101,14 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 	if err := validateParallelTaskItems(params.Tasks); err != nil {
 		return "", err
 	}
+	if p.taskTool == nil {
+		return "", fmt.Errorf("parallel_tasks is not configured")
+	}
 
 	parentID, sink, _, ok := CallContext(ctx)
 	if !ok || sink == nil {
-		// Fallback: no event sink available, use background-jobs approach.
-		return p.runAsBackgroundJobs(ctx, params.Tasks)
+		parentID = "parallel_tasks"
+		sink = event.Discard
 	}
 
 	type subResult struct {
@@ -115,63 +119,30 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 
 	n := len(params.Tasks)
 
-	// Dependency state: remaining = number of deps not yet completed.
-	remaining := make([]int, n)
 	running := make([]bool, n)
 	done := make([]bool, n)
 	outputs := make([]string, n)
 	taskErrs := make([]error, n)
 	statuses := make([]parallelTaskStatus, n)
-	for i, t := range params.Tasks {
-		remaining[i] = len(t.DependsOn)
+	for i := range params.Tasks {
 		statuses[i] = parallelTaskPending
 	}
 
 	doneCh := make(chan subResult, n)
 	var wg sync.WaitGroup
 
-	wisdomDir, _ := os.MkdirTemp("", "parallel-wisdom-*")
-	if wisdomDir != "" {
-		defer os.RemoveAll(wisdomDir)
-	}
-	makePrompt := func(t parallelTaskItem) string {
-		var prefix strings.Builder
-		if len(t.DependsOn) > 0 && wisdomDir != "" {
-			entries, _ := os.ReadDir(wisdomDir)
-			if len(entries) > 0 {
-				fmt.Fprintf(&prefix, "Previous task results are available at %s. Read the relevant files with read_file before starting.\n\n", wisdomDir)
-			}
-		}
-		prefix.WriteString(t.Prompt)
-		return prefix.String()
-	}
 	makeLabel := func(t parallelTaskItem, idx int) string {
 		if t.Description != "" {
 			return t.Description
 		}
 		return fmt.Sprintf("task-%d", idx+1)
 	}
-	writeWisdom := func(r subResult) {
-		if wisdomDir == "" {
-			return
-		}
-		fname := filepath.Join(wisdomDir, fmt.Sprintf("task-%d.md", r.index+1))
-		switch {
-		case r.output != "":
-			summary := fmt.Sprintf("# Task %d Result\n\n%s", r.index+1, strings.TrimSpace(r.output))
-			_ = os.WriteFile(fname, []byte(summary), 0o644)
-		case r.err != nil:
-			summary := fmt.Sprintf("# Task %d Result\n\nFAILED: %s", r.index+1, r.err)
-			_ = os.WriteFile(fname, []byte(summary), 0o644)
-		}
-	}
 	startTask := func(idx int) {
 		t := params.Tasks[idx]
 		running[idx] = true
-		prompt := makePrompt(t)
 		label := makeLabel(t, idx)
 		subID := fmt.Sprintf("%s/sub-%d", parentID, idx+1)
-		dispatchArgs, _ := json.Marshal(map[string]string{"prompt": prompt, "description": label})
+		dispatchArgs, _ := json.Marshal(map[string]string{"prompt": t.Prompt, "description": label})
 		sink.Emit(event.Event{
 			Kind: event.ToolDispatch,
 			Tool: event.Tool{
@@ -185,7 +156,16 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 			defer wg.Done()
 			nested := subSinkFor(subID, sink)
 			modelRef, effortRef := p.taskTool.effectiveProfile(t.Model, t.Effort)
-			subReg := p.taskTool.buildSubReg(t.Tools)
+			childDepth, depthErr := p.taskTool.nextSubagentDepth(ctx)
+			if depthErr != nil {
+				sink.Emit(event.Event{
+					Kind: event.ToolResult,
+					Tool: event.Tool{ID: subID, ParentID: parentID, Name: "task", Err: depthErr.Error()},
+				})
+				doneCh <- subResult{index: idx, err: depthErr}
+				return
+			}
+			subReg := ReadOnlySubagentToolRegistryForDepth(p.taskTool.parentReg, t.Tools, childDepth, p.taskTool.maxDepth())
 
 			max := t.MaxSteps
 			if max <= 0 {
@@ -202,8 +182,8 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 				return
 			}
 
