diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 83f40fdf75..6c9445a2ed 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -452,6 +452,41 @@ jobs:
           if-no-files-found: ignore
           retention-days: 7
 
+      - name: test (Windows bounded root budget)
+        if: env.RUN_STEPS == 'true' && runner.os == 'Windows' && github.event_name == 'pull_request'
+        timeout-minutes: 5
+        shell: bash
+        run: |
+          set -o pipefail
+          go test -count=1 -timeout=3m -json -run '^Test(AuxiliaryUsageObserver|BoundedUsageObserver|RootBounded|EffectRootBounded)' ./internal/contract/provider ./internal/model/boundedllm ./internal/runtime/agent ./internal/assembly/boot | tee "$RUNNER_TEMP/bounded-root-budget.jsonl"
+          jq -se '
+            [.[] | select(.Test == null and .Action == "pass") | .Package] as $packages |
+            [.[] | select(.Test != null and .Action == "pass") | .Test] as $tests |
+            all(.[]; .Action != "skip" and .Action != "fail") and
+            (["reasonix/internal/contract/provider", "reasonix/internal/model/boundedllm",
+              "reasonix/internal/runtime/agent", "reasonix/internal/assembly/boot"] - $packages | length) == 0 and
+            (["TestAuxiliaryUsageObserverReplacesAndCopies",
+              "TestBoundedUsageObserverOwnsOneTerminal", "TestBoundedUsageObserverOwnsOneTerminal/reported",
+              "TestBoundedUsageObserverOwnsOneTerminal/billed_failure", "TestBoundedUsageObserverOwnsOneTerminal/without_sink",
+              "TestBoundedUsageObserverOwnsOneTerminal/missing", "TestBoundedUsageObserverOwnsOneTerminal/input_refused",
+              "TestRootBoundedReviewerBudget", "TestRootBoundedReviewerBudget/reported",
+              "TestRootBoundedReviewerBudget/billed_failure", "TestRootBoundedReviewerBudget/missing_usage",
+              "TestRootBoundedReviewerBudget/disabled", "TestRootBoundedReviewerBudget/no_emission",
+              "TestRootBoundedReviewerBudget/projected_job_context", "TestRootBoundedNestedBudgetChargesOnce",
+              "TestRootBoundedLateProjectionStaysWithOldTask", "TestEffectRootBoundedAdvisorBudget",
+              "TestEffectRootBoundedAdvisorBudget/reported", "TestEffectRootBoundedAdvisorBudget/disabled",
+              "TestEffectRootBoundedAdvisorBudget/missing_usage"] - $tests | length) == 0
+          ' "$RUNNER_TEMP/bounded-root-budget.jsonl"
+
+      - name: upload Windows bounded root budget inventory
+        if: always() && env.RUN_STEPS == 'true' && runner.os == 'Windows' && github.event_name == 'pull_request'
+        uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7
+        with:
+          name: bounded-root-budget-windows-${{ github.run_attempt }}
+          path: ${{ runner.temp }}/bounded-root-budget.jsonl
+          if-no-files-found: ignore
+          retention-days: 7
+
       # The general Windows PR smoke list intentionally omits internal/contract/tool.
       # Keep the session-temp portability contract covered without widening the
       # leg to every tool test: two real PowerShell launches must share the
diff --git a/docs/BOUNDED_ROOT_TOKEN_BUDGET.md b/docs/BOUNDED_ROOT_TOKEN_BUDGET.md
new file mode 100644
index 0000000000..bd30570f0b
--- /dev/null
+++ b/docs/BOUNDED_ROOT_TOKEN_BUDGET.md
@@ -0,0 +1,105 @@
+---
+owner: @SivanCola
+backup: @esengine
+status: active
+reviewed: 2026-10-26
+---
+
+# Bounded calls in root token budgets / 根 token 预算中的有界调用
+
+An active Agent root now receives usage from bounded reviewer/advisor calls
+through one optional internal Go context observer. Bounded calls already emit
+their own usage; they previously bypassed the root token counter.
