refactor(cli,config): remove --fail-on-unverified and agent.host_verification

Both only steered the final-readiness gate, which no longer exists: the flag
chose exit 3 for an unmet readiness verdict and the setting chose whether a
changed turn was held for a check, a review and a sign-off. Neither can do
anything now.

`run` still exits 4 when an untrusted folder refused its edits. A user config
that still carries host_verification loads unchanged; the key is ignored.
Removes the HoldOwedChecks and WaiveOwedChecks plumbing behind them.

docs/CLI.md

@@ -317,7 +317,7 @@ empty when nothing was, and a refusal never changes the exit code. The same
 | `code` | Cause |
 | --- | --- |
 | `permission.unattended` | It needed an approval and nobody could give one. |
-| `permission.untrusted_folder` | The folder is not trusted, so it needed an approval nobody could give. The denial's `remedy` names `reasonix trust --dir <folder>`, which shows what the folder would run before approving; the run exits `4` (`3` under `--fail-on-unverified`) and the result's `unverified_by` carries the code. |
+| `permission.untrusted_folder` | The folder is not trusted, so it needed an approval nobody could give. The denial's `remedy` names `reasonix trust --dir <folder>`, which shows what the folder would run before approving; the run exits `4` and the result's `unverified_by` carries the code. |
 | `permission.read_only` | The session is in `read-only`. |
 | `permission.deny_rule` | A deny rule matched. |
 | `permission.declined` | A person answered no. |
@@ -358,10 +358,9 @@ Exit statuses of `reasonix run`:
 
 | Status | Meaning |
 | --- | --- |
-| `0` | The model finished, including with other refused calls or unmet readiness. |
+| `0` | The model finished, including with other refused calls. |
 | `1` | The run failed: provider, configuration, limit, or cancellation. |
 | `2` | The command line was invalid. |
-| `3` | `--fail-on-unverified` was given and final readiness stayed unmet or the folder's edits were refused for lack of trust. |
 | `4` | The folder is not trusted and its edits or commands were refused, so the work was not done. Trust it with `reasonix trust --dir <folder>` or pass `--permission-mode` knowingly. |
 
 ### Redacted machine interfaces

docs/CLI.zh-CN.md

@@ -269,7 +269,7 @@ reasonix run "运行测试" --output-format stream-json
 }
 ```
 
-`permission_denials` 列出本次运行的权限门拒绝的调用，没有被拒时为空数组；被拒一般不改变退出码；`permission.untrusted_folder` 退出 `4`（指定 `--fail-on-unverified` 时为 `3`）。
+`permission_denials` 列出本次运行的权限门拒绝的调用，没有被拒时为空数组；被拒一般不改变退出码；`permission.untrusted_folder` 退出 `4`。
 同一个 `code` 也随被拒的工具结果出现：`stream-json` 里是 `refusalCode`，`--events-jsonl`
 里是 `refusal_code`。
 
@@ -316,7 +316,6 @@ reasonix run "运行测试" --output-format stream-json
 | `0` | 模型已结束，包括有调用被拒或就绪检查未满足的情况。 |
 | `1` | 运行失败：模型服务、配置、上限或取消。 |
 | `2` | 命令行参数无效。 |
-| `3` | 指定了 `--fail-on-unverified`，且最终就绪检查未满足，或因文件夹未被信任而编辑和命令被拒绝。 |
 | `4` | 文件夹未被信任，编辑和命令被拒绝，任务没有完成。用 `reasonix trust --dir <folder>` 信任它，或明知故犯地传 `--permission-mode`。 |
 
 ### 脱敏机器接口

docs/GUIDE.md

@@ -83,7 +83,6 @@ reasoning_language = "auto"      # visible reasoning text: auto|zh|en