-			sess := NewSession("")
-			output, runErr := RunSubAgentWithSession(ctx, prov, subReg, sess, prompt, Options{
+			sess := NewSession(DefaultReadOnlyTaskSystemPrompt)
+			output, runErr := RunSubAgentWithSession(ctx, prov, subReg, sess, p.taskTool.withWorkspaceContext(t.Prompt), Options{
 				MaxSteps:            max,
 				Temperature:         p.taskTool.temperature,
 				Pricing:             pricing,
@@ -217,6 +197,8 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 				CompactForceRatio:   p.taskTool.compactForceRatio,
 				ArchiveDir:          p.taskTool.archiveDir,
 				KeepPolicy:          p.taskTool.keepPolicy,
+				SubagentDepth:       childDepth,
+				MaxSubagentDepth:    p.taskTool.maxDepth(),
 			}, nested)
 
 			if ctx.Err() != nil && runErr == nil {
@@ -242,16 +224,6 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 		}()
 	}
 
-	startReady := func() {
-		if ctx.Err() != nil {
-			return
-		}
-		for i := range params.Tasks {
-			if remaining[i] == 0 && !running[i] && !done[i] {
-				startTask(i)
-			}
-		}
-	}
 	markCancelled := func(err error) {
 		for i := range params.Tasks {
 			if done[i] {
@@ -269,8 +241,10 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 	}
 
 	completed := 0
-	startReady()
-	processResult := func(r subResult, schedule bool) {
+	for i := range params.Tasks {
+		startTask(i)
+	}
+	processResult := func(r subResult) {
 		if done[r.index] {
 			return
 		}
@@ -286,32 +260,18 @@ func (p *ParallelTasksTool) Execute(ctx context.Context, args json.RawMessage) (
 		default:
 			statuses[r.index] = parallelTaskFailed
 		}
-		writeWisdom(r)
-
-		for i, t := range params.Tasks {
-			if remaining[i] > 0 && !running[i] && !done[i] {
-				for _, dep := range t.DependsOn {
-					if dep == r.index {
-						remaining[i]--
-					}
-				}
-			}
-		}
-		if schedule {
-			startReady()
-		}
 	}
 	for completed < n {
 		select {
 		case r := <-doneCh:
-			processResult(r, true)
+			processResult(r)
 		case <-ctx.Done():
 			err := ctx.Err()
 		drain:
 			for {
 				select {
 				case r := <-doneCh:
-					processResult(r, false)
+					processResult(r)
 				default:
 					break drain
 				}
@@ -381,132 +341,11 @@ func formatParallelTasksAggregate(outputs []string, errs []error, statuses []par
 	return b.String()
 }
 
