Merge remote-tracking branch 'origin/master' into agent-request-messages

docs/core-data-structures/core.md: combined master's canonical tool-order
wording (#196) with this branch's request-advice envelope + wire-order
paragraphs.

packages/core/agent-core/README.md

@@ -35,11 +35,11 @@ This is the [interface/implementation/consumer seam](../../../docs/rfc/implement
 
 ```ts
 import type { Config } from '@deepseek-ai/dsh-agent-core'
-// { agents?, persona? } — the schema is z.intersect([AgentLoop.Config, SystemPrompt.Config]),
+// { agents?, persona?, toolOrder? } — the schema is z.intersect([AgentLoop.Config, SystemPrompt.Config]),
 // so validation and defaulting can never drift from the owners'.
 ```
 
-The bundle FORWARDS each field to the child that owns it: `agents` to `agent-loop` (default `[]`), so each app supplies its own pre-created agents — a stdio app pre-creates a `main`; the ACP app pre-creates none (it creates agents on demand at `session/new`) — and `persona` to `dsh-system-prompt` (default `''`), the deployment's persona section. Forwarding is exactly why the owners can live in the shared spine even though the apps disagree on what to configure.
+The bundle FORWARDS each field to the child that owns it: `agents` to `agent-loop` (default `[]`), so each app supplies its own pre-created agents — a stdio app pre-creates a `main`; the ACP app pre-creates none (it creates agents on demand at `session/new`) — `persona` to `dsh-system-prompt` (default `''`), the deployment's persona section — and `toolOrder` to `dsh-system-prompt` (absent — lexicographic), the explicit model-facing tool order. Forwarding is exactly why the owners can live in the shared spine even though the apps disagree on what to configure.
 
 ## Why a code bundle, not a shared YAML include
 

packages/core/agent-core/src/index.ts

@@ -59,17 +59,20 @@ export const name = 'agent-core'
 /**
  * Bundle config: each field forwarded verbatim to the child that owns it —
  * `agents` to the agent loop (an app that pre-creates no agents, like the ACP
- * bridge, simply omits it), `persona` to the system-prompt plugin (the
- * deployment's persona section). Both are optional INPUT here because each
- * owner's schema supplies the default (`[]` / `''`); the schema is the
- * INTERSECTION of the owners' own schemas, so validation and defaulting can
- * never drift from them.
+ * bridge, simply omits it), `persona` and `toolOrder` to the system-prompt
+ * plugin (the deployment's persona section and the explicit model-facing tool
+ * order). Every field is optional INPUT here because each owner's schema
+ * supplies the default (`[]` / `''` / absent — lexicographic); the schema is
+ * the INTERSECTION of the owners' own schemas, so validation and defaulting
+ * can never drift from them.
  */
 export interface Config {
   /** The agent-loop `agents` list (see dsh-agent-loop's `Config`). */
   agents?: AgentLoopConfig['agents']
   /** The deployment persona (see dsh-system-prompt's `Config`). */
   persona?: SystemPromptConfig['persona']
+  /** The explicit model-facing tool order (see dsh-system-prompt's `Config`). */
+  toolOrder?: SystemPromptConfig['toolOrder']
 }
 
 /** Intersect the owners' schemas so validation + defaulting stay identical. */
@@ -78,11 +81,11 @@ export const Config = z.intersect([AgentLoop.Config, SystemPrompt.Config]) as un
 /**
  * Load the spine. Each `ctx.plugin(...)` mounts one child of the bundle fiber;
  * `agent-loop` receives the forwarded `agents` list and `system-prompt` the
- * forwarded `persona`. Load order is irrelevant (cordis pends each fiber on
- * its `inject` until the services it needs exist), but the listing mirrors the
- * dependency layering for readability: the LLM vocabulary and core registries
- * first, then the dev tripwire and the bash tool consumer, then the loop that
- * drives them.
+ * forwarded `persona` and `toolOrder`. Load order is irrelevant (cordis pends
+ * each fiber on its `inject` until the services it needs exist), but the
+ * listing mirrors the dependency layering for readability: the LLM vocabulary
+ * and core registries first, then the dev tripwire and the bash tool consumer,
+ * then the loop that drives them.
  */
 export function apply(ctx: Context, config: Config): void {
   ctx.plugin(Timer)
@@ -91,8 +94,13 @@ export function apply(ctx: Context, config: Config): void {
   // The forwarded fields are validated + defaulted by this bundle's intersected
   // schema before apply runs, so the ?? fallbacks only narrow the
   // optional-input TYPES — they mirror the owners' schema defaults, never
-  // introduce different ones.
-  ctx.plugin(SystemPrompt, { persona: config.persona ?? '' })
+  // introduce different ones. toolOrder has no owner-supplied default value —
+  // ABSENT means "lexicographic order" — so it is forwarded conditionally
+  // rather than via ??.
+  ctx.plugin(SystemPrompt, {
+    persona: config.persona ?? '',
+    ...config.toolOrder !== undefined ? { toolOrder: config.toolOrder } : {},
+  })
   ctx.plugin(ToolRegistry)
   ctx.plugin(AgentRegistry)
   ctx.plugin(invariants)