 # subagent_claims = "strict"          # user file only; default optimistic: a sub-agent that declares no write_paths holds only the path it is writing; strict makes it hold the whole workspace
 # plan_approval = "auto"             # user file only; a submitted plan starts executing without the approval click, and the transcript says so; default ask
 # default_mode = "plan"              # user file only; a new conversation starts in plan mode (a resumed one keeps its own; Settings has a switch); default normal
-# host_verification = "on"           # user file only; "on" holds a turn that changed files open until a check, a review and a sign-off exist; unset or "off" does not (unset still holds them under the delivery preset); the receipt always says what was checked
 # compact_ratio = 0.85             # sole auto trigger; presets 0.70 / 0.80 / 0.85
 # max_output_tokens = 0            # recommended: automatic (DeepSeek default high → ~64K; not unlimited)
 # max_output_tokens = 32768        # ordinary coding / cost control

docs/GUIDE.zh-CN.md

@@ -92,7 +92,6 @@ reasoning_language = "auto"      # 可见思考过程语言：auto|zh|en
 # plan_approval = "auto"             # 仅用户配置文件；提交的计划不再等待批准点击，直接开始执行，对话里会留下说明，默认 ask
 # default_mode = "plan"              # 仅用户配置文件；新对话从计划模式开始（恢复的对话保持自己的模式；设置里有对应开关），默认 normal
 # subagent_claims = "strict"          # 仅用户配置文件；默认 optimistic：没有声明 write_paths 的子代理只占用正在写的路径；strict 则让它占整个工作区
-# host_verification = "on"           # 仅用户配置文件；设为 on 时，改过文件的回合要先有检查、复核和签收才能结束；不设或设为 off 则不要求（不设时交付预设仍会要求）；交付回执始终如实写明检查了什么
 # compact_ratio 是唯一自动维护阈值（默认 0.85；预设 0.70/0.80/0.85）
 # max_output_tokens = 0            # 推荐：自动（DeepSeek 默认 high → 约 64K；不是无限）
 # max_output_tokens = 32768        # 普通编码 / 控制费用

docs/MIGRATING.md

@@ -85,7 +85,7 @@ When a folder is trustable, its writes are sandbox-confined and no decision is r
 1.x wrote files and ran commands in any folder. In 2.x an untrusted folder has nobody to approve edits and shell commands, so a headless run refuses them. Nothing is loosened: the run now says so instead of looking like success.
 
 - stderr names the refused tools, the code `permission.untrusted_folder` and the remedy.
-- The process exits `4` (`3` under `--fail-on-unverified`); the work was not done.
+- The process exits `4`; the work was not done.
 - `--output-format json` and `stream-json` list each refusal in `permission_denials` with its `code` and `remedy`, and set `unverified_by`.
 
 To make a script behave as it did in 1.x, trust the folder once (`reasonix trust --dir <folder>`), or pass `--permission-mode auto` or `--yolo` knowingly for that run. Other exit codes are as in 1.x: `1` for an error, `2` for a usage error.

internal/assembly/boot/build.go

@@ -539,7 +539,6 @@ func (b *builder) executor() *agent.Agent {
 		ProjectChecks: b.prompt.projectChecks, ProjectSensitivePaths: b.prompt.sensitivePaths,
 		AgentPreset:       b.model.preset,
 		DeliveryProfile:   b.model.delivery,
-		WaiveOwedChecks:   cfg.WaivesOwedChecks(b.model.delivery) && !b.opts.HoldOwedChecks,
 		Ablation:          b.opts.Ablation,
 		WorkspaceLease:    b.session.lease,
 		CapabilityLedger:  t.caps.ledger,

internal/assembly/boot/effect_host_verification_test.go

@@ -1,7 +0,0 @@
-package boot
-
-import ()
-
-// holdChangedTurns is the user-level opt-in that makes a turn which changed files
-// owe its check, review and sign-off; tests of what a turn owes start from it.
-const holdChangedTurns = "\n[agent]\nhost_verification = \"on\"\n"

internal/assembly/boot/effect_plan_check_effects_test.go

@@ -80,7 +80,7 @@ func runPlannedChecks(t *testing.T, files map[string]string, planArgs string, ca
 	isolateConfigHome(t)
 	// The subject is what the checks write, not the jail: a host without an OS
 	// sandbox refuses bash fail-closed, and the checks would never run.
