Merge remote-tracking branch 'origin/master' into worktree/fix-pty-sigint-flake

2026-07-27 00:35:06 +08:00
parent e0e187a6a7 8cb908b9e8
commit 420e0b5fb7
72 changed files with 2131 additions and 172 deletions
--- a/packages/AGENTS.md
+++ b/packages/AGENTS.md
@@ -7,7 +7,7 @@ These package-specific rules supplement the repo-wide [conventions](../AGENTS.md
 - **Product-visible plugins require a non-unit REAL-composition test.** Hand-built `ctx.plugin(...)` suites are insufficient. Boot test-only `cordis.yml` through the Loader and app/process; mock only external/nondeterministic boundaries and assert model-visible, durable, or user-visible output. Keep opt-ins out of shipped defaults. [Policy](../docs/testing.md).
 - **Initiator-owned private chains derive, then capture.** Under `ctx.agents.withInitiator()`, recover the Agent at each orchestration entry, derive `agent.session`, and let operation-local helpers close over it. Keep `Agent` and `Session` explicit at lifecycle, session-log, service, authority, worker/process, persistence, and wire interfaces; do not widen a leaf helper from `Session` to `Context` merely to hide a parameter ([rationale](../.agents/notes/implemented/architecture/2026-07-15-agent-initiator-scope.md)).
 - **Represent one asynchronous operation with one lifecycle controller or transaction.** Separate readiness, cancellation, disposal, reservation, or sentinel state requires an independent owner or settlement boundary; otherwise fold it while preserving rollback, callback containment, and quiescence.
- **Shape capability interfaces around all current consumers.** Keep tool-schema, Loader, UI, transport, and backend-specific behavior in the consumer or adapter; do not let one consumer dictate the interface ([capability-seam rationale](../.agents/notes/implemented/architecture/2026-06-13-capability-seams.md)).
+- **Shape capability interfaces around all current consumers.** Keep tool-schema, Loader, UI, transport, and backend-specific behavior in the consumer or adapter; do not let one consumer dictate the interface ([capability-seam rationale](../.agents/notes/implemented/architecture/2026-06-13-capability-seams.md)). Inverse smell: a public service method with one internal caller — pass a private capability closure instead (`RunCodeBridgeOptions`).
 - **Require a current owner and need.** Tie each abstraction, state machine, option, defensive copy, and compatibility path to a current contract or production consumer, and keep behavior in its owning plugin or service.
 - **Require evidence for public choices.** Configurability does not justify an unsupported default, public operation set, format, or imported external concept. Use current-consumer evidence or relevant prior art; otherwise require an explicit value or defer the choice.
 - **Write model-facing contracts from the model's perspective.** Prompts, tool schemas, results, and diagnostics contain only task-relevant concepts, not UI, transport, or implementation vocabulary. Pin stable model-visible text verbatim and dynamic behavior through snapshots or end-to-end coverage.
--- a/packages/client/ui-conversation/src/client/chat/AssistantMarkdown.tsx
+++ b/packages/client/ui-conversation/src/client/chat/AssistantMarkdown.tsx
@@ -44,7 +44,7 @@ export const AssistantMarkdown = memo(function AssistantMarkdown({ blocks, strea
    <div className={css.root} data-streaming={streaming || undefined}>
      {blocks.map((block, i) => {
        switch (block.kind) {
-          case 'text': return <MarkdownText key={i} text={block.text} />
+          case 'text': return <MarkdownText key={i} text={block.text} streaming={streaming} />
          case 'reasoning': return <ThinkRow key={i} text={block.text} running={streaming && i === last} />
          // Tool-call heads render as tool rows in the chat view's grouping pass.
          case 'tool-call': return null
--- a/packages/client/ui-conversation/src/client/chat/ToolRow.module.css
+++ b/packages/client/ui-conversation/src/client/chat/ToolRow.module.css
@@ -87,14 +87,9 @@ button.leading {
  color: var(--dsw-alias-label-tertiary);
 }

-/* The code variant's expanded body is the run_code program: monospace on the
-   markdown code-block fill so the program reads as code, not prose. */
-.root[data-variant='code'] .body {
-  font-family: var(--ds-font-family-code);
-  font-size: 13px;
-  line-height: 20px;
-  padding: 6px 8px;
-  margin-left: 22px;
-  border-radius: 6px;
-  background: var(--dsw-alias-markdown-code-block);
+/* The code variant's expanded body is the run_code program, rendered through
+   the shared CodeBlock (shiki-highlighted TypeScript); only indentation is
+   this row's concern. */
+.codeBody {
+  margin: 4px 0 4px 22px;
 }
--- a/packages/client/ui-conversation/src/client/chat/ToolRow.tsx
+++ b/packages/client/ui-conversation/src/client/chat/ToolRow.tsx
@@ -6,7 +6,7 @@

 import { useState, type KeyboardEvent, type MouseEvent, type ReactNode } from 'react'
 import clsx from 'clsx'
-import { StateDot } from '@deepseek-ai/dsh-client-ui-primitives'
+import { CodeBlock, StateDot } from '@deepseek-ai/dsh-client-ui-primitives'
 import { IconChevronDownOutline14 } from '@deepseek-ai/dsh-client-ui-primitives'
 import type { ToolRowState, ToolRowVariant } from '../contract/tool-call-model.ts'
 import css from './ToolRow.module.css'
@@ -96,7 +96,9 @@ export function ToolRow({
          </>
        )}
      </div>
-      {open && <div className={css.body}>{body}</div>}
+      {open && (variant === 'code'
+        ? <CodeBlock code={body} lang="typescript" className={css.codeBody} />
+        : <div className={css.body}>{body}</div>)}
    </div>
  )
 }
--- a/packages/client/ui-conversation/src/client/skeleton/DetailsPanel.tsx
+++ b/packages/client/ui-conversation/src/client/skeleton/DetailsPanel.tsx
@@ -5,6 +5,7 @@
 // share the store seat exists for) and derives the call material from the
 // session snapshot — no data of its own.

+import { CodeBlock } from '@deepseek-ai/dsh-client-ui-primitives'
 import { shallowEqual } from '@deepseek-ai/dsh-client-runtime/client'
 import type { ConversationSnapshot, ToolResultNode } from '@deepseek-ai/dsh-client-runtime/client'
 import type { DetailsSlotProps } from '../contract/slots.ts'
@@ -89,7 +90,7 @@ export function DetailsPanel({ useSession, useStore, closeDetails }: DetailsPane
                  {material.argsRaw !== null && (
                    <section className={css.section}>
                      <div className={css.sectionLabel}>Input</div>
-                      <pre className={css.code}>{pretty(material.argsRaw)}</pre>
+                      <CodeBlock code={pretty(material.argsRaw)} lang="json" />
                    </section>
                  )}
                  <section className={css.section}>
--- a/packages/client/ui-conversation/tests/chat-code-subcalls.spec.tsx
+++ b/packages/client/ui-conversation/tests/chat-code-subcalls.spec.tsx
@@ -152,7 +152,7 @@ describe('run_code sub-calls through the real chat machinery', () => {
    expect(view.getByText('Tool call')).toBeTruthy()
  })

-  it('expanding the code row reveals the program body verbatim', async () => {
+  it('expanding the code row reveals the program body verbatim (shiki-tokenized)', async () => {
    const parent = 'call-64'
    const b = await bench(snapshotWith([codeResult(10, parent)], new Map()))
    const view = mountApp(b.slots)
@@ -160,7 +160,12 @@ describe('run_code sub-calls through the real chat machinery', () => {
    const toggle = view.container.querySelector('[data-variant="code"] button[aria-expanded]')
    expect(toggle).not.toBeNull()
    fireEvent.click(toggle!)
-    expect(view.getByText(/const listing = await tools\.bash/)).toBeTruthy()
+    // Shiki splits the program into token spans inside one <pre class="shiki">:
+    // assert the whole text and the highlighted tree rather than one node.
+    const pre = view.container.querySelector('pre.shiki')
+    expect(pre).not.toBeNull()
+    expect(pre!.textContent).toContain('const listing = await tools.bash')
+    expect(pre!.querySelectorAll('span[style]').length).toBeGreaterThan(3)
  })

  it('an isError sub-call renders the error state dot exactly like a failed native row', async () => {
--- a/packages/client/ui-primitives/package.json
+++ b/packages/client/ui-primitives/package.json
@@ -20,11 +20,13 @@
  },
  "license": "BSD-3-Clause",
  "dependencies": {
+    "@shikijs/langs": "^4.3.1",
    "clsx": "^2.0.0",
    "react": "^18.2.0",
    "react-dom": "^18.2.0",
    "react-markdown": "^10.1.0",
-    "remark-gfm": "^4.0.1"
+    "remark-gfm": "^4.0.1",
+    "shiki": "^4.3.1"
  },
  "devDependencies": {
    "@deepseek-ai/dsh-invariants": "workspace:^",
--- a/packages/client/ui-primitives/src/index.ts
+++ b/packages/client/ui-primitives/src/index.ts
@@ -17,6 +17,7 @@ export { FishLogo } from './FishLogo.tsx'
 export { BrandWordmark } from './BrandWordmark.tsx'
 export { Tooltip } from './Tooltip.tsx'
 export type { TooltipSide } from './Tooltip.tsx'
+export { CodeBlock } from './markdown/CodeBlock.tsx'
 export { JsonBlock } from './markdown/JsonBlock.tsx'
 export { MarkdownText } from './markdown/MarkdownText.tsx'
 export { MessageText } from './markdown/MessageText.tsx'
--- a/packages/client/ui-primitives/src/markdown/CodeBlock.module.css
+++ b/packages/client/ui-primitives/src/markdown/CodeBlock.module.css
@@ -0,0 +1,27 @@
+/* One code-block geometry for highlighted and plain arms: the shiki <pre>
+   and the fallback <pre> draw identically except for token colors. */
+
+.block :where(pre) {
+  margin: 0;
+  padding: 8px 10px;
+  border-radius: 8px;
+  overflow-x: auto;
+  background: var(--dsw-alias-markdown-code-block);
+  font: var(--dsw-font-markdown-code-block);
+}
+
+/* Shiki inlines its theme background var; route it to the repo token. */
+.block :where(pre.shiki) {
+  background: var(--dsw-alias-markdown-code-block) !important;
+}
+
+.block :where(pre) code {
+  font: inherit;
+  background: none;
+  padding: 0;
+}
+
+.plain {
+  color: var(--dsw-alias-label-primary);
+  white-space: pre;
+}
--- a/packages/client/ui-primitives/src/markdown/CodeBlock.tsx
+++ b/packages/client/ui-primitives/src/markdown/CodeBlock.tsx
@@ -0,0 +1,37 @@
+// CodeBlock: one code surface for every consumer — markdown fences, the
+// run_code program body, and the details panel's raw args/output — with
+// shiki highlighting for the registered grammars and an identical-geometry
+// plain fallback for everything else. Shiki emits a single <pre class="shiki">
+// tree of nested spans whose colors are --shiki-* custom properties
+// (token sheets own the values); it produces no scripts or event handlers,
+// so injecting its output is safe by construction.
+
+import { useMemo } from 'react'
+import clsx from 'clsx'
+import { highlightToHtml } from './highlight.ts'
+import css from './CodeBlock.module.css'
+
+export interface CodeBlockProps {
+  /** The source text, rendered verbatim (trailing newline trimmed for display). */
+  code: string
+  /** Grammar hint (markdown fence info string or a fixed caller id); unknown = plain. */
+  lang?: string | undefined
+  /** Extra class merged onto the wrapper (callers position; this component draws). */
+  className?: string | undefined
+}
+
+export function CodeBlock({ code, lang, className }: CodeBlockProps) {
+  const trimmed = code.endsWith('\n') ? code.slice(0, -1) : code
+  const html = useMemo(() => highlightToHtml(trimmed, lang), [trimmed, lang])
+  if (html === undefined) {
+    return (
+      <div className={clsx(css.block, className)}>
+        <pre className={css.plain}><code>{trimmed}</code></pre>
+      </div>
+    )
+  }
+  // eslint-disable-next-line react/no-danger -- shiki's output is a static
+  // span tree it generated from `code` (no user HTML passes through), the
+  // sanctioned innerHTML consumption path per shiki's own docs.
+  return <div className={clsx(css.block, className)} dangerouslySetInnerHTML={{ __html: html }} />
+}
--- a/packages/client/ui-primitives/src/markdown/MarkdownText.tsx
+++ b/packages/client/ui-primitives/src/markdown/MarkdownText.tsx
@@ -1,6 +1,8 @@
+import { isValidElement } from 'react'
 import ReactMarkdown from 'react-markdown'
 import type { Components, UrlTransform } from 'react-markdown'
 import remarkGfm from 'remark-gfm'
+import { CodeBlock } from './CodeBlock.tsx'
 import css from './MarkdownText.module.css'

 const remarkPlugins = [remarkGfm]
@@ -22,7 +24,9 @@ function sanitizeUrl(url: string): string {

 const safeUrl: UrlTransform = url => sanitizeUrl(url)

-const components: Components = {
+/** Build the component table; while `streaming`, fences render the plain arm (see CodeBlock). */
+function buildComponents(streaming: boolean): Components {
+  return {
  a: ({ href = '', children }) => {
    const safeHref = sanitizeUrl(href)
    if (safeHref === '') return <>{children}</>
@@ -42,19 +46,40 @@ const components: Components = {
      <table>{children}</table>
    </div>
  ),
+    // Fenced blocks route through the shared CodeBlock (shiki for registered
+    // grammars, identical-geometry plain fallback for unknown/absent
+    // languages); inline code keeps the default <code> path (the :not(pre)
+    // rule styles it). While the message streams, the fence renders the
+    // plain arm — retokenizing a growing fence on every chunk is quadratic
+    // main-thread work; the finalize swap highlights it once.
+    pre: ({ children }) => {
+      /* v8 ignore next 2 -- the markdown pipeline always hands `pre` its single `code` element; the undefined arm guards a react-markdown representation change. */
+      const child = isValidElement<{ className?: string; children?: unknown }>(children) ? children : undefined
+      const raw = child?.props.children
+      // A fence whose content isn't one plain string (e.g. an empty fence)
+      // keeps the stock <pre> rather than guessing.
+      if (typeof raw !== 'string') return <pre>{children}</pre>
+      const lang = /language-([\w-]+)/.exec(child?.props.className ?? '')?.[1]
+      return <CodeBlock code={raw} lang={streaming ? undefined : lang} />
+    },
+  }
 }