+
+活跃 Agent 根任务现在通过单个可选的内部 Go context observer 接收有界 reviewer/
+advisor 调用的用量。此前这些调用已有自己的用量事件，却绕过了根 token 计数。
+
+## Ownership / 所有权
+
+Agent root binding installs the observer, and the budget-only job projection
+copies it with the opaque scope. A child replaces the inherited observer rather
+than stacking callbacks, so one logical bounded terminal charges the root once.
+
+Agent 根绑定安装 observer；仅投射预算的 job context 将其与不透明 scope 一起携带。
+子任务替换继承 observer，不叠加回调，因此一个逻辑有界调用终态只计一次。
+
+The callback captures the original counter, not the Agent's mutable current-task
+field. Late projected calls remain charged to the old task after a replacement.
+Job cancellation and deadlines remain owned by the target job context.
+
+回调捕获原 counter，不读取 Agent 可变的当前 task 字段。替换任务后的延后调用
+仍计入旧任务。job 的取消与 deadline 继续由目标 job context 掌握。
+
+The provider contract adds two internal Go helpers: one replaces the observer,
+and one reports a copy of finalized usage. An observer cannot mutate the record
+that the existing usage sink receives. Callbacks are trusted host code and own
+their synchronization; this is not a callback panic-isolation boundary.
+
+The helpers require a non-nil context, as usual in Go. The bounded owner retains
+its existing context normalization before it calls them.
+
+provider 契约新增两个内部 Go helper：一个替换 observer，一个上报终态用量副本。
+observer 不能改写既有 usage sink 收到的记录。回调属于可信 host 代码，自行负责
+同步；本机制不是回调 panic 隔离边界。
+
+helper 要求非 nil context；有界 owner 调用前保留既有 context 规范化。
+
+## Accounting and limits / 计账与阈值
+
+The bounded owner reports its finalized usage once, including usage received
+before a stream error. Input refusal without a request does not charge it.
+Accounting does not depend on whether the optional usage sink/source is set.
+
+有界调用 owner 仅上报一次终态用量，包括流错误前收到的用量。请求前的输入拒绝
+不会扣计。计账不依赖可选 usage sink/source 是否启用。
+
+Positive prompt/completion counts follow the existing saturating root rule,
+including positive estimates. Missing counts remain unknown, not inferred from
+request counts. Normal-round local telemetry and cost/currency rules stay as
+they were; this does not add a monetary ledger.
+
+正 prompt/completion 计数沿用既有根饱和规则，包括正估算值。缺失计数仍未知，
+不能由请求次数推导。正常轮次的局部遥测与费用/币种规则保持原样，不新增费用账本。
+
+The existing cooperative check runs after a tool round and keeps finalization.
+Explicit limits can land earlier after an advisor call; zero still means
+unbounded. There is no request reservation, provider cancellation or process
+termination in this change.
+
+既有协作式检查仍在工具轮后执行，并保留收尾轮。显式阈值可能在 advisor 调用后
+更早收尾；零值仍不限。本改动不做请求预留、供应商取消或进程终止。
+
+## Caller and compatibility boundary / 调用者与兼容边界
+
+Advisor, recovery reviewer, Goal evaluator, best-of judge, commit-message and
+prompt-refinement callers share the bounded owner. Participation depends on an
+active context observer, not their label. Calls before/after Agent Run or
+outside its context remain outside that root scope.
+
+advisor、恢复 reviewer、Goal evaluator、best-of judge、提交信息和提示改写调用者
+共用有界 owner。是否计入取决于活跃 context observer，而不是调用者标签。
+Agent Run 前后或其 context 之外的调用仍不属于该根 scope。
+
+The two helpers add no RPC, config, persisted or provider-wire field. Standalone
+reviewers retain their two-message, no-tool requests and existing time/output
+bounds. Main system messages, tool schemas and prefix construction are unchanged.
+
+两个 helper 不新增 RPC、配置、持久化或 provider-wire 字段。独立 reviewer 保留
+两条 message、无工具请求及既有时间/输出边界。主 system message、工具 schema
+与前缀构造保持不变。
+
+Real Agent and boot HTTP/SSE regressions cover actual advisor answers, nested
+scope replacement, missing usage, disabled limits and serialized prefix guards.