packages/core/agent-core/tests/agent-core.spec.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest'
 import { Context } from 'cordis'
 import Loader from '@cordisjs/plugin-loader'
+import { TOOL_ORDER_REST } from '@deepseek-ai/dsh-system-prompt'
 import * as agentCore from '../src/index.ts'
 import { AgentId } from '@deepseek-ai/dsh-agent'
 
@@ -67,6 +68,23 @@ describe('dsh-agent-core bundle', () => {
     await ctx.fiber.dispose()
   })
 
+  it('forwards toolOrder to the system-prompt assembly', async () => {
+    const ctx = await mount({ toolOrder: ['zulu', TOOL_ORDER_REST] })
+    // The bundle's own bash tools pend on the absent `ctx.bash` executor in
+    // this providerless mount, so register two plain tools to order.
+    for (const name of ['alpha', 'zulu']) {
+      ctx.get('tools')!.register({
+        name,
+        description: name,
+        parameters: {},
+        execute: async () => [],
+      })
+    }
+    const assembly = await ctx.get('systemPrompt')!.assemble()
+    expect(assembly.tools.map(tool => tool.name)).toEqual(['zulu', 'alpha'])
+    await ctx.fiber.dispose()
+  })
+
   it('re-exports the loop config schema as its own', () => {
     expect(agentCore.Config).toBeDefined()
     expect(agentCore.name).toBe('agent-core')