+const staticComponents = buildComponents(false)
+const streamingComponents = buildComponents(true)
+
 /**
 * Render untrusted assistant-authored Markdown as semantic React elements.
- * @param props - Markdown source text preserved by the session projection.
+ * @param props - Markdown source text preserved by the session projection;
+ * `streaming` renders fences plain (highlighting lands on the finalize swap).
 * @returns A GFM document with raw HTML, relative links, unsafe protocols, and remote images disabled.
 */
-export function MarkdownText({ text }: { text: string }) {
+export function MarkdownText({ text, streaming = false }: { text: string; streaming?: boolean }) {
  return (
    <div className={css.markdown}>
      <ReactMarkdown
        remarkPlugins={remarkPlugins}
-        components={components}
+        components={streaming ? streamingComponents : staticComponents}
        urlTransform={safeUrl}
      >
        {text}
--- a/packages/client/ui-primitives/src/markdown/highlight.ts
+++ b/packages/client/ui-primitives/src/markdown/highlight.ts
@@ -0,0 +1,82 @@
+/**
+ * The client's ONE syntax highlighter: a synchronous fine-grained shiki core
+ * (JavaScript regex engine — no oniguruma WASM, bundle-friendly) with an
+ * explicit grammar allowlist and a CSS-variables theme. Colors live in the
+ * theme package's token sheets as `--shiki-*` custom properties (light and
+ * dark blocks), never here — the repo's tokens-only styling rule.
+ *
+ * Grammars are the set the harness actually renders: TypeScript programs
+ * (`run_code` bodies; TS pulls in JS via grammar embedding), shell commands,
+ * and JSON payloads. An unknown or absent language falls back to plain text
+ * (no highlighting, still monospace) — never an error.
+ */
+
+import { createHighlighterCoreSync, createCssVariablesTheme } from 'shiki/core'
+import { createJavaScriptRegexEngine } from 'shiki/engine/javascript'
+import langTs from '@shikijs/langs/typescript'
+import langBash from '@shikijs/langs/shellscript'
+import langJson from '@shikijs/langs/json'
+import type { HighlighterCore } from 'shiki/core'
+
+/**
+ * Language ids (and aliases) the singleton registers; everything else renders
+ * plain. A Map, not an object: fence info strings are assistant-authored, so
+ * a label like `constructor` or `__proto__` must miss instead of resolving an
+ * inherited property and crashing the renderer inside shiki.
+ */
+const LANG_ALIASES = new Map<string, string>([
+  ['typescript', 'typescript'],
+  ['ts', 'typescript'],
+  ['tsx', 'typescript'],
+  ['javascript', 'typescript'],
+  ['js', 'typescript'],
+  ['shellscript', 'shellscript'],
+  ['bash', 'shellscript'],
+  ['sh', 'shellscript'],
+  ['shell', 'shellscript'],
+  ['zsh', 'shellscript'],
+  ['json', 'json'],
+  ['jsonc', 'json'],
+])
+
+/** All token colors resolve through `--shiki-*` custom properties (theme package sheets). */
+const cssVariablesTheme = createCssVariablesTheme({
+  name: 'css-variables',
+  variablePrefix: '--shiki-',
+  fontStyle: true,
+})
+
+let singleton: HighlighterCore | undefined
+
+/** The synchronous highlighter (one instance per document); pre-warmed below, lazy as the fallback. */
+function highlighter(): HighlighterCore {
+  singleton ??= createHighlighterCoreSync({
+    themes: [cssVariablesTheme],
+    langs: [langTs, langBash, langJson],
+    engine: createJavaScriptRegexEngine({ forgiving: true }),
+  })
+  return singleton
+}
+
+// Engine + grammar construction costs a long task (~120-175ms); building it
+// during the first finalized fence's render would jank exactly when a stream
+// completes. Warm the singleton in a deferred task at module load (= plugin
+// boot) instead; the lazy path above stays as the correctness fallback for a
+// fence that renders before the timer fires. `unref` (Node-only) keeps a
+// non-browser import from pinning the event loop.
+const warmupTimer = setTimeout(() => { highlighter() }, 0)
+;(warmupTimer as { unref?: () => void }).unref?.()
+
+/**
+ * Highlight `code` into shiki's HTML (a single `<pre class="shiki">` tree)
+ * when `lang` maps to a registered grammar; `undefined` means the caller
+ * renders its plain fallback.
+ * @param code - the source text.
+ * @param lang - the language hint (a markdown fence info string or a fixed caller id).
+ * @returns the highlighted HTML, or `undefined` for unknown languages.
+ */
+export function highlightToHtml(code: string, lang: string | undefined): string | undefined {
+  const resolved = lang === undefined ? undefined : LANG_ALIASES.get(lang.toLowerCase())
+  if (resolved === undefined) return undefined
+  return highlighter().codeToHtml(code, { lang: resolved, theme: 'css-variables' })
+}
--- a/packages/client/ui-primitives/tests/code-block.spec.tsx
+++ b/packages/client/ui-primitives/tests/code-block.spec.tsx
@@ -0,0 +1,53 @@
+// @vitest-environment jsdom
+// CodeBlock + the shiki singleton: registered grammars highlight into token
+// spans colored by --shiki-* custom properties; unknown/absent languages take
+// the identical-geometry plain arm; aliases resolve; the trailing newline is
+// display-trimmed. MarkdownText's fence route is pinned in markdown.spec.tsx
+// alongside the rest of the markdown family.
+
+import { describe, expect, it } from 'vitest'
+import { cleanup, render } from '@testing-library/react'
+import { afterEach } from 'vitest'
+import { CodeBlock } from '../src/markdown/CodeBlock.tsx'
+import { highlightToHtml } from '../src/markdown/highlight.ts'
+
+afterEach(cleanup)
+
+describe('highlightToHtml', () => {
+  it('highlights a registered grammar into css-variables token spans', () => {
+    const html = highlightToHtml('const x: number = 1', 'typescript')
+    expect(html).toContain('pre class="shiki css-variables"')
+    expect(html).toContain('var(--shiki-')
+  })
+
+  it.each([['ts'], ['js'], ['bash'], ['sh'], ['jsonc']])('resolves the %s alias', (alias) => {
+    expect(highlightToHtml('x', alias)).toContain('shiki')
+  })
+
+  it('returns undefined for unknown or absent languages', () => {
+    expect(highlightToHtml('x', 'cobol')).toBeUndefined()
+    expect(highlightToHtml('x', undefined)).toBeUndefined()
+  })
+})
+
+describe('CodeBlock', () => {
+  it('renders the highlighted tree for TypeScript', () => {
+    const view = render(<CodeBlock code={'const a = 1\n'} lang="ts" />)
+    const pre = view.container.querySelector('pre.shiki')
+    expect(pre).not.toBeNull()
+    expect(pre!.textContent).toBe('const a = 1')
+    expect(pre!.querySelectorAll('span[style]').length).toBeGreaterThan(1)
+  })
+
+  it('renders the plain arm for an unknown language with the text verbatim', () => {
+    const view = render(<CodeBlock code={'IDENTIFICATION DIVISION.'} lang="cobol" />)
+    expect(view.container.querySelector('pre.shiki')).toBeNull()
+    expect(view.getByText('IDENTIFICATION DIVISION.')).toBeTruthy()
+  })
+
+  it('renders the plain arm when no language is given', () => {
+    const view = render(<CodeBlock code="plain text" />)
+    expect(view.container.querySelector('pre.shiki')).toBeNull()
+    expect(view.getByText('plain text')).toBeTruthy()
+  })
+})
--- a/packages/client/ui-primitives/tests/markdown.spec.tsx
+++ b/packages/client/ui-primitives/tests/markdown.spec.tsx
@@ -57,11 +57,41 @@ describe('MarkdownText', () => {
    expect(container.querySelector('table')?.textContent).toContain('alphabeta')
    expect(container.querySelector('hr')).not.toBeNull()
    expect(container.querySelector('pre code')?.textContent).toContain('const answer = 42')
+    // The ts fence routed through the shared CodeBlock: shiki token spans present.
+    expect(container.querySelector('pre.shiki')).not.toBeNull()
    expect(container.querySelector('br')).not.toBeNull()
    expect(screen.getByRole('link', { name: 'safe' }).getAttribute('target')).toBe('_blank')
    expect(screen.getByRole('link', { name: 'https://deepseek.com' })).toBeTruthy()
  })

+  it('a fence labeled with an inherited object key renders plain, never crashing shiki', () => {
+    for (const label of ['constructor', '__proto__', 'toString', 'hasOwnProperty']) {
+      const { container, unmount } = render(<MarkdownText text={'```' + label + '\ncode body\n```'} />)
+      expect(container.querySelector('pre.shiki')).toBeNull()
+      expect(container.querySelector('pre code')?.textContent).toContain('code body')
+      unmount()
+    }
+  })
+
+  it('an empty fence keeps the stock pre; a language-less fence renders the plain CodeBlock arm', () => {
+    const empty = render(<MarkdownText text={'```\n```'} />)
+    expect(empty.container.querySelector('pre')?.outerHTML).toBe('<pre><code></code></pre>')
+
+    const plain = render(<MarkdownText text={'```\nno language here\n```'} />)
+    expect(plain.container.querySelector('pre.shiki')).toBeNull()
+    expect(plain.container.querySelector('pre code')?.textContent).toContain('no language here')
+  })
+
+  it('streaming renders fences plain; the finalize swap highlights them', () => {
+    const fence = '```ts\nconst answer = 42\n```'
+    const live = render(<MarkdownText text={fence} streaming />)
+    expect(live.container.querySelector('pre.shiki')).toBeNull()
+    expect(live.container.querySelector('pre code')?.textContent).toContain('const answer = 42')
+    live.unmount()
+    const done = render(<MarkdownText text={fence} />)
+    expect(done.container.querySelector('pre.shiki')).not.toBeNull()
+  })
+
  it('neutralizes raw HTML, unsafe or relative links, and remote images', () => {
    const markdown = [
      '<script>globalThis.compromised = true</script>',
--- a/packages/client/ui-theme/src/styles/shiki.css
+++ b/packages/client/ui-theme/src/styles/shiki.css
@@ -0,0 +1,31 @@
+/* Syntax-highlight token palette: the values behind shiki's css-variables
+   theme (--shiki-* custom properties emitted by the ui-primitives CodeBlock).
+   Light values on :root, dark overrides on the body attribute — the same
+   cascade as every other token sheet. Background/foreground deliberately
+   alias the markdown code-block tokens so highlighted and plain blocks agree. */
+
+:root {
+  --shiki-foreground: var(--dsw-alias-label-primary);
+  --shiki-background: var(--dsw-alias-markdown-code-block);
+  --shiki-token-constant: #1c7ed6;
+  --shiki-token-string: #2f9e44;
+  --shiki-token-comment: #868e96;
+  --shiki-token-keyword: #d6336c;
+  --shiki-token-parameter: #e8590c;
+  --shiki-token-function: #6741d9;
+  --shiki-token-string-expression: #2b8a3e;
+  --shiki-token-punctuation: #495057;
+  --shiki-token-link: #1971c2;
+}
+
+body[data-ds-dark-theme] {
+  --shiki-token-constant: #4dabf7;
+  --shiki-token-string: #69db7c;
+  --shiki-token-comment: #adb5bd;
+  --shiki-token-keyword: #faa2c1;
+  --shiki-token-parameter: #ffa94d;
+  --shiki-token-function: #b197fc;
+  --shiki-token-string-expression: #8ce99a;
+  --shiki-token-punctuation: #ced4da;
+  --shiki-token-link: #74c0fc;
+}
--- a/packages/client/ui-trajectory/src/client/TrajectoryCell.module.css
+++ b/packages/client/ui-trajectory/src/client/TrajectoryCell.module.css
@@ -61,6 +61,17 @@
  background: var(--dsw-alias-state-warn-tertiary);
 }

+/* run_code sub-dispatch cells: the business tint plus an indent so the
+   nesting under the parent Tool cell reads at a glance. */
+.tagSubtool {
+  color: var(--dsw-alias-state-business-primary);
+  background: var(--dsw-alias-state-business-tertiary);
+}
+
+.root[data-kind='subtool'] {
+  padding-left: 28px;
+}
+
 .text {
  flex: 1 1 auto;
  min-width: 0;
--- a/packages/client/ui-trajectory/src/client/TrajectoryCell.tsx
+++ b/packages/client/ui-trajectory/src/client/TrajectoryCell.tsx
@@ -4,20 +4,23 @@
 import type { HTMLAttributes } from 'react'
 import css from './TrajectoryCell.module.css'

-/** Closed set of trajectory step kinds (call+result fold into Tool; no Think). */
-export type TrajectoryCellKind = 'user' | 'message' | 'tool'
+/** Closed set of trajectory step kinds (call+result fold into Tool; no Think;
+ *  subtool = one run_code sub-dispatch nested under its Tool cell). */
+export type TrajectoryCellKind = 'user' | 'message' | 'tool' | 'subtool'