+Late projection is a constructed control, not a production background incident;
+mock request equality is not a live-provider cache-rate measurement.
+
+真实 Agent 与 boot HTTP/SSE 回归覆盖实际 advisor 答案、嵌套 scope 替换、缺失
+用量、禁用阈值与序列化前缀守卫。延后投射属于构造控制，不是生产后台事故；mock
+请求相等不是供应商命中率实测。
+
+This is not complete auxiliary-provider admission, a shared Coordinator root,
+durable accounting, root currency aggregation or the full isolation plan.
+See [root token budgets](ROOT_TOKEN_BUDGET_SECURITY.md).
+
+本机制不等于完整辅助 provider 准入、共享 Coordinator 根预算、持久计账、根币种
+汇总或完整隔离方案。作者分析不等于独立 review 或原生平台执行。
diff --git a/internal/assembly/boot/effect_root_bounded_budget_test.go b/internal/assembly/boot/effect_root_bounded_budget_test.go
new file mode 100644
index 0000000000..896025708c
--- /dev/null
+++ b/internal/assembly/boot/effect_root_bounded_budget_test.go
@@ -0,0 +1,142 @@
+package boot
+
+import (
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+
+	"reasonix/internal/contract/event"
+	"reasonix/internal/runtime/agent"
+	"reasonix/internal/session/control"
+	"reasonix/internal/state/sessionstore"
+)
+
+func TestEffectRootBoundedAdvisorBudget(t *testing.T) {
+	for _, tc := range []struct {
+		name       string
+		limit      int
+		missing    bool
+		wantAdvice int32
+	}{
+		{name: "reported", limit: 1500, wantAdvice: 1},
+		{name: "disabled", wantAdvice: 4},
+		{name: "missing_usage", limit: 1500, missing: true, wantAdvice: 2},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			home, root := statedBootHome(t), t.TempDir()
+			var normal, advice, results, usageEvents atomic.Int32
+			var mu sync.Mutex
+			var systems, schemas []string
+			srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				var req struct {
+					Model    string          `json:"model"`
+					Tools    json.RawMessage `json:"tools"`
+					Messages []struct {
+						Role    string          `json:"role"`
+						Content json.RawMessage `json:"content"`
+					} `json:"messages"`
+				}
+				if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+					t.Error(err)
+					w.WriteHeader(http.StatusBadRequest)
+					return
+				}
+				w.Header().Set("Content-Type", "text/event-stream")
+				advisor := req.Model == "advisor-model"
+				if advisor {
+					advice.Add(1)
+					if len(req.Messages) != 2 || req.Messages[0].Role != "system" || req.Messages[1].Role != "user" || (len(req.Tools) > 0 && string(req.Tools) != "null") {
+						t.Error("advisor lost its bounded isolated request shape")
+					}
+					_, _ = io.WriteString(w, "data: {\"choices\":[{\"delta\":{\"content\":\"ADVICE: inspect the lock owner\"},\"finish_reason\":\"stop\"}]}\n\n")
+				} else {
+					mu.Lock()
+					for _, m := range req.Messages {
+						if m.Role == "system" {
+							systems = append(systems, string(m.Content))
+							break
+						}
+					}
+					schemas = append(schemas, string(req.Tools))
+					mu.Unlock()
+					if n := normal.Add(1); n <= 4 {
+						_, _ = fmt.Fprintf(w, "data: {\"choices\":[{\"delta\":{\"tool_calls\":[{\"index\":0,\"id\":\"advise-%d\",\"type\":\"function\",\"function\":{\"name\":\"advise\",\"arguments\":\"{\\\"question\\\":\\\"Which lock owns this operation?\\\"}\"}}]},\"finish_reason\":\"tool_calls\"}]}\n\n", n)