packages/core/agent-loop/tests/tool-order.spec.ts

@@ -0,0 +1,117 @@
+/**
+ * Loop-level tool-order determinism: the request/header event — and therefore
+ * the frozen request the adapter receives — carries the assembly's canonical
+ * tool order (system-prompt's `toolOrder` config, or lexicographic name
+ * order), regardless of the order tool plugins happened to register in.
+ * Registration order is a plugin-load artifact (concurrent dynamic imports
+ * race), so nothing downstream of the registry may depend on it.
+ */
+
+import { describe, expect, it } from 'vitest'
+import { Context } from 'cordis'
+import LlmService from '@deepseek-ai/dsh-llm'
+import SessionStore, { foldRequestHeader } from '@deepseek-ai/dsh-session'
+import SystemPrompt, { TOOL_ORDER_REST } from '@deepseek-ai/dsh-system-prompt'
+import type { Config as SystemPromptConfig } from '@deepseek-ai/dsh-system-prompt'
+import ToolRegistry, { defineTool } from '@deepseek-ai/dsh-tools'
+import AgentRegistry, { AgentId } from '@deepseek-ai/dsh-agent'
+import AgentLoop, { ReactLoopAgent } from '@deepseek-ai/dsh-agent-loop'
+import { MockAdapter, textResponse } from './mock-adapter.ts'
+
+async function harness(adapter: MockAdapter, toolOrder?: SystemPromptConfig['toolOrder']) {
+  const ctx = new Context()
+  await ctx.plugin(LlmService)
+  await ctx.plugin(SessionStore)
+  await ctx.plugin(SystemPrompt, { persona: 'stable base', ...toolOrder !== undefined ? { toolOrder } : {} })
+  await ctx.plugin(ToolRegistry)
+  await ctx.plugin(AgentRegistry)
+  await ctx.plugin(AgentLoop, { agents: [] })
+  ctx.llm.registerAdapter(['mock'], adapter)
+  return ctx
+}
+
+function waitForIdle(ctx: Context, agent: ReactLoopAgent): Promise<void> {
+  return new Promise((resolve) => {
+    const dispose = ctx.on('agent/status', (subject, status) => {
+      if (subject === agent && status === 'idle') {
+        dispose()
+        resolve()
+      }
+    })
+  })
+}
+
+function registerNamed(ctx: Context, name: string) {
+  ctx.tools.register(defineTool({
+    name,
+    description: `the ${name} tool`,
+    parameters: {},
+    async execute() {
+      return [{ type: 'text', text: name }]
+    },
+  }))
+}
+
+/** Run one text-only turn and return the harness context + agent. */
+async function runTurn(registrationOrder: string[], toolOrder?: SystemPromptConfig['toolOrder']) {
+  const adapter = new MockAdapter([textResponse('done')])
+  const ctx = await harness(adapter, toolOrder)
+  for (const name of registrationOrder) registerNamed(ctx, name)
+  const agent = ctx.agentLoop.create(AgentId('a1'), { model: 'mock' })
+  agent.send([{ type: 'text', text: 'go' }])
+  await waitForIdle(ctx, agent)
+  return { ctx, agent, adapter }
+}
+
+describe('loop-level canonical tool order', () => {
+  it('logs the request/header with tools in canonical order, not registration order', async () => {
+    const { agent, adapter } = await runTurn(['zulu', 'alpha', 'mike'])
+    const header = foldRequestHeader(agent.session.events)
+    expect(header?.tools?.map(tool => tool.name)).toEqual(['alpha', 'mike', 'zulu'])
+    // The dispatched request is built FROM the logged header (whose tools the
+    // assembly already canonicalized) and reaches the adapter deep-frozen —
+    // the marker the reconstruction invariant keys on.
+    expect(adapter.requests[0]?.tools?.map(tool => tool.name)).toEqual(['alpha', 'mike', 'zulu'])
+    expect(Object.isFrozen(adapter.requests[0])).toBe(true)
+    expect(adapter.requests[0]?.sessionId).toBe(agent.session.id)
+  })
+
+  it('produces the same header order for any registration order', async () => {
+    const first = await runTurn(['alpha', 'mike', 'zulu'])
+    const second = await runTurn(['zulu', 'mike', 'alpha'])
+    const names = (run: typeof first) => foldRequestHeader(run.agent.session.events)?.tools?.map(tool => tool.name)
+    expect(names(first)).toEqual(['alpha', 'mike', 'zulu'])
+    expect(names(second)).toEqual(names(first))
+  })
+
+  it('honors a configured toolOrder in the logged header and the dispatched request', async () => {
+    const { agent, adapter } = await runTurn(['alpha', 'zulu', 'mike'], ['zulu', TOOL_ORDER_REST])
+    const header = foldRequestHeader(agent.session.events)
+    expect(header?.tools?.map(tool => tool.name)).toEqual(['zulu', 'alpha', 'mike'])
+    expect(adapter.requests[0]?.tools?.map(tool => tool.name)).toEqual(['zulu', 'alpha', 'mike'])
+    expect(Object.isFrozen(adapter.requests[0])).toBe(true)
+  })
+
+  it('fails the turn — no model request — when toolOrder names an unregistered tool', async () => {
+    // The assemble rejection escapes to runTurn's outer catch: the open turn
+    // closes with an `error` reason (agent/error mirrors it), no step opens,
+    // no request/header is logged, the adapter never sees a request, and the
+    // agent returns to idle — a misconfigured deployment fails every turn
+    // deterministically instead of silently reordering nothing.
+    const adapter = new MockAdapter([textResponse('never sent')])
+    const ctx = await harness(adapter, ['ghost', TOOL_ORDER_REST])
+    registerNamed(ctx, 'alpha')
+    const errors: Error[] = []
+    ctx.on('agent/error', (_agent, _turn, _step, error) => void errors.push(error))
+    const agent = ctx.agentLoop.create(AgentId('a1'), { model: 'mock' })
+    agent.send([{ type: 'text', text: 'go' }])
+    await waitForIdle(ctx, agent)
+    expect(adapter.requests).toHaveLength(0)
+    expect(errors.map(e => e.message)).toEqual(['toolOrder lists unregistered tool "ghost"; registered tools: alpha'])
+    expect(foldRequestHeader(agent.session.events)).toBeUndefined()
+    const end = agent.session.events.find(e => e.type === 'turn/end')
+    expect(end?.type === 'turn/end' && end.data.reason).toMatchObject({ kind: 'error', step: 1 })
+    // The turn is balanced (turn/start → turn/end) with no step events inside.
+    expect(agent.session.events.some(e => e.type === 'step/start')).toBe(false)
+  })
+})