 /** Display label per kind (matches the design tags). */
 const KIND_LABEL: Record<TrajectoryCellKind, string> = {
  user: 'User',
  message: 'Message',
  tool: 'Tool',
+  subtool: 'Sub',
 }

 const TAG_CLASS: Record<TrajectoryCellKind, string> = {
  user: css.tagUser!,
  message: css.tagMessage!,
  tool: css.tagTool!,
+  subtool: css.tagSubtool!,
 }

 export interface TrajectoryCellProps extends HTMLAttributes<HTMLDivElement> {
--- a/packages/client/ui-trajectory/src/client/TrajectoryView.tsx
+++ b/packages/client/ui-trajectory/src/client/TrajectoryView.tsx
@@ -12,9 +12,10 @@ export function TrajectoryView({ useSession }: ConvViewProps) {
  const nodes = useSession((s) => s.nodes)
  const partial = useSession((s) => s.partial)
  const runningCalls = useSession((s) => s.runningCalls)
+  const codeDispatches = useSession((s) => s.codeDispatches)
  const turns = useMemo(
-    () => deriveTrajectoryLayout({ nodes, partial, runningCalls }),
-    [nodes, partial, runningCalls],
+    () => deriveTrajectoryLayout({ nodes, partial, runningCalls, codeDispatches }),
+    [nodes, partial, runningCalls, codeDispatches],
  )
  if (turns.length === 0) {
    return <div className={css.root}><p className={css.empty}>暂无轨迹数据</p></div>
--- a/packages/client/ui-trajectory/src/client/WaterfallView.tsx
+++ b/packages/client/ui-trajectory/src/client/WaterfallView.tsx
@@ -1,16 +1,20 @@
-// WaterfallView: P-I placeholder body for the waterfall tab — span stats
-// header over node-count bars per turn standing in for duration lanes (no
-// timing data yet; deviation ledger #3 defers real rendering to P-III).
+// WaterfallView: span stats header over per-turn node-count lanes (P-I
+// stand-in for duration lanes; deviation ledger #3). run_code turns
+// additionally draw TRUTHFUL sub-call lanes: the dispatch start/settle pair
+// carries per-sub-call wall time, so each sub-span's width is its real
+// duration against the parent turn's dispatch window.

 import { useMemo } from 'react'
 import type { ConvViewProps } from '@deepseek-ai/dsh-client-ui-conversation/client'
-import { deriveSpans } from './spans.ts'
+import { deriveSpans, deriveSubSpans } from './spans.ts'
 import { TrajectoryStatsHeader } from './TrajectoryStatsHeader.tsx'
 import css from './views.module.css'

 /** Bar width scale: px per node, clamped so tiny windows still show a bar. */
 const PX_PER_NODE = 14
 const MIN_BAR_PX = 8
+/** Sub-span lane width budget (the parent window scales into this). */
+const SUB_LANE_PX = 220

 /** Optional density override (test/standalone knob; the register site passes nothing). */
 export interface WaterfallExtraProps {
@@ -21,27 +25,48 @@ export interface WaterfallExtraProps {
 export function WaterfallView({ useSession, pxPerNode }: ConvViewProps & WaterfallExtraProps) {
  const scale = pxPerNode ?? PX_PER_NODE
  const nodes = useSession((s) => s.nodes)
+  const codeDispatches = useSession((s) => s.codeDispatches)
  const spans = useMemo(() => deriveSpans(nodes), [nodes])
+  const subSpans = useMemo(() => deriveSubSpans(nodes, codeDispatches), [nodes, codeDispatches])
  if (spans.length === 0) return <div className={css.root}><p className={css.empty}>暂无瀑布数据</p></div>
  return (
    <>
      <TrajectoryStatsHeader useSession={useSession} />
      <div className={css.root}>
        {spans.map((span, i) => (
-          <div key={span.turn} className={css.row} style={{ paddingLeft: i * 12 }}>
-            <span className={css.turnTag}>turn {span.turn}</span>
-            <span
-              className={css.bar}
-              style={{ width: Math.max(span.nodes * scale, MIN_BAR_PX) }}
-              title={`${span.nodes} nodes`}
-            />
-            {span.calls > 0 && (
+          <div key={span.turn}>
+            <div className={css.row} style={{ paddingLeft: i * 12 }}>
+              <span className={css.turnTag}>turn {span.turn}</span>
              <span
-                className={`${css.bar} ${css.barCalls}`}
-                style={{ width: Math.max(span.calls * scale, MIN_BAR_PX) }}
-                title={`${span.calls} tool calls`}
+                className={css.bar}
+                style={{ width: Math.max(span.nodes * scale, MIN_BAR_PX) }}
+                title={`${span.nodes} nodes`}
              />
-            )}
+              {span.calls > 0 && (
+                <span
+                  className={`${css.bar} ${css.barCalls}`}
+                  style={{ width: Math.max(span.calls * scale, MIN_BAR_PX) }}
+                  title={`${span.calls} tool calls`}
+                />
+              )}
+            </div>
+            {(subSpans.get(span.turn) ?? []).map((lane) => (
+              <div key={lane.callId} className={css.subRow} data-subspan style={{ paddingLeft: i * 12 + 24 }}>
+                <span className={css.subTag}>{lane.name}</span>
+                <span
+                  className={`${css.bar} ${css.barSub}`}
+                  data-timing={lane.timing}
+                  style={{
+                    marginLeft: Math.round(lane.offsetFraction * SUB_LANE_PX),
+                    width: Math.max(Math.round(lane.widthFraction * SUB_LANE_PX), 4),
+                  }}
+                  title={lane.timing === 'measured'
+                    /* durationMs is non-null exactly when timing is measured. */
+                    ? `${lane.name} · ${((lane.durationMs ?? 0) / 1000).toFixed(2)}s`
+                    : lane.timing === 'running' ? `${lane.name} · running` : `${lane.name} · duration unknown`}
+                />
+              </div>
+            ))}
          </div>
        ))}
      </div>
--- a/packages/client/ui-trajectory/src/client/layout.ts
+++ b/packages/client/ui-trajectory/src/client/layout.ts
@@ -4,6 +4,7 @@
 */
 import type {
  AssistantMessageNode,
+  CodeSubCall,
  ConversationSnapshot,
  ToolResultNode,
 } from '@deepseek-ai/dsh-client-runtime/client'
@@ -27,6 +28,8 @@ export interface TrajectoryLayoutInput {
  nodes: ConversationSnapshot['nodes']
  partial: ConversationSnapshot['partial']
  runningCalls: ConversationSnapshot['runningCalls']
+  /** run_code sub-dispatches by parent callId (sub-cells nest under the parent Tool cell). */
+  codeDispatches: ConversationSnapshot['codeDispatches']
 }

 interface UsageLike {
@@ -49,7 +52,7 @@ interface LaidCell {
 * @returns turns ordered by first appearance.
 */
 export function deriveTrajectoryLayout(input: TrajectoryLayoutInput): readonly TrajectoryTurnModel[] {
-  const { nodes, partial, runningCalls } = input
+  const { nodes, partial, runningCalls, codeDispatches } = input
  const resultByCall = indexResults(nodes)
  const turns = new Map<number, { message: LaidCell[]; steps: Map<number, LaidCell[]> }>()
  let index = 0
@@ -96,7 +99,7 @@ export function deriveTrajectoryLayout(input: TrajectoryLayoutInput): readonly T
      continue
    }
    if (node.kind === 'assistant') {
-      const laidList = expandAssistant(node, index + 1, prevAbsTime, resultByCall)
+      const laidList = withSubCalls(expandAssistant(node, index + 1, prevAbsTime, resultByCall), codeDispatches)
      for (const laid of laidList) {
        if (node.step > 0) pushStep(node.turn, node.step, laid)
        else pushMessage(node.turn, laid)
@@ -128,6 +131,10 @@ export function deriveTrajectoryLayout(input: TrajectoryLayoutInput): readonly T
            timeSeconds: durationSeconds(node.time, node.callTime),
          },
        })
+        for (const laid of expandSubCalls(codeDispatches.get(node.callId), index)) {
+          pushStep(0, 1, laid)
+          index = laid.cell.index
+        }
      }
      prevAbsTime = finiteTime(node.time) ?? prevAbsTime
    }
@@ -161,6 +168,10 @@ export function deriveTrajectoryLayout(input: TrajectoryLayoutInput): readonly T
        timeSeconds: null,
      },
    })
+    for (const laid of expandSubCalls(codeDispatches.get(call.callId), index)) {
+      pushStep(call.turn, call.step > 0 ? call.step : 1, laid)
+      index = laid.cell.index
+    }
  }

  // Orphan turn-0 cells (orphaned tools / steering turn 0) fold into Turn 1.
@@ -387,6 +398,53 @@ function collectCallIds(
  return ids
 }

+
+
+/** Interleave each tool cell's run_code sub-dispatch cells right after it, reindexing followers. */
+function withSubCalls(laidList: LaidCell[], codeDispatches: ConversationSnapshot['codeDispatches']): LaidCell[] {
+  if (codeDispatches.size === 0) return laidList
+  const out: LaidCell[] = []
+  let index = laidList[0] !== undefined ? laidList[0].cell.index - 1 : 0
+  for (const laid of laidList) {
+    out.push({ ...laid, cell: { ...laid.cell, index: ++index } })
+    if (laid.callId === undefined) continue
+    for (const sub of expandSubCalls(codeDispatches.get(laid.callId), index)) {
+      out.push(sub)
+      index = sub.cell.index
+    }
+  }
+  return out
+}
+
+/** Sub-dispatch cells for one run_code parent, in start order (running = null duration). */
+function expandSubCalls(
+  subs: readonly CodeSubCall[] | undefined,
+  startIndex: number,
+): LaidCell[] {
+  if (subs === undefined || subs.length === 0) return []
+  const out: LaidCell[] = []
+  let index = startIndex
+  for (const sub of subs) {
+    const settled = 'kind' in sub
+    out.push({
+      absTime: settled ? finiteTime(sub.callTime ?? sub.time) : finiteTime(sub.time),
+      toolName: settled ? sub.call?.name ?? sub.callId : sub.name,
+      callId: sub.callId,
+      cell: {
+        index: ++index,
+        kind: 'subtool',
+        text: settled
+          ? (sub.call !== null ? summarizeCall(sub.call.name, sub.call.argsRaw) : summarizeResult(sub))
+          : summarizeCall(sub.name, sub.argsRaw),
+        // PR3's start/settle pair carries per-sub-call wall time; a running
+        // (unsettled) or pre-pair log entry shows the em dash.
+        timeSeconds: settled ? durationSeconds(sub.time, sub.callTime) : null,
+      },
+    })
+  }
+  return out
+}
+
 function summarizeCall(name: string, argsRaw: string): string {
  const args = argsRaw.replace(/\s+/g, ' ').trim()
  if (args === '') return name
--- a/packages/client/ui-trajectory/src/client/spans.ts
+++ b/packages/client/ui-trajectory/src/client/spans.ts
@@ -5,6 +5,24 @@
 */
 import type { ConversationNode, ConversationSnapshot } from '@deepseek-ai/dsh-client-runtime/client'

+/** One run_code sub-dispatch lane in the waterfall: real timing off the start/settle pair. */
+export interface SubSpanLane {
+  callId: string
+  name: string
+  /** Wall duration in ms; null unless both endpoints were observed (`timing: 'measured'`). */
+  durationMs: number | null
+  /**
+   * Timing provenance: `measured` = start/settle pair observed; `running` =
+   * start seen, settle pending; `unknown` = settle-only replay window (the
+   * start fell outside), so no duration claim is possible.
+   */
+  timing: 'measured' | 'running' | 'unknown'
+  /** Start offset as a fraction of the parent turn's dispatch window [0, 1). */
+  offsetFraction: number
+  /** Width as a fraction of the window (running lanes extend to the window end). */
+  widthFraction: number
+}
+
 /** One turn's worth of activity, folded from the snapshot node window. */
 export interface TurnSpan {
  turn: number
@@ -69,3 +87,61 @@ export function deriveSpanStats(spans: readonly TurnSpan[]): SpanStats {
 function hasTurn(node: ConversationNode): node is ConversationNode & { turn: number } {
  return node.kind === 'assistant' || node.kind === 'steering'
 }
+
+/**
+ * Fold the dispatch index into per-turn sub-span lanes with REAL timing: each
+ * lane's offset/width scale against its parent turn's dispatch window (first
+ * start → last settle). Running (unsettled) lanes extend to the window end
+ * with a null duration.
+ * @param nodes - snapshot nodes (locates each parent run_code call's turn).
+ * @param codeDispatches - the snapshot's dispatch index.
+ * @returns lanes keyed by turn, in start order.
+ */
+export function deriveSubSpans(
+  nodes: ConversationSnapshot['nodes'],
+  codeDispatches: ConversationSnapshot['codeDispatches'],
+): ReadonlyMap<number, readonly SubSpanLane[]> {
+  const out = new Map<number, SubSpanLane[]>()
+  if (codeDispatches.size === 0) return out
+  const turnByCall = new Map<string, number>()
+  let currentTurn = 0
+  for (const node of nodes) {
+    if (node.kind === 'assistant' || node.kind === 'steering') currentTurn = node.turn
+    if (node.kind === 'tool-result') turnByCall.set(node.callId, currentTurn)
+  }
+  for (const [parent, subs] of codeDispatches) {
+    if (subs.length === 0) continue
+    const turn = turnByCall.get(parent) ?? currentTurn
+    // A settle-only entry (callTime null: its start fell outside the replay
+    // window) anchors the window by its settle time — a real observation —
+    // but must never masquerade as a measured zero-duration span.
+    const starts: number[] = []
+    const ends: number[] = []
+    for (const sub of subs) {
+      const settled = 'kind' in sub
+      const start = settled ? sub.callTime ?? sub.time : sub.time
+      starts.push(start)
+      ends.push(settled ? sub.time : start)
+    }
+    const windowStart = Math.min(...starts)
+    const windowEnd = Math.max(...ends, windowStart + 1)
+    const windowSpan = windowEnd - windowStart
+    const lanes: SubSpanLane[] = subs.map((sub, i) => {
+      const settled = 'kind' in sub
+      const timing = settled ? (sub.callTime === null ? 'unknown' as const : 'measured' as const) : 'running' as const
+      const start = starts[i] ?? windowStart
+      const end = settled ? sub.time : windowEnd
+      return {
+        callId: sub.callId,
+        name: settled ? sub.call?.name ?? sub.callId : sub.name,
+        durationMs: timing === 'measured' ? Math.max(0, end - start) : null,
+        timing,
+        offsetFraction: (start - windowStart) / windowSpan,
+        widthFraction: Math.max((end - start) / windowSpan, 0.02),
+      }
+    })
+    const existing = out.get(turn) ?? []
+    out.set(turn, [...existing, ...lanes])
+  }
+  return out
+}
--- a/packages/client/ui-trajectory/src/client/views.module.css
+++ b/packages/client/ui-trajectory/src/client/views.module.css
@@ -45,3 +45,38 @@
  color: var(--dsw-alias-label-caption);
  font: var(--dsw-font-xs-13);
 }
+
+/* run_code sub-span lanes: one row per sub-dispatch under its turn row,
+   offset/width scaled to the dispatch window (real wall time). A running
+   lane pulses via reduced opacity until its settle arrives. */
+.subRow {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  margin-top: 2px;
+}
+
+.subTag {
+  flex: none;
+  width: 88px;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  color: var(--dsw-alias-label-tertiary);
+  font: var(--dsw-font-xs-13);
+}
+
+.barSub {
+  height: 8px;
+  background: var(--dsw-alias-state-business-primary);
+}
+
+.barSub[data-timing='running'] {
+  opacity: 0.45;
+}
+
+/* Settle-only replay entries: no measured span — hollow, not a solid bar. */
+.barSub[data-timing='unknown'] {
+  background: transparent;
+  border: 1px dashed var(--dsw-alias-state-business-primary);
+}
--- a/packages/client/ui-trajectory/tests/layout.spec.tsx
+++ b/packages/client/ui-trajectory/tests/layout.spec.tsx
@@ -70,7 +70,7 @@ describe('deriveTrajectoryLayout', () => {
        content: [{ type: 'text', text: 'a.txt' }], isError: false, callView: null, resultView: null,
      },
    ] as unknown as ConversationSnapshot['nodes']
-    const turns = deriveTrajectoryLayout({ nodes, partial: null, runningCalls: [] })
+    const turns = deriveTrajectoryLayout({ codeDispatches: new Map(), nodes, partial: null, runningCalls: [] })
    expect(turns).toHaveLength(1)
    expect(turns[0]?.turn).toBe(1)
    const kinds = turns[0]?.groups.flatMap((g) => g.cells.map((c) => c.kind))
@@ -86,6 +86,7 @@ describe('deriveTrajectoryLayout', () => {