-	writeUserConfig(t, holdChangedTurns+"\n[sandbox]\nbash = \"off\"\n")
+	writeUserConfig(t, "[sandbox]\nbash = \"off\"\n")
 	dir := robustTempDir(t)
 	t.Chdir(dir)
 	// A declared project check makes the turn answer to readiness in the

internal/assembly/boot/effect_script_test.go

@@ -24,7 +24,6 @@ print("rewrote")
 func TestEffectRunScriptReadEditWriteKeepsTheBytes(t *testing.T) {
 	t.Parallel()
 	home := statedBootHome(t)
-	writeFile(t, home, "config.toml", holdChangedTurns)
 	dir := testenv.TempDir(t)
 	big := strings.Repeat("// filler line that pads the file past one tool result\n", 1200) + "func OldName() {}\n"
 	files := map[string]string{

internal/assembly/boot/options.go

@@ -139,7 +139,6 @@ type Options struct {
 	Observe *ObserveOptions
 
 	GoalTurnsUnreachable bool // this assembly never arms a Goal turn; see GoalOnlyToolNames
-	HoldOwedChecks       bool // the caller asked for a verified result, so a changed turn owes its check whatever host_verification says
 	UnattendedChild      bool // an unattended child run: it offers neither best_of_n nor isolation
 	// SessionRecoveryMeta and OnSessionRecovered let richer frontends attach
 	// local UI metadata to automatic transcript recovery branches.

internal/contract/config/config.go

@@ -1137,12 +1137,11 @@ type AgentConfig struct {
 	// declare non-overlapping write_paths. 0, the default, sets no limit: the
 	// path claims keep writers apart, not a count. The user file only.
 	MaxParallelWriters int    `toml:"max_parallel_writers"`
-	SubagentClaims     string `toml:"subagent_claims"`   // "optimistic" (default) or "strict"; the user file only
-	WriteLease         string `toml:"write_lease"`       // "optimistic" (default) or "strict"; the user file only
-	PlanApproval       string `toml:"plan_approval"`     // "ask" (default) or "auto": the plan card answers itself; the user file only
-	DefaultMode        string `toml:"default_mode"`      // "normal" (default) or "plan": how a new conversation starts; the user file only
-	GoalAutoRetry      int    `toml:"goal_auto_retry"`   // 0 (default) stops at an unavailable completion check; N retries it up to N times in a row; the user file only
-	HostVerification   string `toml:"host_verification"` // unset (the default), "on" or "off"; the user file only
+	SubagentClaims     string `toml:"subagent_claims"` // "optimistic" (default) or "strict"; the user file only
+	WriteLease         string `toml:"write_lease"`     // "optimistic" (default) or "strict"; the user file only
+	PlanApproval       string `toml:"plan_approval"`   // "ask" (default) or "auto": the plan card answers itself; the user file only
+	DefaultMode        string `toml:"default_mode"`    // "normal" (default) or "plan": how a new conversation starts; the user file only
+	GoalAutoRetry      int    `toml:"goal_auto_retry"` // 0 (default) stops at an unavailable completion check; N retries it up to N times in a row; the user file only
 
 	// WriteLeaseMaxWaitMinutes bounds one wait behind another session's write
 	// claim; 0, the default, waits for as long as the turn lasts. The user file only.

internal/contract/config/host_verification.go

@@ -1,21 +0,0 @@
-package config
-
-import "strings"
-
-// WaivesOwedChecks reports whether the host drops the check, review and sign-off
-// a changed turn owes. Unset drops them except under the Delivery preset; "off"
-// and "on" decide under any preset; any other value reads as "on", so a typo
-// never weakens what the host asks.
-func (c *Config) WaivesOwedChecks(delivery bool) bool {
-	if c == nil {
-		return !delivery
-	}
-	switch strings.ToLower(strings.TrimSpace(c.Agent.HostVerification)) {
-	case "":
-		return !delivery
-	case "off":
-		return true
-	default:
-		return false
-	}
-}

internal/contract/config/host_verification_test.go