+					} else {
+						_, _ = io.WriteString(w, "data: {\"choices\":[{\"delta\":{\"content\":\"Finished\"},\"finish_reason\":\"stop\"}]}\n\n")
+					}
+				}
+				if !advisor || !tc.missing {
+					_, _ = io.WriteString(w, "data: {\"choices\":[],\"usage\":{\"prompt_tokens\":1000,\"completion_tokens\":100,\"total_tokens\":1100}}\n\n")
+				}
+				_, _ = io.WriteString(w, "data: [DONE]\n\n")
+			}))
+			t.Cleanup(srv.Close)
+			writeFile(t, root, "reasonix.toml", fmt.Sprintf(`default_model = "worker"
+[agent]
+advisor_model = "advisor"
+task_token_budget = %d
+[codegraph]
+enabled = false
+[[providers]]
+name = "worker"
+kind = "openai"
+model = "worker-model"
+base_url = %q
+api_key_env = "BOUNDED_ROOT_EFFECT_KEY"
+[[providers]]
+name = "advisor"
+kind = "openai"
+model = "advisor-model"
+base_url = %q
+api_key_env = "BOUNDED_ROOT_EFFECT_KEY"
+`, tc.limit, srv.URL, srv.URL))
+			t.Setenv("BOUNDED_ROOT_EFFECT_KEY", "test")
+			approveWorkspaceAt(t, home, root)
+			sink := event.FuncSink(func(e event.Event) {
+				if e.Kind == event.ToolResult && e.Tool.Name == "advise" && e.Tool.Err == "" && strings.Contains(e.Tool.Output, "ADVICE: inspect") {
+					results.Add(1)
+				}
+				if e.Kind == event.Usage && e.UsageSource == event.UsageSourceAdvisor && e.Usage != nil {
+					usageEvents.Add(1)
+					if e.Usage.RequestCount != 1 || e.Usage.Estimated != tc.missing {
+						t.Errorf("advisor provenance=%+v", e.Usage)
+					}
+				}
+			})
+			ctrl, err := Build(t.Context(), Options{Home: home, WorkspaceRoot: root, Sink: sink})
+			if err != nil {
+				t.Fatal(err)
+			}
+			t.Cleanup(ctrl.Close)
+			ctrl.SetSessionPath(sessionstore.NewSessionPath(ctrl.SessionDir(), ctrl.Label()))
+			ctrl.SetToolApprovalMode(control.ToolApprovalYolo)
+			err = ctrl.Run(t.Context(), "review lock ownership")
+			if tc.limit == 0 {
+				if err != nil {
+					t.Fatal(err)
+				}
+			} else if pause, ok := agent.InspectRunPause(err); !ok || pause.Key != "token" {
+				t.Fatalf("pause=%+v err=%v", pause, err)
+			}
+			if advice.Load() != tc.wantAdvice || results.Load() != tc.wantAdvice || usageEvents.Load() != tc.wantAdvice || normal.Load() != tc.wantAdvice+1 {
+				t.Fatalf("normal/advisor/result/usage=%d/%d/%d/%d, want%d/%d/%d/%d", normal.Load(), advice.Load(), results.Load(), usageEvents.Load(), tc.wantAdvice+1, tc.wantAdvice, tc.wantAdvice, tc.wantAdvice)
+			}
+			mu.Lock()
+			defer mu.Unlock()
+			if len(systems) != int(normal.Load()) || len(schemas) != len(systems) {
+				t.Fatal("missing main request prefix")
+			}
+			for i := 1; i < len(systems); i++ {
+				if systems[i] != systems[0] || schemas[i] != schemas[0] {
+					t.Fatal("root accounting changed stable system/schema prefix")
+				}
+			}
+		})
+	}
+}
diff --git a/internal/contract/provider/auxiliary_usage.go b/internal/contract/provider/auxiliary_usage.go
new file mode 100644
index 0000000000..9609bb9589
--- /dev/null
+++ b/internal/contract/provider/auxiliary_usage.go
@@ -0,0 +1,26 @@
+package provider
+
+import "context"
+
+type auxiliaryUsageObserverKey struct{}
+
+// WithAuxiliaryUsageObserver binds one trusted host accounting observer. It
+// replaces an inherited observer so nested scopes do not charge a root twice.