  it('adds runningCalls not already present and leaves their time blank', () => {
    const turns = deriveTrajectoryLayout({
+      codeDispatches: new Map(),
      nodes: [] as unknown as ConversationSnapshot['nodes'],
      partial: null,
      runningCalls: [{
@@ -111,7 +112,7 @@ describe('deriveTrajectoryLayout', () => {
        usage: { inputTokens: 1, outputTokens: 2, reasoningTokens: 3 },
      },
    ] as unknown as ConversationSnapshot['nodes']
-    const turns = deriveTrajectoryLayout({ nodes, partial: null, runningCalls: [] })
+    const turns = deriveTrajectoryLayout({ codeDispatches: new Map(), nodes, partial: null, runningCalls: [] })
    const cells = turns[0]?.groups.flatMap((g) => g.cells) ?? []
    expect(cells.find((c) => c.kind === 'message')?.timeSeconds).toBeNull()
    expect(turns[0]?.groups.find((g) => g.title === 'Step 1')?.description).toBeUndefined()
@@ -137,7 +138,7 @@ describe('deriveTrajectoryLayout', () => {
        content: [], isError: false, callView: null, resultView: null,
      },
    ] as unknown as ConversationSnapshot['nodes']
-    const turns = deriveTrajectoryLayout({ nodes, partial: null, runningCalls: [] })
+    const turns = deriveTrajectoryLayout({ codeDispatches: new Map(), nodes, partial: null, runningCalls: [] })
    expect(turns[0]?.groups[0]?.description).toBe('2.9s bash×2')
  })

@@ -154,7 +155,7 @@ describe('deriveTrajectoryLayout', () => {
        blocks: [{ kind: 'text', text: 'ok2' }],
      },
    ] as unknown as ConversationSnapshot['nodes']
-    const turns = deriveTrajectoryLayout({ nodes, partial: null, runningCalls: [] })
+    const turns = deriveTrajectoryLayout({ codeDispatches: new Map(), nodes, partial: null, runningCalls: [] })
    expect(turns.map((t) => t.turn)).toEqual([1, 2])
    expect(turns[0]?.groups.flatMap((g) => g.cells.map((c) => c.text))).toEqual(['first', 'ok1'])
    expect(turns[1]?.groups.flatMap((g) => g.cells.map((c) => c.text))).toEqual(['second', 'ok2'])
@@ -168,7 +169,7 @@ describe('deriveTrajectoryLayout', () => {
        usage: { inputTokens: 11, outputTokens: 22, reasoningTokens: 3 },
      },
    ] as unknown as ConversationSnapshot['nodes']
-    const turns = deriveTrajectoryLayout({ nodes, partial: null, runningCalls: [] })
+    const turns = deriveTrajectoryLayout({ codeDispatches: new Map(), nodes, partial: null, runningCalls: [] })
    const message = turns[0]?.groups.flatMap((g) => g.cells).find((c) => c.kind === 'message')
    expect(message).toMatchObject({
      text: '', input: 11, output: 22, think: 3,
@@ -196,7 +197,7 @@ describe('deriveTrajectoryLayout', () => {
        blocks: [{ kind: 'text', text: 'done' }],
      },
    ] as unknown as ConversationSnapshot['nodes']
-    const turns = deriveTrajectoryLayout({ nodes, partial: null, runningCalls: [] })
+    const turns = deriveTrajectoryLayout({ codeDispatches: new Map(), nodes, partial: null, runningCalls: [] })
    const message = turns[0]?.groups
      .flatMap((g) => g.cells)
      .find((c) => c.kind === 'message' && c.text === 'done')
@@ -204,3 +205,51 @@ describe('deriveTrajectoryLayout', () => {
    expect(message?.timeSeconds).toBe(1)
  })
 })
+
+describe('run_code sub-dispatch cells', () => {
+  const runCodeNodes = [
+    {
+      kind: 'assistant', seq: 2, time: 6_000, turn: 1, step: 1,
+      blocks: [
+        { kind: 'tool-call', callId: 'p1', name: 'run_code', argsRaw: '{"code":"…","description":"批量读取"}' },
+      ],
+    },
+    {
+      kind: 'tool-result', seq: 3, time: 9_000, callId: 'p1',
+      call: { name: 'run_code', argsRaw: '{"code":"…","description":"批量读取"}' }, callTime: 6_200,
+      content: [{ type: 'text', text: 'done' }], isError: false, callView: null, resultView: null,
+    },
+  ] as unknown as ConversationSnapshot['nodes']
+
+  const settledSub = (n: number, name: string, start: number, end: number) => ({
+    kind: 'tool-result' as const, seq: 100 + n, time: end,
+    callId: `p1:code:${n}`,
+    call: { name, argsRaw: '{"x":1}' }, callTime: start,
+    content: [{ type: 'text' as const, text: 'ok' }], isError: false, callView: null, resultView: null,
+  })
+
+  it('nests settled sub-cells after their parent Tool cell with real durations', () => {
+    const codeDispatches = new Map([['p1', [
+      settledSub(1, 'bash', 6_300, 7_300),
+      settledSub(2, 'read', 7_300, 7_800),
+    ]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const turns = deriveTrajectoryLayout({ codeDispatches, nodes: runCodeNodes, partial: null, runningCalls: [] })
+    const cells = turns[0]!.groups.flatMap((g) => g.cells)
+    expect(cells.map((c) => c.kind)).toEqual(['tool', 'subtool', 'subtool'])
+    // Sequential indexes across the interleave; durations from the pair times.
+    expect(cells.map((c) => c.index)).toEqual([1, 2, 3])
+    expect(cells[1]).toMatchObject({ text: 'bash · {"x":1}', timeSeconds: 1 })
+    expect(cells[2]).toMatchObject({ timeSeconds: 0.5 })
+  })
+
+  it('a running (unsettled) sub-call renders a subtool cell with blank time', () => {
+    const running = {
+      callId: 'p1:code:1', name: 'grep', argsRaw: '{"pattern":"x"}',
+      turn: 0, step: 0, time: 6_400, callView: null,
+    }
+    const codeDispatches = new Map([['p1', [running]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const turns = deriveTrajectoryLayout({ codeDispatches, nodes: runCodeNodes, partial: null, runningCalls: [] })
+    const sub = turns[0]!.groups.flatMap((g) => g.cells).find((c) => c.kind === 'subtool')
+    expect(sub).toMatchObject({ text: 'grep · {"pattern":"x"}', timeSeconds: null })
+  })
+})
--- a/packages/client/ui-trajectory/tests/views.spec.tsx
+++ b/packages/client/ui-trajectory/tests/views.spec.tsx
@@ -21,7 +21,7 @@ import type { ConvViewProps, ViewTab } from '@deepseek-ai/dsh-client-ui-conversa
 import { ConversationRoot, type ConversationRootProps } from '@deepseek-ai/dsh-client-ui-conversation/src/client/skeleton/ConversationRoot.tsx'
 import { createChatStore } from '@deepseek-ai/dsh-client-ui-conversation/src/client/stores.ts'
 import { apply, inject } from '@deepseek-ai/dsh-client-ui-trajectory/client'
-import { deriveSpans, deriveSpanStats } from '@deepseek-ai/dsh-client-ui-trajectory/src/client/spans.ts'
+import { deriveSpans, deriveSpanStats, deriveSubSpans } from '@deepseek-ai/dsh-client-ui-trajectory/src/client/spans.ts'
 import { TrajectoryStatsHeader } from '@deepseek-ai/dsh-client-ui-trajectory/src/client/TrajectoryStatsHeader.tsx'
 import { TrajectoryView } from '@deepseek-ai/dsh-client-ui-trajectory/src/client/TrajectoryView.tsx'
 import { WaterfallView } from '@deepseek-ai/dsh-client-ui-trajectory/src/client/WaterfallView.tsx'
@@ -54,7 +54,7 @@ const NODES = [