@@ -1,63 +0,0 @@
-package config
-
-import (
-	"slices"
-	"strings"
-	"testing"
-)
-
-func TestOwedChecksAreWaivedUnlessTheUserOrTheDeliveryPresetHoldsThem(t *testing.T) {
-	for _, tc := range []struct {
-		value    string
-		delivery bool
-		want     bool
-	}{
-		{"", false, true}, {"", true, false},
-		{"on", false, false}, {"on", true, false},
-		{"off", false, true}, {"off", true, true}, {"Off", false, true}, {" off ", true, true},
-		{"false", false, false}, {"nonsense", false, false}, {"nonsense", true, false},
-	} {
-		cfg := Default()
-		cfg.Agent.HostVerification = tc.value
-		if got := cfg.WaivesOwedChecks(tc.delivery); got != tc.want {
-			t.Errorf("host_verification = %q, delivery %v -> waived %v, want %v", tc.value, tc.delivery, got, tc.want)
-		}
-	}
-}
-
-func TestUserFileTurnsHostVerificationOff(t *testing.T) {
-	cfg, _ := loadScoped(t, "[agent]\nhost_verification = \"off\"\n", "")
-	if !cfg.WaivesOwedChecks(false) {
-		t.Fatalf("a user file's off was not honoured: %q", cfg.Agent.HostVerification)
-	}
-}
-
-func TestProjectCannotChangeHostVerificationEitherWay(t *testing.T) {
-	for _, value := range []string{"on", "off"} {
-		cfg, _ := loadScoped(t, "", "[agent]\nhost_verification = \""+value+"\"\n")
-		if !cfg.WaivesOwedChecks(false) || cfg.WaivesOwedChecks(true) {
-			t.Fatalf("a cloned repo's reasonix.toml (%q) moved what the host asks of a turn: waived %v, under delivery %v", value, cfg.WaivesOwedChecks(false), cfg.WaivesOwedChecks(true))
-		}
-		if !slices.Contains(ignoredKeys(cfg), "agent.host_verification") {
-			t.Fatalf("ignored = %v, want agent.host_verification reported", ignoredKeys(cfg))
-		}
-	}
-}
-
-func TestProjectCannotTurnBackOnWhatTheUserTurnedOff(t *testing.T) {
-	cfg, _ := loadScoped(t, "[agent]\nhost_verification = \"off\"\n", "[agent]\nhost_verification = \"on\"\n")
-	if !cfg.WaivesOwedChecks(false) {
-		t.Fatal("the project file overrode the user's own host_verification")
-	}
-}
-
-func TestHostVerificationOffSurvivesASave(t *testing.T) {
-	cfg := Default()
-	cfg.Agent.HostVerification = "off"
-	if out := RenderTOMLForScope(cfg, RenderScopeUser); !strings.Contains(out, "host_verification = \"off\"") {
-		t.Fatalf("rendered config lost the choice:\n%s", out)
-	}
-	if out := RenderTOMLForScope(Default(), RenderScopeUser); strings.Contains(out, "\nhost_verification = ") {
-		t.Fatalf("the default was written as an active key:\n%s", out)
-	}
-}

internal/contract/config/project_scope.go

@@ -31,7 +31,6 @@ type heldScope struct {
 	limits         agentLimits
 	goalAutoRetry  int
 	subagentClaims string
-	hostVerif      string
 	shell          ShellConfig
 	rgPath         string
 	lsp            map[string]LSPServer
@@ -92,7 +91,6 @@ func holdUserScope(c *Config) heldScope {
 		limits:          holdAgentLimits(c),
 		goalAutoRetry:   c.Agent.GoalAutoRetry,
 		subagentClaims:  c.Agent.SubagentClaims,
-		hostVerif:       c.Agent.HostVerification,
 		shell:           shell,
 		rgPath:          c.Tools.Search.RgPath,
 		lsp:             maps.Clone(c.LSP.Servers),
@@ -259,12 +257,6 @@ func (h heldScope) narrow(c *Config, r Roots, root string, projectMeta toml.Meta
 		c.ignoreProject("agent.subagent_claims", c.Agent.SubagentClaims, ProjectUserOnly)
 		c.Agent.SubagentClaims = h.subagentClaims
 	}
-	// What the host asks of a turn before it may finish is the user's own call,
-	// so a cloned repo can neither switch it off nor undo the user's choice.
-	if c.Agent.HostVerification != h.hostVerif {
-		c.ignoreProject("agent.host_verification", c.Agent.HostVerification, ProjectUserOnly)
-		c.Agent.HostVerification = h.hostVerif
-	}
 	// Keeping paired phones across restarts is the user's alone: it widens how
 	// long a device stays trusted, which is not a clone's to give away.