-// runAsBackgroundJobs is the fallback path when no event sink is available.
-func (p *ParallelTasksTool) runAsBackgroundJobs(ctx context.Context, tasks []parallelTaskItem) (string, error) {
-	jm, ok := jobs.FromContext(ctx)
-	if !ok {
-		return "", fmt.Errorf("background jobs are not available in this context")
-	}
-	session := jobs.SessionFromContext(ctx)
-
-	if err := validateParallelTaskItems(tasks); err != nil {
-		return "", err
-	}
-
-	type jobRef struct {
-		id    string
-		label string
-		idx   int
-	}
-	var refs []jobRef
-
-	for i, t := range tasks {
-		if strings.TrimSpace(t.Prompt) == "" {
-			return "", fmt.Errorf("task %d: prompt is required", i+1)
-		}
-		label := t.Description
-		if label == "" {
-			label = fmt.Sprintf("task-%d", i+1)
-		}
-
-		subArgs := map[string]interface{}{
-			"prompt":            t.Prompt,
-			"description":       label,
-			"run_in_background": true,
-		}
-		if len(t.Tools) > 0 {
-			subArgs["tools"] = t.Tools
-		}
-		if t.MaxSteps > 0 {
-			subArgs["max_steps"] = t.MaxSteps
-		}
-		if t.Model != "" {
-			subArgs["model"] = t.Model
-		}
-		if t.Effort != "" {
-			subArgs["effort"] = t.Effort
-		}
-
-		subJSON, err := json.Marshal(subArgs)
-		if err != nil {
-			return "", fmt.Errorf("task %d: marshal: %w", i+1, err)
-		}
-
-		result, err := p.taskTool.Execute(ctx, subJSON)
-		if err != nil {
-			return "", fmt.Errorf("task %d dispatch: %w", i+1, err)
-		}
-		refs = append(refs, jobRef{id: extractJobID(result), label: label, idx: i})
-		_ = result
-	}
-
-	if len(refs) == 0 {
-		return "", fmt.Errorf("no tasks were dispatched")
-	}
-
-	jobIDs := make([]string, 0, len(refs))
-	order := make(map[string]int)
-	for _, r := range refs {
-		jobIDs = append(jobIDs, r.id)
-		order[r.id] = r.idx
-	}
-
-	results := jm.WaitForSession(ctx, session, jobIDs, 0)
-	if len(results) == 0 {
-		return "No parallel task results available.", nil
-	}
-
-	var b strings.Builder
-	fmt.Fprintf(&b, "Completed %d parallel tasks:\n", len(results))
-	for _, r := range results {
-		idx := order[r.ID]
-		label := r.ID
-		if r.Label != "" {
-			label = r.Label
-		}
-		fmt.Fprintf(&b, "── %s ──\n[%s] %s\n%s", label, r.ID, r.Status, strings.TrimSpace(r.Output))
-		_ = idx
-	}
-	return b.String(), nil
-}
-
 func validateParallelTaskItems(tasks []parallelTaskItem) error {
 	for i, t := range tasks {
 		if strings.TrimSpace(t.Prompt) == "" {
 			return fmt.Errorf("task %d: prompt is required", i+1)
 		}
-		for j, dep := range t.DependsOn {
-			if dep < 0 || dep >= len(tasks) {
-				return fmt.Errorf("task %d: depends_on[%d] = %d out of range (0-%d)", i+1, j, dep, len(tasks)-1)
-			}
-			if dep == i {
-				return fmt.Errorf("task %d: self-referencing depends_on", i+1)
-			}
-		}
-	}
-
-	state := make([]int, len(tasks)) // 0 = unvisited, 1 = visiting, 2 = done
-	var visit func(int) error
-	visit = func(i int) error {
-		switch state[i] {
-		case 1:
-			return fmt.Errorf("task dependency cycle detected at task %d", i+1)
-		case 2:
-			return nil
-		}
-		state[i] = 1
-		for _, dep := range tasks[i].DependsOn {
-			if err := visit(dep); err != nil {
-				return err
-			}
-		}
-		state[i] = 2
-		return nil
-	}
-	for i := range tasks {
-		if err := visit(i); err != nil {
-			return err
-		}
 	}
 	return nil
 }
@@ -520,15 +359,3 @@ func resolveSubagentProvider(tt *TaskTool, modelRef, effortRef string) (provider
 	// Use the task tool's own defaults.
 	return tt.prov, tt.pricing, tt.contextWindow, nil
 }
-
-func extractJobID(msg string) string {
-	quote := strings.Index(msg, `"`)
-	if quote < 0 {
-		return ""
-	}
-	end := strings.Index(msg[quote+1:], `"`)
-	if end < 0 {
-		return ""
-	}
-	return msg[quote+1 : quote+1+end]
-}

internal/agent/parallel_tasks_test.go

@@ -4,7 +4,9 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
+	"strconv"
 	"strings"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -13,9 +15,13 @@ import (
 	"reasonix/internal/tool"
 )
 
-func TestParallelTasksToolIsWriteCapable(t *testing.T) {
-	if (&ParallelTasksTool{}).ReadOnly() {
-		t.Fatal("parallel_tasks must be write-capable because spawned sub-agents can invoke writer tools")
+func TestParallelTasksToolIsReadOnly(t *testing.T) {
+	p := &ParallelTasksTool{}
+	if !p.ReadOnly() {
+		t.Fatal("parallel_tasks must be read-only because spawned sub-agents receive only read-only tools")
+	}
+	if !p.PlanModeSafe() {
+		t.Fatal("parallel_tasks must be plan-mode safe because spawned sub-agents receive only read-only tools")
 	}
 }
 
@@ -45,22 +51,22 @@ func TestParallelTasksValidatesAllTasksBeforeRuntimeLookup(t *testing.T) {
 	}
 }
 