 function fakeSession(nodes: ConversationSnapshot['nodes']) {
  const store = createSnapshotStore({
-    nodes, partial: null, runningCalls: [] as ConversationSnapshot['runningCalls'],
+    nodes, partial: null, runningCalls: [] as ConversationSnapshot['runningCalls'], codeDispatches: new Map(),
  })
  return { store, useSession: bindSnapshotSelector(store) as unknown as UseSession<ConversationSnapshot> }
 }
@@ -117,7 +117,7 @@ function tabsOf(slots: SlotsService): ViewTab[] {
 function mount(slots: SlotsService, nodes: ConversationSnapshot['nodes'] = NODES) {
  const sessionSnapshot = createSnapshotStore({
    running: false, removed: false, promptError: null, nodes,
-    partial: null, runningCalls: [] as ConversationSnapshot['runningCalls'],
+    partial: null, runningCalls: [] as ConversationSnapshot['runningCalls'], codeDispatches: new Map(),
  })
  const useSession = bindSnapshotSelector(sessionSnapshot) as unknown as UseSession<ConversationSnapshot>
  const chat = createChatStore().create()
@@ -260,3 +260,115 @@ describe('node half', () => {
    expect(nodeApply()).toBeUndefined()
  })
 })
+
+describe('deriveSubSpans (waterfall lanes)', () => {
+  const dispatchNodes = [
+    { kind: 'assistant', seq: 2, time: 6_000, turn: 3, step: 1, blocks: [] },
+    {
+      kind: 'tool-result', seq: 3, time: 9_000, callId: 'p1',
+      call: { name: 'run_code', argsRaw: '{}' }, callTime: 6_100,
+      content: [], isError: false, callView: null, resultView: null,
+    },
+  ] as unknown as ConversationSnapshot['nodes']
+
+  it('scales settled lanes into the dispatch window with real durations', () => {
+    const codeDispatches = new Map([['p1', [
+      {
+        kind: 'tool-result', seq: 101, time: 7_000, callId: 'p1:code:1',
+        call: { name: 'bash', argsRaw: '{}' }, callTime: 6_200,
+        content: [], isError: false, callView: null, resultView: null,
+      },
+      {
+        kind: 'tool-result', seq: 102, time: 8_200, callId: 'p1:code:2',
+        call: { name: 'read', argsRaw: '{}' }, callTime: 7_000,
+        content: [], isError: false, callView: null, resultView: null,
+      },
+    ]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const lanes = deriveSubSpans(dispatchNodes, codeDispatches)
+    const turn3 = lanes.get(3)
+    expect(turn3).toHaveLength(2)
+    // Window = 6200..8200 (2000ms). bash: 0..0.4; read: 0.4..1.0.
+    expect(turn3?.[0]).toMatchObject({ name: 'bash', durationMs: 800, timing: 'measured', offsetFraction: 0 })
+    expect(turn3?.[0]?.widthFraction).toBeCloseTo(0.4)
+    expect(turn3?.[1]).toMatchObject({ name: 'read', durationMs: 1200 })
+    expect(turn3?.[1]?.offsetFraction).toBeCloseTo(0.4)
+  })
+
+  it('a running lane extends to the window end with a null duration', () => {
+    const codeDispatches = new Map([['p1', [
+      {
+        kind: 'tool-result', seq: 101, time: 8_000, callId: 'p1:code:1',
+        call: { name: 'bash', argsRaw: '{}' }, callTime: 6_200,
+        content: [], isError: false, callView: null, resultView: null,
+      },
+      { callId: 'p1:code:2', name: 'grep', argsRaw: '{}', turn: 0, step: 0, time: 7_000, callView: null },
+    ]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const lanes = deriveSubSpans(dispatchNodes, codeDispatches)
+    const running = lanes.get(3)?.find((lane) => lane.name === 'grep')
+    expect(running).toMatchObject({ durationMs: null, timing: 'running' })
+    // Extends from its start to the window end.
+    expect(running!.offsetFraction + running!.widthFraction).toBeCloseTo(1)
+  })
+
+  it('a settle-only entry (null callTime) is unknown timing, never a measured 0 ms', () => {
+    const codeDispatches = new Map([['p1', [
+      {
+        kind: 'tool-result', seq: 101, time: 8_000, callId: 'p1:code:1',
+        call: { name: 'bash', argsRaw: '{}' }, callTime: null,
+        content: [], isError: false, callView: null, resultView: null,
+      },
+    ]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const lane = deriveSubSpans(dispatchNodes, codeDispatches).get(3)?.[0]
+    expect(lane).toMatchObject({ durationMs: null, timing: 'unknown' })
+  })
+
+  it('waterfall renders sub-span lanes under the owning turn row', () => {
+    const codeDispatches = new Map([['p1', [
+      {
+        kind: 'tool-result', seq: 101, time: 8_000, callId: 'p1:code:1',
+        call: { name: 'bash', argsRaw: '{}' }, callTime: 6_200,
+        content: [], isError: false, callView: null, resultView: null,
+      },
+    ]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const store = createSnapshotStore({
+      nodes: dispatchNodes, partial: null,
+      runningCalls: [] as ConversationSnapshot['runningCalls'], codeDispatches,
+    })
+    const props = {
+      sessionId: SID,
+      useSession: bindSnapshotSelector(store) as unknown as UseSession<ConversationSnapshot>,
+      useSessions: emptySessions(),
+      useWorkspaces: emptyWorkspaces(),
+    } as unknown as ConvViewProps
+    const view = render(createElement(WaterfallView as FC<ConvViewProps>, props))
+    const lane = view.container.querySelector('[data-subspan]')
+    expect(lane).not.toBeNull()
+    expect(lane!.textContent).toContain('bash')
+    expect(lane!.querySelector('[title*="1.80s"]')).not.toBeNull()
+    expect(lane!.querySelector('[data-timing="measured"]')).not.toBeNull()
+  })
+
+  it('waterfall labels a settle-only lane as duration unknown', () => {
+    const codeDispatches = new Map([['p1', [
+      {
+        kind: 'tool-result', seq: 101, time: 8_000, callId: 'p1:code:1',
+        call: { name: 'read', argsRaw: '{}' }, callTime: null,
+        content: [], isError: false, callView: null, resultView: null,
+      },
+    ]]]) as unknown as ConversationSnapshot['codeDispatches']
+    const store = createSnapshotStore({
+      nodes: dispatchNodes, partial: null,
+      runningCalls: [] as ConversationSnapshot['runningCalls'], codeDispatches,
+    })
+    const props = {
+      sessionId: SID,
+      useSession: bindSnapshotSelector(store) as unknown as UseSession<ConversationSnapshot>,
+      useSessions: emptySessions(),
+      useWorkspaces: emptyWorkspaces(),
+    } as unknown as ConvViewProps
+    const view = render(createElement(WaterfallView as FC<ConvViewProps>, props))
+    const bar = view.container.querySelector('[data-timing="unknown"]')
+    expect(bar).not.toBeNull()
+    expect(bar!.getAttribute('title')).toContain('duration unknown')
+  })
+})
--- a/packages/client/web/src/base.css
+++ b/packages/client/web/src/base.css
@@ -1,9 +1,10 @@
 /* Shell-owned global base: full-height mount plus the theme token sheets.
- * The three ui-theme sheets are the sole token source (--dsw-*); the shell
+ * The four ui-theme sheets are the sole token source (--dsw-*); the shell
 * links them here so tokens exist before any plugin CSS lands. */
@import '@deepseek-ai/dsh-client-ui-theme/styles/base.css';
@import '@deepseek-ai/dsh-client-ui-theme/styles/design-platform.css';
@import '@deepseek-ai/dsh-client-ui-theme/styles/gradient-shadow-text.css';
+@import '@deepseek-ai/dsh-client-ui-theme/styles/shiki.css';

 html,
 body,
--- a/packages/cordis/tool-cordis/src/api-catalog.ts
+++ b/packages/cordis/tool-cordis/src/api-catalog.ts
@@ -1218,6 +1218,13 @@ export const EVENT_API: readonly EventApiEntry[] = [
    jsDoc: '/**\n * A tool was registered or unregistered, or a scoped restriction changed\n * (the available tool set changed — possibly for one scope only). An\n * UNFILTERED registry-subject notification, deliberately not scope-filtered\n * dispatch: a global change concerns every agent\'s next assembly, so a\n * scoped listener subscribing here sees every change, not just its own\n * scope\'s.\n * @mode emit\n */',
    summary: 'A tool was registered or unregistered, or a scoped restriction changed (the available tool set changed — possibly for one scope only).',
  },
+  {
+    name: 'tools/code-dispatch-log',
+    mode: 'waterfall',
+    signature: '\'tools/code-dispatch-log\'(this: Scoped<ToolRegistry>, dispatch: CodeDispatchLog, next: () => Promise<ContentBlock[]>): Promise<ContentBlock[]>',
+    jsDoc: '/**\n * Shape the DURABLE LOG COPY of one `run_code` sub-dispatch outcome before\n * the bridge appends its `tool/code-dispatch` event. `next()` keeps the\n * content unchanged; a listener may return replacement blocks (e.g. the\n * spill policy\'s preview + locator for an oversized text result). Only the\n * logged copy is affected — the program already received the complete\n * value, and the model sees neither. A throwing listener is contained:\n * the bridge falls back to logging the unshaped content.\n * Scope-filtered dispatch (`@deepseek-ai/dsh-scope`): agent-scoped listeners receive only that agent\'s dispatches.\n * @param dispatch - the parent execution, sub-call identity, and the settled content to log.\n * @mode waterfall\n */',
+    summary: 'Shape the DURABLE LOG COPY of one `run_code` sub-dispatch outcome before the bridge appends its `tool/code-dispatch` event.',
+  },
  {
    name: 'tools/execute',
    mode: 'waterfall',
--- a/packages/core/scope/src/scoped-events.generated.ts
+++ b/packages/core/scope/src/scoped-events.generated.ts
@@ -35,6 +35,7 @@ const scopedSubjectResolvers: Readonly<Record<string, ScopedSubjectResolver | nu
  'subagent/end': null,
  'subagent/start': null,
  'system-prompt/assemble': args => (args[1] as Record<string, unknown>)['scope'],
+  'tools/code-dispatch-log': args => (args[0] as Record<string, unknown>)['agent'],
  'tools/execute': args => (args[0] as Record<string, unknown>)['agent'],
  'tools/post-execute': args => (args[0] as Record<string, unknown>)['agent'],
  'tools/pre-execute': args => (args[0] as Record<string, unknown>)['agent'],
--- a/packages/core/scope/tests/invariant.spec.ts
+++ b/packages/core/scope/tests/invariant.spec.ts
@@ -63,6 +63,7 @@ describe('scoped-dispatch invariants', () => {
      ['approval/request', [{ agent, toolName: 'echo' }, () => Promise.resolve('unavailable')]],
      ['goal/changed', [agent, { operation: 'create', ref: { id: 'goal-a', revision: 1 } }]],
      ['system-prompt/assemble', [[], { scope: agent }]],
+      ['tools/code-dispatch-log', [{ exec: { callId: 'c', name: 't', arguments: {} }, agent, subCallId: 'c:code:1', name: 't', isError: false, content: [] }, () => Promise.resolve([])]],
      ['tools/execute', [{ callId: 'c', name: 't', arguments: {}, agent }, () => Promise.resolve({ content: [], isError: false })]],
      ['tools/post-execute', [{ callId: 'c', name: 't', arguments: {}, agent }, { content: [], isError: false }, () => Promise.resolve({ kind: 'accept' })]],
      ['tools/pre-execute', [{ callId: 'c', name: 't', arguments: {}, agent }, () => Promise.resolve({ kind: 'allow' })]],
--- a/packages/core/tools/README.i18n.yaml
+++ b/packages/core/tools/README.i18n.yaml
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write
-README.md: 893ba3afef71ea0fb6b4bca267d929b220e3d506
-README.zh.md: cefb35bbc3556a1aca97d9e2fc5f8e6e06391e2e
+README.md: 3a9082f99663586936ce1eafaf09264a91798e28
+README.zh.md: 763ae1195d2c7a2606806970a3ba6371114d9d1c
--- a/packages/core/tools/README.md
+++ b/packages/core/tools/README.md
@@ -191,5 +191,5 @@ Append-only; newly visible content follows the reusable request prefix and does
 - **Caller-defined subagent and workflow structured outputs remain object-rooted** — this is a consumer-level guard; the shared schema vocabulary and tool outputs support every JSON root.
 - **`timeoutMs` on a definition is declarative only** — the registry never enforces deadlines; enforcement requires the `@deepseek-ai/dsh-timeout-policy` wrapper.
 - **Code Mode is TypeScript-only and the presentation mode is service-wide** — `mode: code`/`both` rejects prompt assembly unless `ctx.codeRuntime.language === 'typescript'`; scoped restrictions/shadows still choose each agent's visible bindings, but one tool cannot be native-only while another is code-only.
- **Code Mode intermediate values are execution-local and unbounded by bytes** — the canonical typed values cannot be reconstructed from session replay and may exhaust process or worker memory; only the outer `run_code` output has the worker's configurable hard cap. The rendered `content` of every sub-call IS logged verbatim on `tool/code-dispatch`, uncapped and outside spill policy, so programs that read huge files grow the session log by the same bytes (spill integration for the logged copy is deferred work).
+- **Code Mode intermediate values are execution-local and unbounded by bytes** — the canonical typed values cannot be reconstructed from session replay and may exhaust process or worker memory; only the outer `run_code` output has the worker's configurable hard cap. The durable log copy of each sub-call IS bounded: the `tools/code-dispatch-log` waterfall lets the spill policy replace an oversized `tool/code-dispatch` content with a preview + locator ([rationale](../../../.agents/notes/implemented/feature/2026-07-26-code-dispatch-log-spill.md)).
 - **`run_code` state is fresh per run** — a persistent REPL-style kernel is rejected for the MVP (cross-call state would be invisible to the log); see [the Code Mode Agent Note](../../../.agents/notes/implemented/feature/2026-06-15-code-mode.md).
--- a/packages/core/tools/README.zh.md
+++ b/packages/core/tools/README.zh.md
@@ -191,5 +191,5 @@ The available tools:
 - **调用方定义的 subagent 与工作流结构化输出仍要求对象根**：这是消费方层面的守卫；共享 schema 词汇和工具输出支持任意 JSON 根。
 - **定义上的 `timeoutMs` 仅为声明**：注册表绝不会强制执行截止时间；要强制执行，必须使用 `@deepseek-ai/dsh-timeout-policy` 包装层。
 - **Code Mode 只支持 TypeScript，且呈现模式在服务内统一**：`mode: code`/`both` 会拒绝组装提示词，除非 `ctx.codeRuntime.language === 'typescript'`；作用域限制／遮蔽仍会选择每个 agent 的可见绑定，但不能让一个工具仅使用 Native，而另一个仅使用 Code。
- **Code Mode 中间值只存在于执行局部，且没有字节上限**：这些规范的类型化值无法从会话回放重建，并可能耗尽进程或 worker 内存；只有外层 `run_code` 输出受 worker 可配置的硬上限约束。每个子调用渲染后的 `content` 确实会原样记录在 `tool/code-dispatch` 中，不受字节上限约束，也不在 spill 策略范围内。因此，读取超大文件的程序会使会话日志增加等量字节（日志中的副本尚未接入 spill，相关工作留待后续完成）。
+- **Code Mode 中间值只存在于执行局部，且没有字节上限**：这些规范的类型化值无法从会话回放重建，并可能耗尽进程或 worker 内存；只有外层 `run_code` 输出受 worker 可配置的硬上限约束。每个子调用的持久日志副本则**有**上限：`tools/code-dispatch-log` waterfall 允许 spill 策略把过大的 `tool/code-dispatch` 内容替换为预览加定位符（[原理](../../../.agents/notes/implemented/feature/2026-07-26-code-dispatch-log-spill.md)）。
 - **每次运行都会获得全新的 `run_code` 状态**：MVP 不采用持久 REPL 风格内核（跨调用状态不会出现在日志中）；参见 [Code Mode Agent Note](../../../.agents/notes/implemented/feature/2026-06-15-code-mode.md)。
--- a/packages/core/tools/src/code-mode.ts
+++ b/packages/core/tools/src/code-mode.ts
@@ -13,7 +13,7 @@ import { snapshotJsonValue } from '@deepseek-ai/dsh-session'
 import type { JsonValue } from '@deepseek-ai/dsh-session'
 import { defineTool } from './schema.ts'
 import { TOOL_REGISTRY_SCHEDULER } from './index.ts'
-import type { ToolDefinition, ToolExecutionResult, ToolRegistry, ToolRunContext } from './index.ts'
+import type { CodeDispatchLog, ToolDefinition, ToolExecutionResult, ToolRegistry, ToolRunContext } from './index.ts'

 declare module '@deepseek-ai/dsh-session' {
  interface SessionEventMap {
@@ -186,6 +186,20 @@ function renderValue(value: JsonValue): string {
 /** Canonical value returned by the outer Code Mode transport. */
 type RunCodeOutput = { logs: string[]; result?: JsonValue }