 	if c.Serve.RememberPairedDevices != h.rememberDevices {

internal/contract/config/render.go

@@ -211,7 +211,6 @@ func RenderTOMLForScope(c *Config, scope RenderScope) string {
 	agentSetting(&b, "write_lease_max_wait_minutes", c.Agent.WriteLeaseMaxWaitMinutes, defaults.Agent.WriteLeaseMaxWaitMinutes, "minutes a write waits behind another session's claim before it fails with workspace.claim_timeout; 0 = wait as long as the turn lasts (default), 120 is a sensible limit; user file only")
 	agentSetting(&b, "plan_approval", c.Agent.PlanApproval, defaults.Agent.PlanApproval, "auto: the plan card answers itself and execution starts; user file only")
 	agentSetting(&b, "default_mode", c.Agent.DefaultMode, defaults.Agent.DefaultMode, "plan: a new conversation starts in plan mode (the mode picker still switches it); a resumed one keeps its own; user file only")
-	agentSetting(&b, "host_verification", c.Agent.HostVerification, defaults.Agent.HostVerification, "on: the host holds a turn that changed files open until a check, a review and a sign-off exist; unset or off: it does not (unset still holds them under the delivery preset); the receipt always says what was and was not checked; user file only")
 	if c.Agent.OutputStyle != "" {
 		fmt.Fprintf(&b, "output_style = %q   # persona/tone folded into the prompt\n", c.Agent.OutputStyle)
 	} else {

internal/contract/settings/declare.go

@@ -141,7 +141,6 @@ var table = []Entry{
 	{ID: "agent.plan_approval", Class: Guarded, Effect: OnOpen},
 	{ID: "agent.default_mode", Class: Preference, Effect: OnOpen},
 	{ID: "agent.goal_auto_retry", Class: Guarded, Effect: OnOpen},
-	{ID: "agent.host_verification", Class: Guarded, Effect: OnOpen},
 	{ID: "agent.output_style", Class: Preference, Effect: OnOpen},
 	{ID: "agent.auto_plan", Class: Preference, Effect: OnOpen},
 	{ID: "agent.reasoning_language", Class: Preference, Effect: OnOpen},

internal/contract/settings/settings_test.go

@@ -61,7 +61,7 @@ func TestBoundaryKeysAreGuarded(t *testing.T) {
 		}
 	}
 	for _, id := range []string{
-		"desktop.default_tool_approval_mode", "agent.plan_approval", "agent.host_verification", "agent.subagent_claims",
+		"desktop.default_tool_approval_mode", "agent.plan_approval", "agent.subagent_claims",
 		"agent.max_steps", "agent.max_subagent_depth", "agent.max_subagent_concurrency", "agent.max_parallel_writers",
 		"agent.task_token_budget", "agent.guardian_model", "agent.system_prompt", "agent.screen_external_content",
 		"tools.shell.path", "tools.enabled", "statusline.command", "skills.paths", "providers", "plugins",

internal/frontend/cli/build_options.go

@@ -30,7 +30,6 @@ type cliBuildOverrides struct {
 	WorkspaceRoot        string
 	HeadlessApprovalMode string
 	GoalTurnsUnreachable bool
-	HoldOwedChecks       bool
 	Stderr               io.Writer
 	OnSessionRecovered   func(control.SessionRecoveryInfo) error
 	Ablation             ablation.Set
@@ -78,7 +77,6 @@ func cliProfileBuildOptions(modelName string, maxStepsOverride int, requireKey b
 		AdditionalDirs:       overrides.AdditionalDirs,
 		HeadlessApprovalMode: overrides.HeadlessApprovalMode,
 		GoalTurnsUnreachable: overrides.GoalTurnsUnreachable,
-		HoldOwedChecks:       overrides.HoldOwedChecks,
 		StatsSource:          surface.CLI,
 		FeedbackSurface:      feedback.SurfaceTUI,
 		Stderr:               overrides.Stderr,
@@ -95,7 +93,7 @@ func cliProfileBuildOptions(modelName string, maxStepsOverride int, requireKey b
 // exits, so nothing in this path can reach SetGoal and goal-only tools are left
 // out of the provider schema.
 func runBuildOverrides(effort *string, allow, dirs []string, workspaceRoot, approval string,
-	onRecovered func(control.SessionRecoveryInfo) error, ablated ablation.Set, holdOwedChecks bool) cliBuildOverrides {
+	onRecovered func(control.SessionRecoveryInfo) error, ablated ablation.Set) cliBuildOverrides {
 	return cliBuildOverrides{
 		Effort:               effort,
 		PermissionAllow:      allow,
@@ -105,6 +103,5 @@ func runBuildOverrides(effort *string, allow, dirs []string, workspaceRoot, appr
 		GoalTurnsUnreachable: true,
 		OnSessionRecovered:   onRecovered,
 		Ablation:             ablated,
-		HoldOwedChecks:       holdOwedChecks,
 	}
 }

internal/frontend/cli/cli.go

@@ -501,7 +501,7 @@ func runAgent(args []string, version string) int {
 	// executor, not just the top-level one. Ask fails closed because no UI can
 	// answer; with no mode named, the build's sandbox claim and the folder's
 	// trust decide between that and auto once the controller exists.