-func TestParallelTasksRejectsDependencyCyclesBeforeRuntimeLookup(t *testing.T) {
+func TestParallelTasksRejectsHiddenDependencyFieldBeforeRuntimeLookup(t *testing.T) {
 	tool := &ParallelTasksTool{}
 	_, err := tool.Execute(context.Background(), json.RawMessage(`{
 		"tasks": [
 			{"prompt": "first", "depends_on": [1]},
-			{"prompt": "second", "depends_on": [0]}
+			{"prompt": "second"}
 		]
 	}`))
 	if err == nil {
-		t.Fatal("Execute returned nil error for cyclic dependencies")
+		t.Fatal("Execute returned nil error for a hidden dependency field")
 	}
-	if !strings.Contains(err.Error(), "cycle") {
-		t.Fatalf("Execute error = %v, want dependency cycle validation", err)
+	if !strings.Contains(err.Error(), "depends_on") {
+		t.Fatalf("Execute error = %v, want hidden dependency field rejection", err)
 	}
 	if strings.Contains(err.Error(), "background jobs are not available") {
-		t.Fatalf("Execute looked up background jobs before validating dependencies: %v", err)
+		t.Fatalf("Execute looked up background jobs before rejecting hidden dependencies: %v", err)
 	}
 }
 
@@ -98,6 +104,50 @@ func TestParallelTasksForegroundCompletesAndClosesWorkers(t *testing.T) {
 	}
 }
 
+func TestParallelTasksInjectsWorkspaceContextIntoChildren(t *testing.T) {
+	workspace := t.TempDir()
+	task := NewTaskTool(promptRoutingProvider{}, nil, tool.NewRegistry(), 20, 0, 0, 0, 0, 0, 0, 0.0, "", "sys", nil, 0, "", "", nil).
+		WithTranscripts(NewSubagentStore(t.TempDir()), workspace, "base-model", "base-effort")
+	parallel := NewParallelTasksTool(task, tool.NewRegistry())
+	ctx := withCallContext(context.Background(), "parallel-call", event.Discard, nil, false)
+
+	out, err := parallel.Execute(ctx, json.RawMessage(`{"tasks":[{"prompt":"inspect one"},{"prompt":"inspect two"}]}`))
+	if err != nil {
+		t.Fatalf("Execute: %v", err)
+	}
+	if !strings.Contains(out, "Current workspace: "+strconv.Quote(workspace)) ||
+		!strings.Contains(out, `prefer "." or relative paths`) ||
+		!strings.Contains(out, "inspect one ok") ||
+		!strings.Contains(out, "inspect two ok") {
+		t.Fatalf("parallel output = %q, want child workspace context and prompt", out)
+	}
+}
+
+func TestParallelTasksDoesNotExposeWriterToolsToChildren(t *testing.T) {
+	var writerCalls int32
+	parentReg := tool.NewRegistry()
+	parentReg.Add(fakeTool{name: "write_file", readOnly: false, calls: &writerCalls})
+	task := newTestTaskTool(t, writerCallingProvider{}, parentReg, "sys", "", "", nil)
+	parallel := NewParallelTasksTool(task, parentReg)
+	ctx := withCallContext(context.Background(), "parallel-call", event.Discard, nil, false)
+
+	out, err := parallel.Execute(ctx, json.RawMessage(`{
+		"tasks": [
+			{"prompt": "try writer one"},
+			{"prompt": "try writer two"}
+		]
+	}`))
+	if err != nil {
+		t.Fatalf("Execute returned error: %v\n%s", err, out)
+	}
+	if got := atomic.LoadInt32(&writerCalls); got != 0 {
+		t.Fatalf("writer tool was exposed to read-only sub-agents and called %d times", got)
+	}
+	if !strings.Contains(out, "Completed 2 parallel tasks") {
+		t.Fatalf("missing aggregate output: %s", out)
+	}
+}
+
 func TestParallelTasksCancelReturnsPartialAggregate(t *testing.T) {
 	task := newTestTaskTool(t, promptRoutingProvider{}, tool.NewRegistry(), "sys", "", "", nil)
 	parallel := NewParallelTasksTool(task, tool.NewRegistry())
@@ -170,6 +220,33 @@ func (promptRoutingProvider) Stream(_ context.Context, req provider.Request) (<-
 	return ch, nil
 }
 