packages/core/system-prompt/README.md

@@ -7,15 +7,16 @@ System prompt assembly registry. Plugins contribute ordered text sections, tool-
 | Key | Default | Meaning |
 |---|---|---|
 | `persona` | `''` | The deployment persona: the ONE deployment-authored prompt fragment, rendered as the order-0 `deployment:persona` section and shared by every agent in the context (subagents included). A template — complete `{{…}}` groups are interpreted strictly against the registered variables (the shipped loop registers `{{model}}`/`{{cwd}}`), with no escape syntax for literal braces yet. Empty ⇒ the section is dropped at render. |
+| `toolOrder` | — | Explicit model-facing tool order, as a list of `ToolSchema.name`s with one `'<unlisted-tools>'` rest entry (`TOOL_ORDER_REST`): listed tools take their listed position, unlisted tools land at the rest entry in lexicographic name order. Absent ⇒ plain lexicographic name order. Applied to the collected tools BEFORE the `system-prompt/assemble` waterfall — like the sections' `order` sort, it canonicalizes what the registry contributed (registration order is a plugin-load artifact), and a waterfall listener that mutates the list owns the determinism of what it emits. Misconfiguration fails loud: a list without exactly one rest entry, or with duplicates, throws at load; a listed name with no registered tool rejects every `assemble()`; a tool provider returning the reserved rest-entry name also rejects. Under the shipped loop the turn fails before any model request. Why a central list and not per-plugin weights: [Explicit model-facing tool order](../../../docs/rfc/implemented/feature/2026-07-06-explicit-tool-order.md). |
 
 ## Service: `SystemPrompt` (ctx key: `systemPrompt`)
 
 ### Public API
 
 - `ctx.systemPrompt.section(section: PromptSection): () => void` Contribute a section. Duplicate names throw. Disposed with the calling fiber.
-- `ctx.systemPrompt.tools(provider: () => ToolSchema[]): () => void` Contribute tool schemas (evaluated at each assembly). Disposed with the calling fiber.
+- `ctx.systemPrompt.tools(provider: () => ToolSchema[]): () => void` Contribute tool schemas (evaluated at each assembly). A provider must not return a schema named `TOOL_ORDER_REST`; that name is reserved for `toolOrder`'s rest entry. Disposed with the calling fiber.
 - `ctx.systemPrompt.variable(name: string, provider: (context) => string | undefined): () => void` Contribute a prompt variable, referenced from section text as `{{name}}`. Duplicate or unreferenceable names throw; `undefined` means "no value for this assembly". Disposed with the calling fiber.
-- `ctx.systemPrompt.assemble(context?: AssembleContext): Promise<PromptAssembly>` Assemble the prompt for one caller. Runs through the `system-prompt/assemble` waterfall.
+- `ctx.systemPrompt.assemble(context?: AssembleContext): Promise<PromptAssembly>` Assemble the prompt for one caller. Runs through the `system-prompt/assemble` waterfall. Rejects when a configured `toolOrder` names a tool no provider contributed, or when a provider returns the reserved rest-entry name.
 
 ### Events
 

packages/core/system-prompt/src/index.ts

@@ -88,7 +88,8 @@ export interface AssembledSection {
  *
  * Tool schemas are part of the assembly by design: "what the model is told it
  * can do" is one coherent thing managed here, even though adapters transmit
- * `tools` as a separate wire field rather than prompt text.
+ * `tools` as a separate wire field rather than prompt text. They arrive in
+ * the canonical model-facing order (see {@link Config.toolOrder}).
  *
  * `variables` carries every registered prompt variable resolved against this
  * assembly's context — key present means registered, `undefined` value means
@@ -110,6 +111,70 @@ const VARIABLE_NAME = /^[a-z][a-z0-9_]*$/
 /** A complete `{{...}}` reference group at the scan position (validated after). */
 const GROUP_AT = /^\{\{([^{}]*)\}\}/
 