-	overrides := runBuildOverrides(effortOverride, allowedTools, f.additionalDirs, workspaceRoot, permissions.approval, cliSessionRecoveredHandler(leases), ablated, *f.failOnUnverified)
+	overrides := runBuildOverrides(effortOverride, allowedTools, f.additionalDirs, workspaceRoot, permissions.approval, cliSessionRecoveredHandler(leases), ablated)
 	overrides.Version = version
 	overrides = withComposition(overrides, *f.agentComposition, resumeSession)
 	ctrl, err := setupProfileWithOverrides(ctx, *f.model, *f.maxSteps, true, sink, profile, overrides)
@@ -527,7 +527,7 @@ func runAgent(args []string, version string) int {
 	chain.begin(ctrl, version, prompt)
 	runErr := ctrl.Run(ctx, prompt)
 	reporter.RecordRecovery(ctrl.DrainRecoveryMetrics())
-	completion := classifyRunCompletion(runErr).withFolderRefusal(chain.refusedByFolderTrust(os.Stderr)).withFailOnUnverified(*f.failOnUnverified)
+	completion := classifyRunCompletion(runErr).withFolderRefusal(chain.refusedByFolderTrust(os.Stderr))
 	if cfg != nil {
 		notify.SendEvent(newNotificationSender(), i18n.M, cfg.Notifications, event.Event{
 			Kind:    event.TurnDone,

internal/frontend/cli/run_completion_test.go

@@ -27,16 +27,10 @@ func TestClassifyRunCompletionNamesTheFailureClass(t *testing.T) {
 }
 
 // An unmet final-readiness judgement is a verdict on a finished run: it keeps
-// its benchmark class but exits 0 unless --fail-on-unverified asks for 3.
+// its benchmark class and exits 0.
 func TestFinalReadinessIsAVerdictNotAFailure(t *testing.T) {
 	readiness := classifyRunCompletion(&agent.FinalReadinessError{Attempts: 3})
 	if readiness.isError || readiness.exitCode != 0 || !readiness.unverified || readiness.subtype != "success" {
 		t.Fatalf("final-readiness completion = %+v, want a successful, unverified run", readiness)
 	}
-	if got := readiness.withFailOnUnverified(true).exitCode; got != runExitUnverified {
-		t.Fatalf("--fail-on-unverified exit = %d, want %d", got, runExitUnverified)
-	}
-	if got := classifyRunCompletion(errors.New("provider 500")).withFailOnUnverified(true).exitCode; got != 1 {
-		t.Fatalf("--fail-on-unverified changed a run error's exit to %d", got)
-	}
 }

internal/frontend/cli/run_flags.go

@@ -17,7 +17,6 @@ type runFlags struct {
 	takeover                          *bool
 	autoApprove, yolo                 *bool
 	printOnly, eventsJSONL            *bool
-	failOnUnverified                  *bool
 	additionalDirs, allowedToolValues []string
 }
 
@@ -45,7 +44,6 @@ func newRunFlags() *runFlags {
 	f.permissionMode = fs.String("permission-mode", "", "permission mode: read-only | manual | ask | auto | acceptEdits | dontAsk | bypassPermissions (default: auto where the OS sandbox confines writes and the folder is trusted, otherwise ask)")
 	f.autoApprove, f.yolo = registerRunApprovalFlags(fs)
 	f.printOnly = fs.BoolP("print", "p", false, "print only the final response")
-	f.failOnUnverified = fs.Bool("fail-on-unverified", false, "exit 3 when the model finished but the host could not verify the result (final readiness unmet, or edits and commands refused because the folder is not trusted)")
 	f.eventsJSONL = fs.Bool("events-jsonl", false, "emit a redacted structured event stream as JSONL")
 	f.outputFormat = fs.String("output-format", "text", "output format: text | json | stream-json")
 	fs.StringArrayVar(&f.additionalDirs, "add-dir", nil, "allow tool access to an additional directory (repeatable)")

internal/frontend/cli/run_output.go

@@ -98,14 +98,10 @@ type runResult struct {
 	// refusal's code; a run that wrote nothing says why here, not only in prose.
 	PermissionDenials []runPermissionDenial `json:"permission_denials"`
 	// PermissionMode is the posture the run settled on, named or defaulted.