+// A nil observer disables observation. Cancellation and deadlines are unchanged.
+func WithAuxiliaryUsageObserver(ctx context.Context, observe func(*Usage)) context.Context {
+	return context.WithValue(ctx, auxiliaryUsageObserverKey{}, observe)
+}
+
+// ObserveAuxiliaryUsage reports one logical call's finalized usage without
+// exposing its caller-owned record to mutation. Observers may run concurrently
+// and own their synchronization. No observer or unknown usage is a no-op.
+func ObserveAuxiliaryUsage(ctx context.Context, usage *Usage) {
+	if usage == nil {
+		return
+	}
+	observe, _ := ctx.Value(auxiliaryUsageObserverKey{}).(func(*Usage))
+	if observe != nil {
+		copy := *usage
+		observe(&copy)
+	}
+}
diff --git a/internal/contract/provider/auxiliary_usage_test.go b/internal/contract/provider/auxiliary_usage_test.go
new file mode 100644
index 0000000000..2d227adc35
--- /dev/null
+++ b/internal/contract/provider/auxiliary_usage_test.go
@@ -0,0 +1,30 @@
+package provider
+
+import (
+	"context"
+	"errors"
+	"testing"
+)
+
+func TestAuxiliaryUsageObserverReplacesAndCopies(t *testing.T) {
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+	outer, inner := 0, 0
+	ctx = WithAuxiliaryUsageObserver(ctx, func(*Usage) { outer++ })
+	ctx = WithAuxiliaryUsageObserver(ctx, func(u *Usage) { inner++; u.PromptTokens = 999; u.Estimated = true })
+	cancel()
+	if !errors.Is(ctx.Err(), context.Canceled) {
+		t.Fatal("observer changed cancellation")
+	}
+	u := Usage{PromptTokens: 10, CompletionTokens: 2, RequestCount: 3}
+	ObserveAuxiliaryUsage(ctx, &u)
+	if outer != 0 || inner != 1 || u.PromptTokens != 10 || u.Estimated {
+		t.Fatalf("observer leaked or duplicated:outer%d inner%d usage%+v", outer, inner, u)
+	}
+	ObserveAuxiliaryUsage(ctx, nil)
+	ObserveAuxiliaryUsage(context.Background(), &u)
+	ObserveAuxiliaryUsage(WithAuxiliaryUsageObserver(ctx, nil), &u)
+	if inner != 1 {
+		t.Fatal("unknown usage or disabled observer caused a charge")
+	}
+}
diff --git a/internal/model/boundedllm/bounded.go b/internal/model/boundedllm/bounded.go
index 9fa4b8528b..da8407e624 100644
--- a/internal/model/boundedllm/bounded.go
+++ b/internal/model/boundedllm/bounded.go
@@ -117,6 +117,7 @@ func Call(ctx context.Context, cfg Config, system, evidence string) (string, err
 	var usage *provider.Usage
 	defer func() {
 		usage = provider.UsageWithRequestAttemptCount(callCtx, usage)
+		provider.ObserveAuxiliaryUsage(callCtx, usage)
 		if usage != nil && cfg.UsageSource != "" && cfg.Sink != nil {
 			cfg.Sink.Emit(event.Event{
 				Kind:        event.Usage,
diff --git a/internal/model/boundedllm/usage_observer_test.go b/internal/model/boundedllm/usage_observer_test.go
new file mode 100644
index 0000000000..9467115e50
--- /dev/null
+++ b/internal/model/boundedllm/usage_observer_test.go
@@ -0,0 +1,72 @@
+package boundedllm
+
+import (
+	"errors"
+	"testing"
+
+	"reasonix/internal/contract/event"
+	"reasonix/internal/contract/provider"
+)
+
+func TestBoundedUsageObserverOwnsOneTerminal(t *testing.T) {
+	for _, mode := range []string{"reported", "billed_failure", "without_sink", "missing", "input_refused"} {
+		t.Run(mode, func(t *testing.T) {
+			observed, emitted := 0, 0
+			original := provider.Usage{PromptTokens: 1000, CompletionTokens: 100, TotalTokens: 1100, RequestCount: 2}