+type writerCallingProvider struct{}
+
+func (writerCallingProvider) Name() string { return "writer-calling" }
+
+func (writerCallingProvider) Stream(_ context.Context, req provider.Request) (<-chan provider.Chunk, error) {
+	ch := make(chan provider.Chunk, 2)
+	if !hasToolResult(req, "write_file") {
+		ch <- toolCallChunk("write-1", "write_file", `{"path":"x","content":"y"}`)
+		ch <- provider.Chunk{Type: provider.ChunkDone}
+		close(ch)
+		return ch, nil
+	}
+	ch <- provider.Chunk{Type: provider.ChunkText, Text: "writer unavailable"}
+	ch <- provider.Chunk{Type: provider.ChunkDone}
+	close(ch)
+	return ch, nil
+}
+
+func hasToolResult(req provider.Request, name string) bool {
+	for _, m := range req.Messages {
+		if m.Role == provider.RoleTool && m.Name == name {
+			return true
+		}
+	}
+	return false
+}
+
 type stringsError string
 
 func (e stringsError) Error() string { return string(e) }

internal/agent/planmode_test.go

@@ -152,6 +152,14 @@ func TestPlanModeAllowedToolsOverride(t *testing.T) {
 	}
 }
 
+func TestPlanModeReadOnlyCommandsOverride(t *testing.T) {
+	a := &Agent{planModeReadOnlyCommands: []string{"gh issue view"}}
+	blocked, msg := a.planModeBlocked("bash", false, false, planmode.PlanSafetyUnknown, bashCommandArgs(t, "gh issue view 4572 --json title"))
+	if blocked {
+		t.Fatalf("command in planModeReadOnlyCommands should not be blocked: %s", msg)
+	}
+}
+
 func TestPlanModeGenericWriterBlocked(t *testing.T) {
 	blocked, msg := (&Agent{}).planModeBlocked("some_writer_tool", false, false, planmode.PlanSafetyUnknown, nil)
 	if !blocked {
@@ -213,6 +221,18 @@ func (g *fakePlanModeReadOnlyTrustGate) CheckPlanModeReadOnlyTrust(ctx context.C
 	return g.allow, g.reason, nil
 }
 
+type readOnlyRecordingGate struct {
+	readOnly []bool
+}
+
+func (g *readOnlyRecordingGate) Check(ctx context.Context, toolName string, args json.RawMessage, readOnly bool) (bool, string, error) {
+	g.readOnly = append(g.readOnly, readOnly)
+	if !readOnly {
+		return false, "expected read-only", nil
+	}
+	return true, "", nil
+}
+
 // TestPlanModeUntrustedReadOnlyToolFailsClosed proves the gate does NOT trust an
 // MCP tool's self-reported readOnlyHint: a ReadOnly()==true external tool is
 // still fail-closed in plan mode, and runs only once declared in
@@ -301,6 +321,74 @@ func TestPlanModeUnsafeUntrustedReadOnlyToolDoesNotAskForTrust(t *testing.T) {
 	}
 }
 