-	PermissionMode string `json:"permission_mode,omitempty"`
-	// Readiness is the unmet final-readiness judgement, if any; the run still
-	// exits 0 unless --fail-on-unverified asked otherwise.
-	Readiness *eventwire.FinalReadiness `json:"readiness,omitempty"`
-	// UnverifiedBy is the refusal code that left the work undone when the run
-	// still reports success; `--fail-on-unverified` turns it into exit 3.
-	UnverifiedBy string                       `json:"unverified_by,omitempty"`
-	Completion   *eventwire.CompletionSummary `json:"completion,omitempty"`
+	PermissionMode string                       `json:"permission_mode,omitempty"`
+	Readiness      *eventwire.FinalReadiness    `json:"readiness,omitempty"`
+	UnverifiedBy   string                       `json:"unverified_by,omitempty"`
+	Completion     *eventwire.CompletionSummary `json:"completion,omitempty"`
 }
 
 type machineEventUsage struct {

internal/frontend/cli/run_untrusted_folder_test.go

@@ -76,7 +76,7 @@ func (r *untrustedFolderRun) modelSaw() string {
 
 // In a folder nobody trusted, a headless run that could not write says so on
 // stderr with the refusal's identity and the remedy, tells the model the same
-// cause, and exits 4 (3 under --fail-on-unverified); trusting the folder lets the
+// cause, and exits 4; trusting the folder lets the
 // identical command write.
 func TestRunInUntrustedFolderSaysWhyNothingWasDone(t *testing.T) {
 	r := newUntrustedFolderRun(t)
@@ -98,11 +98,6 @@ func TestRunInUntrustedFolderSaysWhyNothingWasDone(t *testing.T) {
 		}
 	}
 
-	_, _ = captureCLIOutput(t, func() { code = Run([]string{"run", "--fail-on-unverified", "write notes.md"}, "test") })
-	if code != runExitUnverified {
-		t.Fatalf("--fail-on-unverified exited %d, want %d: every write was refused by project trust", code, runExitUnverified)
-	}
-
 	stdout, _ := captureCLIOutput(t, func() { code = Run([]string{"-p", "--output-format", "json", "write notes.md"}, "test") })
 	var result struct {
 		Denials      []runPermissionDenial `json:"permission_denials"`

internal/frontend/cli/run_verdict.go

@@ -15,14 +15,9 @@ import (
 	"reasonix/internal/session/control"
 )
 
-// runExitUnverified is `run --fail-on-unverified`'s exit status when the model
-// finished but the host's final-readiness judgement was left unmet. It is
-// apart from 1 (run error) and 2 (usage) so automation can tell them apart.
-const runExitUnverified = 3
-
 // runExitUntrustedFolder is the exit status of a run whose edits or commands
 // were refused because the folder is untrusted, so the work was not done. It
-// is apart from 1, 2 and 3; --fail-on-unverified keeps reporting 3 for it.
+// is apart from 1 and 2.
 const runExitUntrustedFolder = 4
 
 // runVerdict is what the host decided about the run beside its answer.
@@ -136,15 +131,6 @@ func runReadiness(runErr error) *eventwire.FinalReadiness {
 	return &eventwire.FinalReadiness{Attempts: readiness.Attempts, Missing: append([]string(nil), readiness.Missing...)}
 }
 
-// withFailOnUnverified applies --fail-on-unverified to the exit status only;
-// the machine-readable result reports the same verdict either way.
-func (c runCompletion) withFailOnUnverified(on bool) runCompletion {
-	if on && c.unverified {
-		c.exitCode = runExitUnverified
-	}
-	return c
-}
-
 // withFolderRefusal marks a run whose edits and commands were refused because
 // the folder is untrusted as unverified: whatever the model then said, the work
 // it was asked for did not happen.