+/**
+ * Registry-private capabilities the bridge receives at construction — the
+ * `requireRuntime` idiom: operations only the owning registry can mint stay
+ * off its public service surface and flow here as closures instead.
+ */
+export interface RunCodeBridgeOptions {
+  /** Resolves `ctx.codeRuntime` or throws the loud misconfiguration error (shared with the registry's assembly-time checks). */
+  requireRuntime: () => CodeRuntime
+  /** The run's overlap cap for parallel-classified sub-calls (the registry passes its validated `maxParallelSubCalls`). */
+  maxParallel: number
+  /** Runs the contained `tools/code-dispatch-log` waterfall over one settled sub-dispatch (the registry's private invoker). */
+  shapeDispatchLog: (dispatch: CodeDispatchLog) => Promise<ContentBlock[]>
+}
+
 /**
 * Build the `run_code` {@link ToolDefinition}: required `code` and
 * `description` parameters, executed through the dispatch bridge described
@@ -194,13 +208,11 @@ type RunCodeOutput = { logs: string[]; result?: JsonValue }
 * outside the filterable global/scoped capability layers.
 * @param registry - the owning registry (sub-calls go through its `execute`,
 *   bindings cover its registered tools).
- * @param requireRuntime - resolves `ctx.codeRuntime` or throws the loud
- *   misconfiguration error (shared with the registry's assembly-time checks).
- * @param maxParallel - the run's overlap cap for parallel-classified
- *   sub-calls (the registry passes its validated `maxParallelSubCalls`).
+ * @param options - the registry-private capabilities described above.
 * @returns the registry-ready definition.
 */
-export function createRunCodeTool(registry: ToolRegistry, requireRuntime: () => CodeRuntime, maxParallel: number): ToolDefinition {
+export function createRunCodeTool(registry: ToolRegistry, options: RunCodeBridgeOptions): ToolDefinition {
+  const { requireRuntime, maxParallel, shapeDispatchLog } = options
  return defineTool({
    name: RUN_CODE_NAME,
    description:
@@ -279,6 +291,8 @@ export function createRunCodeTool(registry: ToolRegistry, requireRuntime: () =>
      }
      const pendingQueue: PendingDispatch[] = []
      const inFlight = new Set<Promise<void>>()
+      /** Tracked settle-event side work (log shaping + append), drained at run settlement. */
+      const logWork = new Set<Promise<void>>()
      const commitQueue: PendingDispatch[] = []
      let exclusiveActive = false
      let driving = false
@@ -357,6 +371,9 @@ export function createRunCodeTool(registry: ToolRegistry, requireRuntime: () =>
        // entries, awaits the live pool, and drains the ordered commit lane —
        // including a commit already in progress when the program returned.
        await drive()
+        // Every settle's shaped append lands inside the open run_code turn
+        // (tasks self-remove on settlement).
+        while (logWork.size > 0) await Promise.allSettled([...logWork])
      }

      // Read through a call, not a bare property: the abort state genuinely
@@ -387,22 +404,41 @@ export function createRunCodeTool(registry: ToolRegistry, requireRuntime: () =>
            | { kind: 'post-result' | 'final-result'; exec: ToolRunContext; result: ToolExecutionResult }
            | undefined
          const settle = (result: ToolExecutionResult): void => {
-            exec.agent?.session.append('tool/code-dispatch', {
-              parentCallId: exec.callId,
-              subCallId,
-              name,
-              // The SIBLING parse of the dispatched value: byte-identical JSON,
-              // but a separate object — a tool mutating its args cannot desync
-              // this record from what it actually received.
-              arguments: normalized.logged,
-              isError: result.isError,
-              // The registry deep-froze this projection at result finalization;
-              // append snapshots it again, so the log copy stays detached.
-              content: result.content,
-            })
+            // The program gets its value NOW: log shaping (e.g. a spill
+            // backend) must never delay the binding or occupy a dispatch
+            // slot. The shaped append is tracked side work; the run's
+            // settlement drains logWork so every settle event still lands
+            // inside the open turn (shapeDispatchLog is contained, so this
+            // chain cannot reject).
            resolve(result.isError
              ? { isError: true, message: result.error.message }
              : { isError: false, value: result.value })
+            const agent = exec.agent
+            if (agent === undefined) return
+            const task: Promise<void> = (async () => {
+              // The durable copy may be reshaped (e.g. spilled to a preview +
+              // locator) by the log-shaping waterfall; the program's value
+              // and the model contract are untouched.
+              const logged = await shapeDispatchLog({
+                exec, agent, subCallId, name, isError: result.isError,
+                // The registry deep-froze this projection at result
+                // finalization; append snapshots the final copy again, so
+                // the log stays detached.
+                content: result.content,
+              })
+              agent.session.append('tool/code-dispatch', {
+                parentCallId: exec.callId,
+                subCallId,
+                name,
+                // The SIBLING parse of the dispatched value: byte-identical JSON,
+                // but a separate object — a tool mutating its args cannot desync
+                // this record from what it actually received.
+                arguments: normalized.logged,
+                isError: result.isError,
+                content: logged,
+              })
+            })().finally(() => { logWork.delete(task) })
+            logWork.add(task)
          }
          pendingQueue.push({
            flight: Promise.resolve(),
@@ -444,6 +480,12 @@ export function createRunCodeTool(registry: ToolRegistry, requireRuntime: () =>
                exec.deferContext(context)
              }
              settle(result)
+              // Backpressure on the shaped-append side channel: pending log
+              // tasks (each retaining a full result while a slow backend
+              // stores it) are bounded by the pool cap — beyond it the
+              // ordered lane waits, so later sub-calls cannot start and
+              // pending I/O/memory cannot grow without bound.
+              while (logWork.size > maxParallel) await Promise.race(logWork)
            },
          })
          wakeup()
--- a/packages/core/tools/src/index.ts
+++ b/packages/core/tools/src/index.ts
@@ -123,6 +123,19 @@ declare module 'cordis' {
     * @mode waterfall
     */
    'tools/post-execute'(this: Scoped<ToolRegistry>, exec: ToolExecution, result: Readonly<ToolExecutionResult>, next: () => Promise<PostToolDecision>): Promise<PostToolDecision>
+    /**
+     * Shape the DURABLE LOG COPY of one `run_code` sub-dispatch outcome before
+     * the bridge appends its `tool/code-dispatch` event. `next()` keeps the
+     * content unchanged; a listener may return replacement blocks (e.g. the
+     * spill policy's preview + locator for an oversized text result). Only the
+     * logged copy is affected — the program already received the complete
+     * value, and the model sees neither. A throwing listener is contained:
+     * the bridge falls back to logging the unshaped content.
+     * Scope-filtered dispatch (`@deepseek-ai/dsh-scope`): agent-scoped listeners receive only that agent's dispatches.
+     * @param dispatch - the parent execution, sub-call identity, and the settled content to log.
+     * @mode waterfall
+     */
+    'tools/code-dispatch-log'(this: Scoped<ToolRegistry>, dispatch: CodeDispatchLog, next: () => Promise<ContentBlock[]>): Promise<ContentBlock[]>
    /**
     * Observe the frozen, lossless-JSON final outcome. Listener failures are contained.
     * Scope-filtered dispatch (`@deepseek-ai/dsh-scope`): keyed by `exec.agent`.
@@ -272,6 +285,30 @@ export type ToolExecutionMode =
  | { kind: 'parallel' }
  | { kind: 'exclusive' }

+/**
+ * One settled `run_code` sub-dispatch about to be logged, as seen by the
+ * `tools/code-dispatch-log` waterfall: the parent execution (session owner,
+ * outer call identity), the sub-call identity, and the outcome whose durable
+ * copy a listener may reshape. `content` is the RENDERED result projection
+ * (what a native `tool/result` would carry) — the program itself received
+ * the structured `value` (or just the error message on failure); only the
+ * `tool/code-dispatch` event's copy changes.
+ */
+export interface CodeDispatchLog {
+  /** The outer `run_code` execution. */
+  readonly exec: ToolExecution
+  /** The calling agent (the scope routing key and the spill owner), when the outer call has one. */
+  readonly agent?: Agent
+  /** Deterministic sub-call id (`<parent>:code:<n>`). */
+  readonly subCallId: CallId
+  /** The dispatched sub-tool name. */
+  readonly name: string
+  /** Whether the sub-call settled as an error. */
+  readonly isError: boolean
+  /** The sub-call's complete model-facing content (the settle event's default payload). */
+  readonly content: ContentBlock[]
+}
+
 /**
 * One pending tool call inside the registry pipeline. Parsed arguments cross
 * one lossless-JSON materialization boundary before policy and are deep-frozen;
@@ -690,7 +727,11 @@ export class ToolRegistry extends Service {
    // the filterable global/scoped capability layers.
    this.codeTransport = this.mode === 'native'
      ? undefined
-      : createRunCodeTool(this, () => this.requireCodeRuntime(), resolveMaxParallelSubCalls(config.maxParallelSubCalls))
+      : createRunCodeTool(this, {
+        requireRuntime: () => this.requireCodeRuntime(),
+        maxParallel: resolveMaxParallelSubCalls(config.maxParallelSubCalls),
+        shapeDispatchLog: dispatch => this.shapeDispatchLog(dispatch),
+      })
    ctx.systemPrompt.tools(context => this.wireSchemas(context.scope))
    if (this.mode !== 'native') {
      ctx.systemPrompt.section({
@@ -941,6 +982,27 @@ export class ToolRegistry extends Service {
    }
  }

+  /**
+   * Run the `tools/code-dispatch-log` waterfall over one settled sub-dispatch
+   * and return the content the bridge should log on `tool/code-dispatch`.
+   * Contained: a throwing listener falls back to the unshaped content — log
+   * shaping must never fail the dispatch or lose the settle event. Private:
+   * the ONE consumer is the `run_code` bridge this registry constructs, which
+   * receives it as a capability parameter (the `requireRuntime` idiom) — the
+   * waterfall, not this invoker, is the public extension seam.
+   */
+  private async shapeDispatchLog(dispatch: CodeDispatchLog): Promise<ContentBlock[]> {
+    try {
+      return await this.ctx.waterfall(
+        scopeTarget(this, dispatch.agent), 'tools/code-dispatch-log', dispatch,
+        () => Promise.resolve(dispatch.content),
+      )
+    } catch (error: unknown) {
+      this.ctx.logger.warn(`tools: code-dispatch-log listener failed for ${dispatch.name}: ${errorMessage(error)}; logging the unshaped content`)
+      return dispatch.content
+    }
+  }
+
  /**
   * Execute through pre-policy, guards, around-dispatch, post-policy,
   * definition-owned content finalization, and final notification. Tool and
--- a/packages/core/tools/tests/code-mode.spec.ts
+++ b/packages/core/tools/tests/code-mode.spec.ts
@@ -801,6 +801,21 @@ describe('the run_code dispatch bridge', () => {
    expect(result.content[0]).toEqual({ type: 'text', text: 'caught: deliberate failure' })
  })

+  it('a throwing tools/code-dispatch-log listener is contained: the unshaped content is logged', async () => {
+    const { ctx, runtime } = await setup({ mode: 'code' })
+    registerEcho(ctx)
+    ctx.on('tools/code-dispatch-log', () => { throw new Error('shaper exploded') })
+    const { agent, events } = fakeAgent()
+    runtime.behavior = async (request) => {
+      const value = await request.bindings[0]!.functions.echo!({ value: 'x' })
+      return { logs: [], value: value as string }
+    }
+    const result = await runCode(ctx, 'program', { agent })
+    expect(result.isError).toBe(false)
+    const settle = events.find(event => event.type === 'tool/code-dispatch')
+    expect(settle?.data).toMatchObject({ name: 'echo', isError: false, content: [{ type: 'text', text: 'echo:x' }] })
+  })
+
  it('a throwing tools/pre-execute listener settles the sub-call without post-execute', async () => {
    const { ctx, runtime } = await setup({ mode: 'code' })
    const calls = registerEcho(ctx)
--- a/packages/spill/spill-policy/README.i18n.yaml
+++ b/packages/spill/spill-policy/README.i18n.yaml
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write
-README.md: ed72e0d73ec83a6cb8620245e3793770ce622ab3
-README.zh.md: db0be6153465ee64bce7f9ab9a2a3c1dddf8b53d
+README.md: 7638f62c0426964d83d7a635ebf73d0f117abd78
+README.zh.md: bf6d52a8c6b2b2e56052075af4a2150fba357f84
--- a/packages/spill/spill-policy/README.md
+++ b/packages/spill/spill-policy/README.md
@@ -15,7 +15,7 @@ This plugin registers **no service** and owns no storage or preview mechanics: p
 ## Behavior

 1. Let the tool run (delegates via `next()`, so it bounds whatever a downstream hook accepted).
-2. Skip nested executions (`exec.parent` is present), accepted value replacements (the registry must revalidate and rerender them), `read` (avoids a `read → spill → read again` loop), and any non-`accept` decision (a `block`'s corrective feedback passes through).
+2. Skip nested executions (`exec.parent` is present — their DURABLE copy is bounded by the dispatch-log arm below), accepted value replacements (the registry must revalidate and rerender them), `read` (avoids a `read → spill → read again` loop), and any non-`accept` decision (a `block`'s corrective feedback passes through).
 3. Flatten the accepted content only when it is **plain text** (all `text` blocks); a result with any non-text block is left untouched.
 4. If its UTF-8 size is `≤ maxInlineBytes`, leave it unchanged.
 5. Otherwise save the full text and replace the result with a preview + this notice, sized so the whole replacement (preview + blank line + notice) stays within `maxInlineBytes` — the notice's byte cost is reserved out of the budget, so the preview shrinks to fit and the model-facing result never exceeds the cap:
@@ -30,6 +30,8 @@ This plugin registers **no service** and owns no storage or preview mechanics: p

 **Best-effort:** no session owner, no `ctx.spillStore` backend, or a `saveText` rejection ⇒ the policy logs a warning and returns the original result. A spill failure never turns a successful call into an `isError` or hides the inline result. A successful replacement changes only `content`; the canonical programmatic value is preserved.