+/**
+ * The rest entry for {@link Config.toolOrder}: the position where registered
+ * tools not named in the list are inserted (in lexicographic name order).
+ * Reserved: collected tool schemas using this name are rejected before
+ * ordering, so the marker can never collide with a real model-facing tool.
+ */
+export const TOOL_ORDER_REST = '<unlisted-tools>'
+
+/**
+ * Validate a configured tool-order list's shape at service construction:
+ * the {@link TOOL_ORDER_REST} rest entry exactly once, no duplicate names.
+ * Returns the list (or undefined when unconfigured); throws otherwise,
+ * failing the service at load — a bad order config must never reach an
+ * assembly. Whether every listed name matches a registered tool is checked
+ * at each assembly instead ({@link orderTools}): tool plugins register after
+ * this service constructs, so the tool set does not exist yet here.
+ */
+function validateToolOrder(toolOrder: string[] | undefined): string[] | undefined {
+  if (toolOrder === undefined) return undefined
+  const seen = new Set<string>()
+  for (const name of toolOrder) {
+    if (seen.has(name)) throw new Error(`toolOrder lists "${name}" more than once`)
+    seen.add(name)
+  }
+  if (!seen.has(TOOL_ORDER_REST)) {
+    throw new Error(`toolOrder must contain the "${TOOL_ORDER_REST}" rest entry (where unlisted tools are inserted)`)
+  }
+  return toolOrder
+}
+
+/**
+ * Order collected tool schemas by the validated policy: with no configured
+ * list, plain lexicographic name order; with one, listed names take their
+ * listed position and every unlisted tool lands at the
+ * {@link TOOL_ORDER_REST} rest entry in lexicographic name order. A listed
+ * name with no collected tool throws — misconfiguration fails loud, and this
+ * is the earliest moment the registered tool set exists to check against
+ * (tool plugins register after the service constructs, so load time is too
+ * early): the assembly rejects, failing the caller's turn before any model
+ * request. Never drops a tool, and both sorts are stable, so tools sharing a
+ * name keep their collection order.
+ */
+function orderTools(tools: ToolSchema[], toolOrder: string[] | undefined): ToolSchema[] {
+  const reserved = tools.find(tool => tool.name === TOOL_ORDER_REST)
+  if (reserved !== undefined) {
+    throw new Error(`tool provider returned reserved tool name "${TOOL_ORDER_REST}" (reserved for toolOrder's rest entry)`)
+  }
+  if (toolOrder === undefined) return tools.sort(compareToolNames)
+  const registered = new Set(tools.map(tool => tool.name))
+  const unknown = toolOrder.filter(name => name !== TOOL_ORDER_REST && !registered.has(name))
+  if (unknown.length > 0) {
+    throw new Error(`toolOrder lists unregistered tool${unknown.length > 1 ? 's' : ''} ${unknown.map(name => `"${name}"`).join(', ')}; registered tools: ${[...registered].sort().join(', ') || '(none)'}`)
+  }
+  const listed = new Set(toolOrder)
+  const rest = tools.filter(tool => !listed.has(tool.name)).sort(compareToolNames)
+  return toolOrder.flatMap(name =>
+    name === TOOL_ORDER_REST ? rest : tools.filter(tool => tool.name === name))
+}
+
+/** Lexicographic (code-unit) name comparison — locale-independent, so the order is identical on every machine. */
+function compareToolNames(a: ToolSchema, b: ToolSchema): number {
+  return a.name < b.name ? -1 : a.name > b.name ? 1 : 0
+}
+
 /** Plugin config: the deployment-authored fragment of the system prompt (see {@link Config.persona} for its contract). */
 export interface Config {
   /**
@@ -125,6 +190,29 @@ export interface Config {
    * deployment opens with the harness identity alone.
    */
   persona?: string
+  /**
+   * Explicit model-facing tool order, as a list of `ToolSchema.name`s: listed
+   * tools take their listed position, and tools absent from the list are
+   * inserted at the {@link TOOL_ORDER_REST} (`'<unlisted-tools>'`) entry in
+   * lexicographic name order. A configured list must contain the rest entry
+   * exactly once, no duplicate names, and no name without a registered tool —
+   * a misconfigured order blocks work instead of silently reaching a model
+   * request: shape violations throw at load, and an unregistered name rejects
+   * every assembly. `TOOL_ORDER_REST` is reserved for the list marker and may
+   * not be a collected tool name; such a provider output also rejects the
+   * assembly. The single assembly-time validation rejects either failure
+   * before any model request — the earliest moment the registered tool set
+   * exists to check against, since tool plugins register after this service
+   * constructs. When omitted, tools are ordered lexicographically by name.
+   * Applied to the tools
+   * {@link SystemPrompt.assemble} collects, BEFORE the
+   * `system-prompt/assemble` waterfall — like the sections' `order` sort, it
+   * canonicalizes what the registry contributed (registration order is a
+   * plugin-load artifact); a waterfall listener that mutates the tool list
+   * owns the determinism of what it emits. Rationale (and why not per-plugin
+   * weights): docs/rfc/implemented/feature/2026-07-06-explicit-tool-order.md.
+   */
+  toolOrder?: string[]
 }
 