+func TestPlanModeUnknownBashReadOnlyCommandCanAskForTrust(t *testing.T) {
+	reg := tool.NewRegistry()
+	reg.Add(fakeTool{name: "bash", readOnly: false})
+	gate := &fakePlanModeReadOnlyTrustGate{allow: true}
+	a := New(nil, reg, NewSession(""), Options{PlanModeReadOnlyTrustGate: gate}, event.Discard)
+	a.SetPlanMode(true)
+
+	out := a.executeOne(context.Background(), provider.ToolCall{Name: "bash", Arguments: string(bashCommandArgs(t, "gh issue view 4572 --json title"))})
+	if strings.HasPrefix(out.output, "blocked:") || !strings.Contains(out.output, "done") {
+		t.Fatalf("trusted unknown bash query should run, got: %q", out.output)
+	}
+	if gate.calls != 1 {
+		t.Fatalf("trust gate calls = %d, want 1", gate.calls)
+	}
+	if gate.req.ToolName != PlanModeReadOnlyCommandApprovalTool || gate.req.Command != "gh issue view 4572 --json title" || gate.req.Prefix != "gh issue view" {
+		t.Fatalf("trust request = %+v, want bash command prefix request", gate.req)
+	}
+}
+
+func TestPlanModeTrustedUnknownBashCommandReachesGateAsReadOnly(t *testing.T) {
+	reg := tool.NewRegistry()
+	reg.Add(fakeTool{name: "bash", readOnly: false})
+	trustGate := &fakePlanModeReadOnlyTrustGate{allow: true}
+	permissionGate := &readOnlyRecordingGate{}
+	a := New(nil, reg, NewSession(""), Options{
+		Gate:                      permissionGate,
+		PlanModeReadOnlyTrustGate: trustGate,
+	}, event.Discard)
+	a.SetPlanMode(true)
+
+	out := a.executeOne(context.Background(), provider.ToolCall{Name: "bash", Arguments: string(bashCommandArgs(t, "gh issue view 4572"))})
+	if strings.HasPrefix(out.output, "blocked:") || !strings.Contains(out.output, "done") {
+		t.Fatalf("trusted unknown bash query should run through the normal gate as read-only, got: %q", out.output)
+	}
+	if len(permissionGate.readOnly) != 1 || !permissionGate.readOnly[0] {
+		t.Fatalf("permission gate readOnly calls = %v, want [true]", permissionGate.readOnly)
+	}
+}
+
+func TestPlanModeUnknownBashReadOnlyCommandDeclineBlocks(t *testing.T) {
+	reg := tool.NewRegistry()
+	reg.Add(fakeTool{name: "bash", readOnly: false})
+	gate := &fakePlanModeReadOnlyTrustGate{allow: false, reason: "not read-only"}
+	a := New(nil, reg, NewSession(""), Options{PlanModeReadOnlyTrustGate: gate}, event.Discard)
+	a.SetPlanMode(true)
+
+	out := a.executeOne(context.Background(), provider.ToolCall{Name: "bash", Arguments: string(bashCommandArgs(t, "gh issue view 4572"))})
+	if !strings.Contains(out.output, "not read-only") {
+		t.Fatalf("declined bash trust should block with reason, got: %q", out.output)
+	}
+}
+
+func TestPlanModeUnsafeUnknownBashCommandDoesNotAskForTrust(t *testing.T) {
+	reg := tool.NewRegistry()
+	reg.Add(fakeTool{name: "bash", readOnly: false})
+	gate := &fakePlanModeReadOnlyTrustGate{allow: true}
+	a := New(nil, reg, NewSession(""), Options{PlanModeReadOnlyTrustGate: gate}, event.Discard)
+	a.SetPlanMode(true)
+
+	out := a.executeOne(context.Background(), provider.ToolCall{Name: "bash", Arguments: string(bashCommandArgs(t, "gh issue view 4572 && rm -rf /"))})
+	if !strings.HasPrefix(out.output, "blocked:") {
+		t.Fatalf("unsafe shell syntax should remain blocked, got: %q", out.output)
+	}
+	if gate.calls != 0 {
+		t.Fatalf("trust gate calls = %d, want 0 for unsafe shell syntax", gate.calls)
+	}
+}
+
 // --- planModeBashBlocked tests ---
 
 func TestPlanModeBashBlocked_SafeCommands(t *testing.T) {

internal/agent/preview.go

@@ -6,7 +6,7 @@ import (
 	"strings"
 )
 
-var reTransientUserBlock = regexp.MustCompile(`(?s)^\s*<(?:response-language|reasoning-language|memory-update|background-jobs)>.*?</(?:response-language|reasoning-language|memory-update|background-jobs)>\s*\n?`)
+var reTransientUserBlock = regexp.MustCompile(`(?s)^\s*<(?:response-language|reasoning-language|memory-update|background-jobs|active-goal|hook-context|capability-route)(?:\s+[^>]*)?>.*?</(?:response-language|reasoning-language|memory-update|background-jobs|active-goal|hook-context|capability-route)>\s*\n?`)
 
 const memoryCompilerExecutionOpen = "<memory-compiler-execution>"
 

internal/agent/preview_test.go

@@ -63,6 +63,28 @@ func TestStripTransientUserBlocksUnwrapsMemoryCompilerExecution(t *testing.T) {
 			in:   "just a normal prompt",
 			want: "just a normal prompt",
 		},
+		{
+			name: "hook context prefix is stripped",
+			in:   "<hook-context event=\"SessionStart\">\nLoad conventions.\n</hook-context>\n\nship it",
+			want: "ship it",
+		},
+		{
+			name: "active goal prefix is stripped",
+			in:   "<active-goal>\nFix all bugs\n</active-goal>\n\nfix the auth bug",
+			want: "fix the auth bug",
+		},
+		{
+			name: "active goal after other transient prefixes is stripped",
+			in: "<reasoning-language>\nuse Chinese\n</reasoning-language>\n\n" +
+				"<memory-update>\n- note\n</memory-update>\n\n" +
+				"<active-goal>\nDo X\n</active-goal>\n\nhelp me",
+			want: "help me",
+		},
+		{
+			name: "capability route prefix is stripped",
+			in:   "<capability-route version=\"1\">\nRelevant capabilities:\n- skill:review prefer\n</capability-route>\n\nreview this",
+			want: "review this",
+		},
 	}
 	for _, tc := range cases {
 		t.Run(tc.name, func(t *testing.T) {