+**The dispatch-log arm:** a second listener on `tools/code-dispatch-log` applies the same cap, replacement pipeline, and best-effort fallbacks to the DURABLE copy of each `run_code` sub-call result (artifact label `dispatch`, keyed by the sub-call id). The program's value is untouched — it already crossed the worker boundary whole — and `read` sub-calls are bounded too: a log copy is not model context, so the read-again loop cannot occur, and `read` is precisely the tool that produces huge logs ([rationale](../../../.agents/notes/implemented/feature/2026-07-26-code-dispatch-log-spill.md)).
+
 ## Scope

 The policy sees only the FINAL formatted surface result—not a tool's internal resource or canonical value. If a provider already truncated (e.g. `web-fetch-local.maxBodyChars`), the spill artifact holds the full formatted result the tool returned, not the full original source. Provider/resource caps stay mandatory and separate. `glob`/`grep` own item-level surface spill because their complete acquired values still exist before rendering; bash streams own acquisition-time spill. The generic policy prepends its waterfall listener, then delegates, so ordinary tool-owned asynchronous projections complete before generic byte bounding regardless of plugin load order. See the [tool output spill Agent Note](../../../.agents/notes/implemented/architecture/2026-07-08-tool-output-spill-files.md).
--- a/packages/spill/spill-policy/README.zh.md
+++ b/packages/spill/spill-policy/README.zh.md
@@ -15,7 +15,7 @@
 ## 行为

 1. 允许工具运行（通过 `next()` 委托，因此可以限制任何下游钩子接受的内容）。
-2. 跳过嵌套执行（存在 `exec.parent`）、已接受的值替换（注册表必须重新验证并渲染它们）、`read`（避免 `read → spill → read again` 循环）以及任何非 `accept` 决定（`block` 的纠正反馈会原样通过）。
+2. 跳过嵌套执行（存在 `exec.parent`——其持久副本由下方的 dispatch-log 分支设界）、已接受的值替换（注册表必须重新验证并渲染它们）、`read`（避免 `read → spill → read again` 循环）以及任何非 `accept` 决定（`block` 的纠正反馈会原样通过）。
 3. 仅在已接受的内容为**纯文本**（全部都是 `text` 块）时才将其展平；包含任何非文本块的结果都保持不变。
 4. 如果 UTF-8 大小为 `≤ maxInlineBytes`，则保持不变。
 5. 否则，保存完整文本，并将结果替换为预览和以下通知。系统会调整大小，使整个替换内容（预览、空行和通知）不超过 `maxInlineBytes`：先从预算中保留通知所需字节，再缩小预览以适配剩余空间，因此面向模型的结果绝不会超过上限：
@@ -30,6 +30,8 @@

 **尽力而为**：没有会话 owner、没有 `ctx.spillStore` 后端，或 `saveText` 拒绝 ⇒ 策略记录警告并返回原始结果。spill 失败绝不会将成功调用变为 `isError`，也不会隐藏内联结果。成功替换时只会更改 `content`；规范程序值保持不变。

+**dispatch-log 分支：**注册在 `tools/code-dispatch-log` 上的第二个监听器，把同一套上限、替换流水线与尽力而为的回退应用到每个 `run_code` 子调用结果的持久副本上（工件标签为 `dispatch`，按子调用 id 归档）。程序取得的值不受影响——它早已完整跨过 worker 边界；`read` 子调用同样设界：日志副本不是模型上下文，因此不会发生 read-again 循环，而 `read` 恰恰是最容易产生巨型日志的工具（[原理](../../../.agents/notes/implemented/feature/2026-07-26-code-dispatch-log-spill.md)）。
+
 ## 范围

 该策略只能看到最终格式化接口结果，看不到工具的内部资源或规范值。如果提供方已经截断内容（例如 `web-fetch-local.maxBodyChars`），spill 产物保存的是工具返回的完整格式化结果，而非完整原始源。提供方/资源上限仍必须存在，并且与该策略分离。`glob`/`grep` 负责对项级接口结果执行 spill，因为渲染前仍然存在完整的已获取值；bash 流负责在获取时 spill。通用策略预先注册自己的 waterfall 监听器，然后再委托，因此无论插件加载顺序如何，普通工具拥有的异步投影都会在通用字节限制之前完成。详见[工具输出 spill Agent Note](../../../.agents/notes/implemented/architecture/2026-07-08-tool-output-spill-files.md)。
--- a/packages/spill/spill-policy/src/index.ts
+++ b/packages/spill/spill-policy/src/index.ts
@@ -10,18 +10,26 @@
 * `@deepseek-ai/dsh-retention` (`TextRetainer`), storage is `ctx.spillStore`.
 * The policy only decides WHEN to spill and composes the notice.
 *
+ * A second arm applies the SAME cap to the durable log: the
+ * `tools/code-dispatch-log` waterfall bounds the `tool/code-dispatch` event's
+ * copy of an oversized `run_code` sub-call result (the program's value is
+ * untouched; UIs and replay read the full text through the spill artifact).
+ *
 * ## Deliberately narrow
 *
 * - Omitted `maxInlineBytes` ⇒ the plugin registers nothing (a true no-op).
 * - Plain-text results only: a result carrying any non-text block is left
 *   untouched (the policy knows only the final formatted text, not tool
 *   internals).
- * - Nested composite calls are skipped; only their outer surface result may
- *   become model-facing and spillable.
+ * - Nested composite calls skip the MODEL-facing arm; their durable log copy
+ *   is bounded by the dispatch-log arm instead.
 * - Accepted value replacements pass through for registry revalidation and
 *   rendering; this presentation policy cannot also replace content in the
 *   same mutually exclusive decision.
- * - `read` is skipped to avoid a `read → spill → read again` loop.
+ * - `read` is skipped by the model-facing arm to avoid a
+ *   `read → spill → read again` loop; the dispatch-log arm bounds `read`
+ *   sub-calls too (a log copy is not model context, and `read` is precisely
+ *   the tool that produces huge logs).
 * - Best-effort: no session owner, no `ctx.spillStore` backend, or a save
 *   failure ⇒ log and return the original result. A spill failure must NEVER
 *   turn a successful tool call into an `isError` or hide the inline result.
@@ -42,6 +50,7 @@ import { TextRetainer, describeOmitted } from '@deepseek-ai/dsh-retention'
 import type { Omitted } from '@deepseek-ai/dsh-retention'
 import type { SaveTextSpill, SpillRef } from '@deepseek-ai/dsh-spill'
 import type { SessionId } from '@deepseek-ai/dsh-session'
+import type { CallId } from '@deepseek-ai/dsh-llm'
 import type { PostToolDecision, ToolExecution } from '@deepseek-ai/dsh-tools'
 import type { SpillPolicyExec } from './types.ts'

@@ -108,6 +117,75 @@ export function apply(ctx: Context, config: Config): void {
  if (!Number.isInteger(maxInlineBytes) || maxInlineBytes < 0) {
    throw new Error(`spill-policy: maxInlineBytes must be a non-negative integer (got ${maxInlineBytes})`)
  }
+  // Narrowed once for the nested arms (closure narrowing does not survive awaits).
+  const cap: number = maxInlineBytes
+
+  /**
+   * Spill `text` and build the bounded replacement (preview + notice), or
+   * return `undefined` when the policy must keep the original (no session
+   * owner, no backend, storage failure, or no within-cap replacement).
+   * Shared verbatim by the model-facing post-execute arm and the durable
+   * dispatch-log arm so both produce byte-identical projections.
+   */
+  async function spillReplacement(
+    text: string,
+    totalBytes: number,
+    sessionId: SessionId | undefined,
+    toolName: string,
+    callId: CallId,
+    label: 'result' | 'dispatch',
+  ): Promise<string | undefined> {
+    if (sessionId === undefined) {
+      ctx.logger.warn(`spill-policy: no session owner for ${toolName} ${label}; keeping the inline content`)
+      return undefined
+    }
+    const spillStore = ctx.get('spillStore')
+    if (!spillStore) {
+      ctx.logger.warn('spill-policy: no ctx.spillStore backend loaded; keeping the inline content')
+      return undefined
+    }
+    const save: SaveTextSpill = {
+      owner: { sessionId },
+      source: { toolName, callId, label },
+      suggestedName: `${toolName}.txt`,
+      content: text,
+    }
+    let ref: SpillRef
+    try {
+      ref = await spillStore.saveText(save)
+    } catch (error: unknown) {
+      // Best-effort: a storage failure (permissions, ENOSPC, backend down) must
+      // never fail the call or hide the content — keep the original inline.
+      ctx.logger.warn(`spill-policy: saveText failed for ${toolName}: ${String(error)}; keeping the inline content`)
+      return undefined
+    }
+
+    // Reserve the notice's byte cost INSIDE maxInlineBytes so the replacement
+    // (preview + blank line + notice) never exceeds the documented cap — a naive
+    // preview that spent the whole budget then appended the notice could be
+    // larger than the cap, and for a marginally-over result even larger than the
+    // original. The reservation uses a notice priced at the worst-case omission
+    // count (the full byte total): its digit count bounds the real count's, so
+    // the reserved size is a safe upper bound and the final notice is never
+    // longer than what we reserved. `\n\n` is the 2-byte join.
+    const reserve = Buffer.byteLength(spillNotice({ kind: 'exact', count: totalBytes }, ref), 'utf8') + 2
+    const previewBudget = Math.max(0, cap - reserve)
+    const { text: previewText, omitted } = preview(text, previewBudget)
+    const notice = spillNotice(omitted, ref)
+    const replacedText = previewText.length > 0 ? `${previewText}\n\n${notice}` : notice
+    // Invariant: the policy NEVER emits a replacement larger than the cap. When
+    // the notice alone exceeds maxInlineBytes (a tiny cap or a long spill root),
+    // there is no within-cap replacement, so keep the inline content — spilling
+    // would break the advertised cap. (A within-cap replacement is always
+    // smaller than the original, which is > cap by the entry condition, so this
+    // one check subsumes "not smaller than the original" too. The spill file
+    // already written is a harmless orphan; cleanup is deferred.)
+    if (Buffer.byteLength(replacedText, 'utf8') > cap) {
+      ctx.logger.warn(`spill-policy: spill notice for ${toolName} exceeds maxInlineBytes; keeping the inline content`)
+      return undefined
+    }
+    return replacedText
+  }

  ctx.on('tools/post-execute', async (exec, result, next): Promise<PostToolDecision> => {
    // Delegate first so a downstream listener (e.g. a hook) settles the result;
@@ -124,58 +202,31 @@ export function apply(ctx: Context, config: Config): void {
    const totalBytes = Buffer.byteLength(text, 'utf8')
    if (totalBytes <= maxInlineBytes) return decision