 /**
@@ -203,14 +291,23 @@ function interpolate(section: AssembledSection, variables: Record<string, string
 export class SystemPrompt extends Service {
   static Config: z<Config> = z.object({
     persona: z.string().default(''),
+    // A schemastery array defaults to [] when omitted, but an omitted
+    // toolOrder must stay absent ("lexicographic order"), not become an
+    // explicitly-configured empty list (which is invalid — it lacks the
+    // rest entry). Forcing the default to undefined keeps the key out of the
+    // validated config; the cast is needed because .default() expects the
+    // array type.
+    toolOrder: z.array(z.string()).default(undefined as unknown as string[]),
   })
 
   private sections: PromptSection[] = []
   private toolProviders: (() => ToolSchema[])[] = []
   private variableProviders = new Map<string, (context: AssembleContext) => string | undefined>()
+  private readonly toolOrder: string[] | undefined
 
   constructor(ctx: Context, public config: Config) {
     super(ctx, 'systemPrompt')
+    this.toolOrder = validateToolOrder(config.toolOrder)
     // The harness-owned openers. They live HERE (not on the loop plugin) so a
     // deployment that swaps in a different loop keeps them: the identity is a
     // harness fact stated ahead of everything, and the persona is the
@@ -266,7 +363,10 @@ export class SystemPrompt extends Service {
   /**
    * Contribute a tool-schema provider that is evaluated at each assembly
    * call (so it can reflect the live registry state). The provider is
-   * removed when the calling fiber is disposed. Emits `system-prompt/change`.
+   * removed when the calling fiber is disposed. A provider must not return a
+   * schema named {@link TOOL_ORDER_REST}; that name is reserved for
+   * {@link Config.toolOrder}'s rest entry and rejects the assembly. Emits
+   * `system-prompt/change`.
    * @param provider - evaluated at every {@link assemble} for fresh schemas.
    * @returns the disposer that removes the provider.
    */
@@ -323,19 +423,28 @@ export class SystemPrompt extends Service {
 