+			failure := errors.New("review failure")
+			chunks := chunkProvider{{Type: provider.ChunkText, Text: "reviewed"}}
+			if mode != "missing" {
+				chunks = append(chunks, provider.Chunk{Type: provider.ChunkUsage, Usage: &original})
+			}
+			if mode == "billed_failure" {
+				chunks = append(chunks, provider.Chunk{Type: provider.ChunkError, Err: failure})
+			}
+			ctx := provider.WithAuxiliaryUsageObserver(t.Context(), func(u *provider.Usage) {
+				observed++
+				if u.PromptTokens != 1000 || u.RequestCount != 2 {
+					t.Errorf("observer usage=%+v", u)
+				}
+				u.PromptTokens = 9999
+			})
+			cfg := Config{Provider: chunks, UsageSource: event.UsageSourceAdvisor, Sink: event.FuncSink(func(e event.Event) {
+				emitted++
+				if e.Usage == nil || e.Usage.PromptTokens != 1000 {
+					t.Errorf("observer corrupted emitted usage=%+v", e.Usage)
+				}
+			})}
+			if mode == "without_sink" {
+				cfg.Sink = nil
+				cfg.UsageSource = ""
+			}
+			if mode == "input_refused" {
+				cfg.MaxSystemBytes = 1
+			}
+			text, err := Call(ctx, cfg, "policy", "evidence")
+			if mode == "billed_failure" {
+				if !errors.Is(err, failure) {
+					t.Fatalf("error cause lost:%v", err)
+				}
+			} else if mode == "input_refused" {
+				if err == nil {
+					t.Fatal("oversized input was admitted")
+				}
+			} else if err != nil || text != "reviewed" {
+				t.Fatalf("answer=%q err=%v", text, err)
+			}
+			want := 1
+			if mode == "missing" || mode == "input_refused" {
+				want = 0
+			}
+			if observed != want || original.PromptTokens != 1000 {
+				t.Fatalf("observed=%d want%d original=%+v", observed, want, original)
+			}
+			wantEmit := want
+			if mode == "without_sink" {
+				wantEmit = 0
+			}
+			if emitted != wantEmit {
+				t.Fatalf("usage events=%d,want%d", emitted, wantEmit)
+			}
+		})
+	}
+}
diff --git a/internal/runtime/agent/root_bounded_budget_test.go b/internal/runtime/agent/root_bounded_budget_test.go
new file mode 100644
index 0000000000..627c328544
--- /dev/null
+++ b/internal/runtime/agent/root_bounded_budget_test.go
@@ -0,0 +1,165 @@
+package agent
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"testing"
+
+	"reasonix/internal/contract/event"
+	"reasonix/internal/contract/provider"
+	"reasonix/internal/contract/tool"
+	"reasonix/internal/model/boundedllm"
+	"reasonix/internal/state/sessionstore"
+)
+
+type boundedBudgetProvider struct {
+	calls         int
+	missing, fail bool
+}
+
+func (*boundedBudgetProvider) Name() string { return "bounded-budget" }
+
+func (p *boundedBudgetProvider) Stream(context.Context, provider.Request) (<-chan provider.Chunk, error) {
+	p.calls++
+	ch := make(chan provider.Chunk, 3)
+	ch <- provider.Chunk{Type: provider.ChunkText, Text: "reviewed"}
+	if !p.missing {
+		ch <- provider.Chunk{Type: provider.ChunkUsage, Usage: &provider.Usage{PromptTokens: 1000, CompletionTokens: 100, TotalTokens: 1100}}
+	}
+	if p.fail {
+		ch <- provider.Chunk{Type: provider.ChunkError, Err: errors.New("billed review failure")}
+	}
+	close(ch)
+	return ch, nil
+}
+
+type boundedBudgetTool struct {
+	readProbe
+	prov                  *boundedBudgetProvider
+	noEmission, projected bool
+	capture               context.Context
+	failures              int