internal/frontend/cli/run_verdict_test.go

@@ -49,7 +49,7 @@ func TestRunProseOnlyFinishHasNoReadinessDebt(t *testing.T) {
 	if err := os.Remove(filepath.Join(dir, "note.txt")); err != nil {
 		t.Fatal(err)
 	}
-	code, result, stderr := runCLIJSONResult(t, "run", "-y", "--fail-on-unverified", "--output-format", "json", "write a note")
+	code, result, stderr := runCLIJSONResult(t, "run", "-y", "--output-format", "json", "write a note")
 	if code != 0 || result["is_error"] != false || result["readiness"] != nil {
 		t.Fatalf("prose finish exited %d with %v; stderr: %s", code, result, stderr)
 	}

internal/frontend/cli/shell_completion.go

@@ -109,7 +109,7 @@ func cliCompletionRootSpec() cliCompletionSpec {
 		effort, permissionMode,
 		completionFlag("--auto -y", cliCompletionNoValue), completionFlag("--dangerously-skip-permissions --yolo", cliCompletionNoValue),
 		completionFlag("--print -p", cliCompletionNoValue),
-		completionFlag("--events-jsonl", cliCompletionNoValue), completionFlag("--fail-on-unverified", cliCompletionNoValue),
+		completionFlag("--events-jsonl", cliCompletionNoValue),
 		completionFlag("--output-format", cliCompletionStaticValue, "text", "json", "stream-json"),
 		completionFlag("--add-dir", cliCompletionPathValue),
 		completionFlag("--allowed-tools --allowedTools", cliCompletionStaticValue),

internal/frontend/serve/trace_parity_test.go

@@ -95,15 +95,6 @@ func runTrace(t *testing.T, tc traceCase) traceReading {
 	if err := os.WriteFile(filepath.Join(dir, "reasonix.toml"), []byte(toml), 0o600); err != nil {
 		t.Fatalf("write config: %v", err)
 	}
-	// A turn that changed files owes its check only when the user asks the host to
-	// hold it, and these readings include what that hold emits.
-	userConfig := config.RootsForHome(dir).UserConfigPath()
-	if err := os.MkdirAll(filepath.Dir(userConfig), 0o700); err != nil {
-		t.Fatal(err)
-	}
-	if err := os.WriteFile(userConfig, []byte("[agent]\nhost_verification = \"on\"\n"), 0o600); err != nil {
-		t.Fatalf("write user config: %v", err)
-	}
 	// The home this run states is where the person's approval would be.
 	if _, err := config.RootsForHome(dir).ApproveWorkspacePrograms(dir); err != nil {
 		t.Fatal(err)

internal/runtime/agent/agent_options.go

@@ -167,11 +167,6 @@ type Options struct {
 	// AgentPreset is empty for one compatibility version of direct constructors.
 	DeliveryProfile bool
 
-	// WaiveOwedChecks is the user's choice to drop the check, review and
-	// sign-off a turn that changed files owes before it may finish. It never
-	// comes from a project file.
-	WaiveOwedChecks bool
-
 	// AgentPreset is the session role setting (balanced|delivery). Empty
 	// falls back to balanced unless DeliveryProfile is true (then delivery).
 	// Switching the preset mid-session does not rebuild the agent; the value is

internal/runtime/agent/agent_role.go

@@ -16,16 +16,11 @@ type agentRole struct {
 	// non-destructive MCP targets without readOnlyHint. Writers, bash and
 	// destructive MCP stay blocked, and read-only sub-agents leave it false.
 	plannerMCPExecution bool
-	// waiveOwedChecks is the user's choice to drop the check, review and
-	// sign-off a turn that changed files owes before it may finish. The turn's
-	// own unfinished plan and a capability the user required still hold it.
-	waiveOwedChecks bool
 }
 
 func newAgentRole(opts Options) agentRole {
 	return agentRole{
 		readOnlyExecution:   opts.ReadOnlyExecution,
 		plannerMCPExecution: opts.PlannerMCPExecution,
-		waiveOwedChecks:     opts.WaiveOwedChecks,
 	}
 }

tools/repolint/baseline.json

@@ -13,10 +13,10 @@
     "doc-prose": 2128,
     "error-text": 9,
     "essay": 2579,
-    "file-size": 18429,
+    "file-size": 18427,
     "flat-view": 0,
     "frontend-parity": 13,
-    "function-size": 2486,
+    "function-size": 2485,
     "layer-order": 0,
     "layering": 0,
     "marker": 0,
@@ -265,7 +265,7 @@
     },
     "internal/contract/config/config.go": {
       "essay": 30,
-      "file-size": 846,
+      "file-size": 845,
       "process-input": 3
     },
     "internal/contract/config/credentials.go": {
@@ -343,8 +343,8 @@
     "internal/contract/config/render.go": {
       "complexity": 86,
       "essay": 1,
-      "file-size": 273,
-      "function-size": 375
+      "file-size": 272,
+      "function-size": 374
     },
     "internal/contract/config/vision.go": {
       "essay": 1,