   /**
    * Assemble the current prompt for one caller: section texts are resolved
-   * against `context` and sorted by order, tools collected from all
-   * providers, and every registered variable resolved against `context` into
-   * `assembly.variables`. Tool schemas are deep-cloned because adapters and
-   * request waterfalls may mutate schema objects. Runs through the
-   * `system-prompt/assemble` waterfall, giving listeners the opportunity to
-   * mutate or replace the assembly before it reaches the model. Await the
-   * result before reading the assembly values — waterfall listeners may be
-   * async. Interpolation happens later, in {@link renderPrompt}.
+   * against `context` and sorted by order, tools collected from all providers
+   * and put in the canonical model-facing order ({@link Config.toolOrder}, or
+   * lexicographic name order when unconfigured — provider registration order
+   * is a plugin-load artifact and never reaches the assembly; a configured
+   * order naming a tool no provider contributed rejects the assembly), and every
+   * registered variable resolved against `context` into `assembly.variables`.
+   * Tool schemas are deep-cloned because adapters and request waterfalls may
+   * mutate schema objects. Runs through the `system-prompt/assemble`
+   * waterfall, giving listeners the opportunity to mutate or replace the
+   * assembly before it reaches the model — like the sections' `order` sort,
+   * tool canonicalization happens on the initial assembly, and a listener
+   * owns the determinism of whatever it emits. Await the result before
+   * reading the assembly values — waterfall listeners may be async.
+   * Interpolation happens later, in {@link renderPrompt}.
    * @param context - what this assembly is for (defaults to an empty context;
    *   see {@link AssembleContext}).
    * @returns the assembly after the waterfall has run.
    */
-  assemble(context: AssembleContext = {}): Promise<PromptAssembly> {
+  // async so the misconfigured-toolOrder throw in orderTools surfaces as a
+  // rejection: a Promise-returning method must not throw synchronously
+  // (`assemble().catch(...)` would miss it).
+  async assemble(context: AssembleContext = {}): Promise<PromptAssembly> {
     const variables: Record<string, string | undefined> = {}
     for (const [name, provider] of this.variableProviders) {
       variables[name] = provider(context)
@@ -348,8 +457,10 @@ export class SystemPrompt extends Service {
           text: typeof section.text === 'function' ? section.text(context) : section.text,
         }))
         .sort((a, b) => a.order - b.order),
-      tools: this.toolProviders.flatMap(provider =>
-        provider().map(tool => ({ ...tool, parameters: structuredClone(tool.parameters) }))),
+      tools: orderTools(
+        this.toolProviders.flatMap(provider =>
+          provider().map(tool => ({ ...tool, parameters: structuredClone(tool.parameters) }))),
+        this.toolOrder),
       variables,
     }
     return this.ctx.waterfall(this, 'system-prompt/assemble', assembly, context, () => Promise.resolve(assembly))