-    const sessionId = ownerSessionId(exec)
-    if (sessionId === undefined) {
-      ctx.logger.warn(`spill-policy: no session owner for ${exec.name} result; keeping the inline result`)
-      return decision
-    }
-    const spillStore = ctx.get('spillStore')
-    if (!spillStore) {
-      ctx.logger.warn('spill-policy: no ctx.spillStore backend loaded; keeping the inline result')
-      return decision
-    }
-
-    const save: SaveTextSpill = {
-      owner: { sessionId },
-      source: { toolName: exec.name, callId: exec.callId, label: 'result' },
-      suggestedName: `${exec.name}.txt`,
-      content: text,
-    }
-    let ref: SpillRef
-    try {
-      ref = await spillStore.saveText(save)
-    } catch (error: unknown) {
-      // Best-effort: a storage failure (permissions, ENOSPC, backend down) must
-      // never fail the call or hide the result — keep the original inline.
-      ctx.logger.warn(`spill-policy: saveText failed for ${exec.name}: ${String(error)}; keeping the inline result`)
-      return decision
-    }
-
-    // Reserve the notice's byte cost INSIDE maxInlineBytes so the replacement
-    // (preview + blank line + notice) never exceeds the documented cap — a naive
-    // preview that spent the whole budget then appended the notice could be
-    // larger than the cap, and for a marginally-over result even larger than the
-    // original. The reservation uses a notice priced at the worst-case omission
-    // count (the full byte total): its digit count bounds the real count's, so
-    // the reserved size is a safe upper bound and the final notice is never
-    // longer than what we reserved. `\n\n` is the 2-byte join.
-    const reserve = Buffer.byteLength(spillNotice({ kind: 'exact', count: totalBytes }, ref), 'utf8') + 2
-    const previewBudget = Math.max(0, maxInlineBytes - reserve)
-    const { text: previewText, omitted } = preview(text, previewBudget)
-    const notice = spillNotice(omitted, ref)
-    const replacedText = previewText.length > 0 ? `${previewText}\n\n${notice}` : notice
-    // Invariant: the policy NEVER emits a replacement larger than the cap. When
-    // the notice alone exceeds maxInlineBytes (a tiny cap or a long spill root),
-    // there is no within-cap replacement, so keep the inline result — spilling
-    // would break the advertised context cap. (A within-cap replacement is
-    // always smaller than the original, which is > cap by the entry condition,
-    // so this one check subsumes "not smaller than the original" too. The spill
-    // file already written is a harmless orphan; cleanup is deferred.)
-    if (Buffer.byteLength(replacedText, 'utf8') > maxInlineBytes) {
-      ctx.logger.warn(`spill-policy: spill notice for ${exec.name} exceeds maxInlineBytes; keeping the inline result`)
-      return decision
-    }
+    const replacedText = await spillReplacement(text, totalBytes, ownerSessionId(exec), exec.name, exec.callId, 'result')
+    if (replacedText === undefined) return decision
    const replaced: ContentBlock[] = [{ type: 'text', text: replacedText }]
    return { kind: 'accept', content: replaced, ...decision.additionalContexts ? { additionalContexts: decision.additionalContexts } : {} }
  }, { prepend: true })
+
+  // The durable-log arm: bound the `tool/code-dispatch` event's copy of an
+  // oversized sub-call result the same way the model-facing arm bounds an
+  // outer result. The program's returned value is untouched (it already
+  // crossed the worker boundary whole); only the session log's copy shrinks
+  // to preview + locator, so replay and UIs read the full text through the
+  // spill artifact exactly as they do for spilled native results.
+  ctx.on('tools/code-dispatch-log', async (dispatch, next): Promise<ContentBlock[]> => {
+    const content = await next()
+    // `read` sub-calls spill too: the log copy is not model context, so the
+    // read → spill → read-again loop the post-execute arm avoids cannot
+    // happen here, and read is precisely the tool that produces huge logs.
+    const text = flattenPlainText(content)
+    if (text === undefined) return content
+    const totalBytes = Buffer.byteLength(text, 'utf8')
+    if (totalBytes <= maxInlineBytes) return content
+
+    const replacedText = await spillReplacement(
+      text, totalBytes, ownerSessionId(dispatch.exec), dispatch.name, dispatch.subCallId, 'dispatch')
+    if (replacedText === undefined) return content
+    return [{ type: 'text', text: replacedText }]
+  }, { prepend: true })
 }
--- a/packages/spill/spill-policy/tests/spill-policy.spec.ts
+++ b/packages/spill/spill-policy/tests/spill-policy.spec.ts
@@ -16,6 +16,7 @@ import type { ContentBlock } from '@deepseek-ai/dsh-llm'
 import { SessionId } from '@deepseek-ai/dsh-session'
 import SystemPrompt from '@deepseek-ai/dsh-system-prompt'
 import ToolRegistry, { defineContentToolFixture } from '@deepseek-ai/dsh-tools'
+import type { ToolDefinition } from '@deepseek-ai/dsh-tools'
 import type { PostToolDecision, ToolExecution, ToolExecutionToken } from '@deepseek-ai/dsh-tools'
 import { SpillLocator, SpillStore } from '@deepseek-ai/dsh-spill'
 import type { SaveTextSpill, SpillRef } from '@deepseek-ai/dsh-spill'
@@ -28,9 +29,12 @@ const testToolSignal = new AbortController().signal
 class StubStore extends SpillStore {
  saves: SaveTextSpill[] = []
  fail = false
+  /** Per-save hang hook: each call awaits the returned promise before completing. */
+  gate: (() => Promise<void>) | undefined

  async saveText(input: SaveTextSpill): Promise<SpillRef> {
    if (this.fail) throw new Error('disk full')
+    await this.gate?.()
    this.saves.push(input)
    return {
      locator: SpillLocator(`/spill/${input.suggestedName}`),
@@ -230,6 +234,230 @@ describe('read skip', () => {
  })
 })

+describe('the durable dispatch-log arm', () => {
+  /** Boot code mode + the policy + the worker runtime; run one program via the real bridge. */
+  async function runCodeWith(program: string, maxInlineBytes: number, extraTools: ToolDefinition[] = []) {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    await ctx.plugin(ToolRegistry, { mode: 'code' })
+    await ctx.plugin(StubStore)
+    await ctx.plugin(SpillPolicy, { maxInlineBytes })
+    await ctx.plugin(WorkerCodeRuntime, {})
+    const events: { type: string; data: unknown }[] = []
+    const agent = {
+      session: {
+        header: { id: SessionId('dispatch-spill'), cwd: '/workspace' },
+        append: (type: string, data: unknown) => { events.push({ type, data }) },
+      },
+    }
+    ctx.tools.register(textTool('huge_read', 'H'.repeat(2_000)))
+    ctx.tools.register(textTool('small_read', 'tiny'))
+    for (const tool of extraTools) ctx.tools.register(tool)
+    const result = await ctx.tools.execute({
+      signal: testToolSignal,
+      callId: CallId('parent-1'),
+      name: 'run_code',
+      arguments: { code: program, description: 'Drive dispatch-log spilling' },
+      agent: agent as never,
+    })
+    return { ctx, result, events, spill: ctx.spillStore as StubStore }
+  }
+
+  it('bounds the tool/code-dispatch copy of an oversized sub-result while the program value stays whole', async () => {
+    const { result, events, spill } = await runCodeWith(
+      'const blocks = await tools.huge_read({});\nreturn blocks[0].text.length', 200)
+    expect(result.isError).toBe(false)
+    if (result.isError) throw new Error('expected success')
+    // The program received the COMPLETE text (length 2000), untouched by spill.
+    expect(result.value).toMatchObject({ result: 2_000 })
+    // The durable settle event carries the bounded projection + locator.
+    const settle = events.find(event => event.type === 'tool/code-dispatch')
+    expect(settle).toBeDefined()
+    const logged = (settle!.data as { content: { type: string; text: string }[] }).content
+    expect(logged).toHaveLength(1)
+    const loggedText = logged[0]!.text
+    expect(Buffer.byteLength(loggedText, 'utf8')).toBeLessThanOrEqual(200)
+    expect(loggedText).toContain('Full formatted result stored at: /spill/huge_read.txt')
+    // The artifact holds the full text under the dispatch label and sub-call id.
+    const save = spill.saves.find(entry => entry.source.label === 'dispatch')
+    expect(save).toMatchObject({
+      source: { toolName: 'huge_read', callId: 'parent-1:code:1', label: 'dispatch' },
+    })
+    expect(save?.content).toBe('H'.repeat(2_000))
+  })
+
+  it('leaves a non-text sub-result log unchanged (flatten declines)', async () => {
+    const { events, spill } = await runCodeWith(
+      'return await tools.mixed_read({})', 5, [defineContentToolFixture({
+        name: 'mixed_read',
+        description: 'mixed_read',
+        parameters: {},
+        async execute(): Promise<ContentBlock[]> {
+          return [{ type: 'text', text: 'x'.repeat(100) }, { type: 'reasoning', text: 'why' }]
+        },
+      })])
+    const settle = events.find(event => event.type === 'tool/code-dispatch')
+    expect((settle!.data as { content: unknown[] }).content).toHaveLength(2)
+    expect(spill.saves.filter(entry => entry.source.label === 'dispatch')).toHaveLength(0)
+  })
+
+  it('leaves a within-cap sub-result log untouched and saves nothing for it', async () => {
+    const { events, spill } = await runCodeWith(
+      'return await tools.small_read({})', 200)
+    const settle = events.find(event => event.type === 'tool/code-dispatch')
+    expect((settle!.data as { content: { type: string; text: string }[] }).content)
+      .toEqual([{ type: 'text', text: 'tiny' }])
+    expect(spill.saves.filter(entry => entry.source.label === 'dispatch')).toHaveLength(0)
+  })
+
+  it('a slow spill backend never delays the program value or a later dispatch slot', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    await ctx.plugin(ToolRegistry, { mode: 'code' })
+    await ctx.plugin(StubStore)
+    await ctx.plugin(SpillPolicy, { maxInlineBytes: 100 })
+    await ctx.plugin(WorkerCodeRuntime, {})
+    // A spill backend that hangs until released.
+    let releaseSave!: () => void
+    const gate = new Promise<void>((resolve) => { releaseSave = resolve })
+    const store = ctx.spillStore as StubStore
+    const realSave = store.saveText.bind(store)
+    store.saveText = async (input) => {
+      await gate
+      return realSave(input)
+    }
+    const events: { type: string; data: unknown }[] = []
+    const agent = {
+      session: {
+        header: { id: SessionId('dispatch-slow-spill'), cwd: '/workspace' },
+        append: (type: string, data: unknown) => { events.push({ type, data }) },
+      },
+    }
+    ctx.tools.register(textTool('huge_read', 'H'.repeat(2_000)))
+    ctx.tools.register(textTool('small_read', 'tiny'))
+    let smallAfterHuge = false
+    const runPromise = ctx.tools.execute({
+      signal: testToolSignal,
+      callId: CallId('parent-3'),
+      name: 'run_code',
+      arguments: {
+        // The program takes BOTH values while the spill backend hangs: the
+        // huge read's binding resolves immediately (its logged copy is side
+        // work), so the small read proceeds without waiting.
+        code: 'const big = await tools.huge_read({});\nconst small = await tools.small_read({});\nreturn big[0].text.length + small[0].text.length',
+        description: 'Prove log shaping is off the program path',
+      },
+      agent: agent as never,
+    }).then((result) => {
+      return result
+    })
+    // The run cannot COMPLETE while the settle append is gated (drain waits
+    // for logWork), but the program itself already ran both calls; release
+    // the backend and observe the settle events land inside the turn.
+    await vi.waitFor(() => {
+      // The second dispatch STARTED while the first one's spill hung.
+      smallAfterHuge = events.some(event => event.type === 'tool/code-dispatch-start'
+        && (event.data as { name: string }).name === 'small_read')
+      if (!smallAfterHuge) throw new Error('small_read not started yet')
+    })
+    releaseSave()
+    const result = await runPromise
+    expect(result.isError).toBe(false)
+    if (result.isError) throw new Error('expected success')
+    expect(result.value).toMatchObject({ result: 2_004 })
+    const settles = events.filter(event => event.type === 'tool/code-dispatch')
+    expect(settles).toHaveLength(2)
+    expect(smallAfterHuge).toBe(true)
+  })
+
+  it('a sustained slow backend backpressures the run instead of accumulating unbounded log tasks', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    // Cap 1: once the hung shaped-append backlog exceeds the cap, the ordered
+    // lane holds inside the second commit, so the THIRD dispatch cannot start
+    // until a pending save drains — the bound is observable as its missing
+    // start event.
+    await ctx.plugin(ToolRegistry, { mode: 'code', maxParallelSubCalls: 1 })
+    await ctx.plugin(StubStore)
+    await ctx.plugin(SpillPolicy, { maxInlineBytes: 100 })
+    await ctx.plugin(WorkerCodeRuntime, {})
+    const store = ctx.spillStore as StubStore
+    const releases: (() => void)[] = []
+    store.gate = () => new Promise<void>((resolve) => { releases.push(resolve) })
+    const events: { type: string; data: unknown }[] = []
+    const agent = {
+      session: {
+        header: { id: SessionId('dispatch-spill-bound'), cwd: '/workspace' },
+        append: (type: string, data: unknown) => { events.push({ type, data }) },
+      },
+    }
+    ctx.tools.register(textTool('huge_read', 'H'.repeat(2_000)))
+    const started = (n: number): boolean => events.some(event => event.type === 'tool/code-dispatch-start'
+      && (event.data as { subCallId: string }).subCallId.endsWith(`:code:${n}`))
+    const runPromise = ctx.tools.execute({
+      signal: testToolSignal,
+      callId: CallId('parent-bound'),
+      name: 'run_code',
+      arguments: {
+        code: 'await tools.huge_read({}); await tools.huge_read({}); await tools.huge_read({}); return "done"',
+        description: 'Three oversized reads against a hung backend',
+      },
+      agent: agent as never,
+    })
+    // Two hung saves = backlog above the cap: the lane must hold before
+    // starting dispatch 3.
+    await vi.waitFor(() => {
+      if (releases.length < 2) throw new Error('second hung save not reached yet')
+    })
+    expect(started(2)).toBe(true)
+    expect(started(3)).toBe(false)
+    releases.shift()!()
+    // Draining one pending save releases the lane; dispatch 3 starts.
+    await vi.waitFor(() => {
+      if (!started(3)) throw new Error('third dispatch not started yet')
+    })
+    while (releases.length > 0) releases.shift()!()
+    const result = await runPromise
+    expect(result.isError).toBe(false)
+    await vi.waitFor(() => {
+      if (releases.length > 0) { while (releases.length > 0) releases.shift()!() }
+      if (events.filter(event => event.type === 'tool/code-dispatch').length !== 3) {
+        throw new Error('settle events still pending')
+      }
+    })
+  })
+
+  it('a saveText failure keeps the complete content in the durable log (best-effort)', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    await ctx.plugin(ToolRegistry, { mode: 'code' })
+    await ctx.plugin(StubStore)
+    await ctx.plugin(SpillPolicy, { maxInlineBytes: 100 })
+    await ctx.plugin(WorkerCodeRuntime, {})
+    ;(ctx.spillStore as StubStore).fail = true
+    const warn = vi.spyOn(ctx.logger, 'warn').mockImplementation(() => {})
+    const events: { type: string; data: unknown }[] = []
+    const agent = {
+      session: {
+        header: { id: SessionId('dispatch-spill-fail'), cwd: '/workspace' },
+        append: (type: string, data: unknown) => { events.push({ type, data }) },
+      },
+    }
+    ctx.tools.register(textTool('huge_read', 'H'.repeat(2_000)))
+    const result = await ctx.tools.execute({
+      signal: testToolSignal,
+      callId: CallId('parent-2'),
+      name: 'run_code',
+      arguments: { code: 'return (await tools.huge_read({}))[0].text.length', description: 'Fail the spill backend' },
+      agent: agent as never,
+    })
+    expect(result.isError).toBe(false)
+    const settle = events.find(event => event.type === 'tool/code-dispatch')
+    expect((settle!.data as { content: { text: string }[] }).content[0]!.text).toBe('H'.repeat(2_000))
+    expect(warn).toHaveBeenCalled()
+  })
+})
+
 describe('nested-call skip', () => {
  it('leaves nested composite results complete and spillable only through their outer call', async () => {
    const { ctx, spill } = await setup({ maxInlineBytes: 10 })