+}
+
+func (p *boundedBudgetTool) Execute(ctx context.Context, _ json.RawMessage) (string, error) {
+	p.capture = InheritRootTokenBudget(context.Background(), ctx)
+	if p.projected {
+		ctx = p.capture
+	}
+	cfg := boundedllm.Config{Provider: p.prov, Sink: event.Discard, UsageSource: event.UsageSourceAdvisor}
+	if p.noEmission {
+		cfg.Sink = nil
+		cfg.UsageSource = ""
+	}
+	answer, err := boundedllm.Call(ctx, cfg, "review policy", "bounded evidence")
+	if err != nil {
+		p.failures++
+		return "review failed", nil
+	}
+	return answer, nil
+}
+
+func TestRootBoundedReviewerBudget(t *testing.T) {
+	for _, tc := range []struct {
+		name                                           string
+		missing, fail, disabled, noEmission, projected bool
+		wantCalls                                      int32
+	}{
+		{name: "reported", wantCalls: 2},
+		{name: "billed_failure", fail: true, wantCalls: 2},
+		{name: "missing_usage", missing: true, wantCalls: 3},
+		{name: "disabled", disabled: true, wantCalls: 5},
+		{name: "no_emission", noEmission: true, wantCalls: 2},
+		{name: "projected_job_context", projected: true, wantCalls: 2},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			aux := &boundedBudgetProvider{missing: tc.missing, fail: tc.fail}
+			probe := &boundedBudgetTool{prov: aux, noEmission: tc.noEmission, projected: tc.projected}
+			reg := tool.NewRegistry()
+			reg.Add(probe)
+			main := &spendingProvider{max: 4}
+			root := New(main, reg, sessionstore.NewSession("root"), Options{TaskBudget: TaskBudget{Tokens: 1500}}, event.Discard)
+			ctx := t.Context()
+			if tc.disabled {
+				ctx = WithTaskBudget(ctx, TaskBudget{})
+			}
+			err := root.Run(ctx, "obtain a review")
+			if tc.disabled {
+				if err != nil {
+					t.Fatal(err)
+				}
+			} else if pause, ok := InspectRunPause(err); !ok || pause.Key != "token" {
+				t.Fatalf("pause=%+v err=%v", pause, err)
+			}
+			if main.rounds.Load() != tc.wantCalls || aux.calls != int(tc.wantCalls-1) {
+				t.Fatalf("normal/reviewer calls=%d/%d, want%d/%d", main.rounds.Load(), aux.calls, tc.wantCalls, tc.wantCalls-1)
+			}
+			want := uint64(main.rounds.Load()) * 1100
+			if !tc.missing {
+				want += uint64(aux.calls) * 1100
+			}
+			if root.task.budget.rootTokens.used.Load() != want {
+				t.Fatalf("root tokens=%d,want%d", root.task.budget.rootTokens.used.Load(), want)
+			}
+			if root.task.budget.promptTokens+root.task.budget.outputTokens != int(main.rounds.Load())*1100 {
+				t.Fatal("bounded usage changed local normal telemetry")
+			}
+			if tc.fail && probe.failures != aux.calls {
+				t.Fatal("fixture did not retain each billed failure")
+			}
+		})
+	}
+}
+
+func TestRootBoundedNestedBudgetChargesOnce(t *testing.T) {
+	aux := &boundedBudgetProvider{}
+	childTools := tool.NewRegistry()
+	childTools.Add(&boundedBudgetTool{prov: aux})
+	childProv := &spendingProvider{max: 4}
+	child := New(childProv, childTools, sessionstore.NewSession("child"), Options{}, event.Discard)
+	parentTools := tool.NewRegistry()
+	parentTools.Add(&tokenBudgetChildTool{child: child})
+	parentProv := &spendingProvider{max: 1}
+	root := New(parentProv, parentTools, sessionstore.NewSession("root"), Options{TaskBudget: TaskBudget{Tokens: 2500}}, event.Discard)
+	err := root.Run(t.Context(), "delegate the review")
+	if pause, ok := InspectRunPause(err); !ok || pause.Key != "token" {
+		t.Fatalf("pause=%+v err=%v", pause, err)
+	}
+	if parentProv.rounds.Load() != 2 || childProv.rounds.Load() != 2 || aux.calls != 1 {
+		t.Fatalf("parent/child/aux=%d/%d/%d,want2/2/1", parentProv.rounds.Load(), childProv.rounds.Load(), aux.calls)
+	}
+	if got := root.task.budget.rootTokens.used.Load(); got != 5500 {
+		t.Fatalf("root tokens=%d,want5500 without duplicate observers", got)
+	}
+}
+
+func TestRootBoundedLateProjectionStaysWithOldTask(t *testing.T) {
+	aux := &boundedBudgetProvider{}
+	probe := &boundedBudgetTool{prov: aux}
+	reg := tool.NewRegistry()
+	reg.Add(probe)
+	root := New(&spendingProvider{max: 1}, reg, sessionstore.NewSession("root"), Options{}, event.Discard)
+	ctx, cancel := context.WithCancel(t.Context())
+	t.Cleanup(cancel)
+	if err := root.Run(ctx, "capture budget"); err != nil {
+		t.Fatal(err)
+	}
+	old := root.task.budget.rootTokens
+	before := old.used.Load()
+	cancel()
+	if err := root.Run(t.Context(), "new task"); err != nil {
+		t.Fatal(err)
+	}
+	current := root.task.budget.rootTokens
+	currentBefore := current.used.Load()
+	if old == current {
+		t.Fatal("new task kept old scope")
+	}
+	if _, err := boundedllm.Call(probe.capture, boundedllm.Config{Provider: aux}, "review policy", "late evidence"); err != nil {
+		t.Fatal(err)
+	}
+	if old.used.Load() != before+1100 || current.used.Load() != currentBefore {
+		t.Fatalf("late usage went to wrong scope:old%d/%d,new%d/%d", old.used.Load(), before, current.used.Load(), currentBefore)
+	}
+}
diff --git a/internal/runtime/agent/root_token_budget.go b/internal/runtime/agent/root_token_budget.go
index 2f85f3dc14..82d3722619 100644
--- a/internal/runtime/agent/root_token_budget.go
+++ b/internal/runtime/agent/root_token_budget.go
@@ -20,7 +20,8 @@ type rootTokenBudgetKey struct{}
 // Token and wall bounds share its lifetime; job cancellation remains separate.
 func InheritRootTokenBudget(ctx, parent context.Context) context.Context {
 	if scope, ok := parent.Value(rootTokenBudgetKey{}).(*rootTokenBudget); ok {
-		return context.WithValue(ctx, rootTokenBudgetKey{}, scope)
+		ctx = context.WithValue(ctx, rootTokenBudgetKey{}, scope)
+		return provider.WithAuxiliaryUsageObserver(ctx, scope.observe)
 	}
 	return ctx
 }
@@ -47,7 +48,7 @@ func (a *Agent) bindRootTokenBudget(ctx context.Context, state *turnRuntime) con
 		ctx = context.WithValue(ctx, rootTokenBudgetKey{}, scope)
 	}
 	state.budget.rootTokens = scope
-	return ctx
+	return provider.WithAuxiliaryUsageObserver(ctx, scope.observe)
 }
 
 func (b *rootTokenBudget) observe(usage *provider.Usage) {
diff --git a/release-notes/studio/unreleased/14025-bounded-root-token-usage.md b/release-notes/studio/unreleased/14025-bounded-root-token-usage.md
new file mode 100644
index 0000000000..77223e1068
--- /dev/null
+++ b/release-notes/studio/unreleased/14025-bounded-root-token-usage.md
@@ -0,0 +1,2 @@
+- 修复：活跃 Agent 根 context 内的有界 reviewer/advisor 用量现在计入既有 token 软预算，保留缺失用量、零阈值、局部遥测和独立请求边界。关联 #14025。
+- Fixed: bounded reviewer/advisor usage in an active Agent root now counts toward its token soft budget. Missing usage, zero limits and isolated requests retain their behavior. Refs #14025.