packages/core/system-prompt/tests/tool-order.spec.ts

@@ -0,0 +1,115 @@
+import { describe, expect, it } from 'vitest'
+import { Context } from 'cordis'
+import SystemPrompt, { PromptAssembly, TOOL_ORDER_REST } from '@deepseek-ai/dsh-system-prompt'
+import type { ToolSchema } from '@deepseek-ai/dsh-llm'
+
+function tool(name: string, description = name): ToolSchema {
+  return { name, description, parameters: { type: 'object', properties: {} } }
+}
+
+async function mount(config: { persona?: string; toolOrder?: string[] } = {}): Promise<Context> {
+  const ctx = new Context()
+  await ctx.plugin(SystemPrompt, config)
+  return ctx
+}
+
+function names(assembly: PromptAssembly): string[] {
+  return assembly.tools.map(t => t.name)
+}
+
+describe('SystemPrompt tool order', () => {
+  // The ONE place the public constant's value is pinned; everything else
+  // (tests and deployment configs alike) references TOOL_ORDER_REST.
+  it('exports the rest entry as "<unlisted-tools>"', () => {
+    expect(TOOL_ORDER_REST).toBe('<unlisted-tools>')
+  })
+
+  it('assembles tools in lexicographic name order when no toolOrder is configured', async () => {
+    const ctx = await mount()
+    ctx.systemPrompt.tools(() => [tool('charlie'), tool('alpha')])
+    ctx.systemPrompt.tools(() => [tool('bravo')])
+    expect(names(await ctx.systemPrompt.assemble())).toEqual(['alpha', 'bravo', 'charlie'])
+  })
+
+  it('assembles the same order regardless of provider registration order', async () => {
+    const forward = await mount()
+    forward.systemPrompt.tools(() => [tool('alpha')])
+    forward.systemPrompt.tools(() => [tool('zulu')])
+    const backward = await mount()
+    backward.systemPrompt.tools(() => [tool('zulu')])
+    backward.systemPrompt.tools(() => [tool('alpha')])
+    expect(names(await forward.systemPrompt.assemble())).toEqual(['alpha', 'zulu'])
+    expect(names(await backward.systemPrompt.assemble())).toEqual(['alpha', 'zulu'])
+  })
+
+  it('applies a configured toolOrder: listed positions, rest at the rest entry lexicographically', async () => {
+    const ctx = await mount({ toolOrder: ['todo_write', TOOL_ORDER_REST, 'bash'] })
+    ctx.systemPrompt.tools(() => [tool('bash'), tool('echo_b'), tool('todo_write'), tool('echo_a')])
+    expect(names(await ctx.systemPrompt.assemble())).toEqual(['todo_write', 'echo_a', 'echo_b', 'bash'])
+  })
+
+  it('rejects the assembly when toolOrder names a tool that is not registered (misconfiguration blocks work)', async () => {
+    const ctx = await mount({ toolOrder: ['todo_write', 'ghost', TOOL_ORDER_REST, 'wraith'] })
+    ctx.systemPrompt.tools(() => [tool('bash'), tool('todo_write')])
+    await expect(ctx.systemPrompt.assemble()).rejects.toThrow(
+      'toolOrder lists unregistered tools "ghost", "wraith"; registered tools: bash, todo_write')
+  })
+
+  it('names the single unregistered tool when no tools are registered at all', async () => {
+    const ctx = await mount({ toolOrder: ['ghost', TOOL_ORDER_REST] })
+    await expect(ctx.systemPrompt.assemble()).rejects.toThrow(
+      'toolOrder lists unregistered tool "ghost"; registered tools: (none)')
+  })
+
+  it.each([
+    ['without an explicit toolOrder', undefined],
+    ['with only the rest entry configured', [TOOL_ORDER_REST]],
+  ])('rejects a provider tool named like the reserved rest entry %s', async (_case, toolOrder) => {
+    const ctx = await mount(toolOrder === undefined ? {} : { toolOrder })
+    ctx.systemPrompt.tools(() => [tool(TOOL_ORDER_REST)])
+    await expect(ctx.systemPrompt.assemble()).rejects.toThrow(
+      `tool provider returned reserved tool name "${TOOL_ORDER_REST}"`)
+  })
+
+  it('keeps collection order between tools that share a name (stable sort)', async () => {
+    const ctx = await mount()
+    ctx.systemPrompt.tools(() => [tool('dup', 'first'), tool('anchor'), tool('dup', 'second')])
+    const assembly = await ctx.systemPrompt.assemble()
+    expect(assembly.tools.map(t => t.description)).toEqual(['anchor', 'first', 'second'])
+  })
+
+  it('canonicalizes BEFORE the assemble waterfall: listeners see the ordered list and own their own edits', async () => {
+    const ctx = await mount()
+    ctx.systemPrompt.tools(() => [tool('zulu'), tool('alpha')])
+    let seen: string[] | undefined
+    ctx.on('system-prompt/assemble', function (assembly, _context, next) {
+      seen = assembly.tools.map(t => t.name)
+      // A listener-appended tool is NOT re-sorted — same contract as sections:
+      // canonicalization applies to what the registry contributed, and a
+      // listener owns the determinism of what it emits.
+      assembly.tools.push(tool('aardvark'))
+      return next()
+    })
+    const assembly = await ctx.systemPrompt.assemble()
+    expect(seen).toEqual(['alpha', 'zulu'])
+    expect(names(assembly)).toEqual(['alpha', 'zulu', 'aardvark'])
+  })
+
+  it.each([
+    ['an empty list', []],
+    ['a list without the rest entry', ['bash', 'todo_write']],
+  ])('rejects %s at load (the rest entry is required)', async (_case, toolOrder) => {
+    await expect(new Context().plugin(SystemPrompt, { toolOrder })).rejects.toThrow(`must contain the "${TOOL_ORDER_REST}" rest entry`)
+  })
+
+  it.each([
+    ['a duplicate tool name', ['bash', 'bash', TOOL_ORDER_REST]],
+    ['a duplicate rest entry', [TOOL_ORDER_REST, 'bash', TOOL_ORDER_REST]],
+  ])('rejects %s at load', async (_case, toolOrder) => {
+    await expect(new Context().plugin(SystemPrompt, { toolOrder })).rejects.toThrow('more than once')
+  })
+
+  it('throws from direct construction too', () => {
+    expect(() => new SystemPrompt(new Context(), { toolOrder: ['bash'] })).toThrow('rest entry')
+  })
+})

packages/core/tools/README.md

@@ -71,6 +71,12 @@ A `defineTool` tool also **validates the model-generated arguments against its `
 
 See `defineTool`, `validateArgs`, `ToolArgsError`, `SchemaSpec`, `InferArgs`, and `schemaSpecToJsonSchema` in the public API for details.
 
+### Structured-output schema subset
+
+A separate vocabulary for callers that DEMAND a machine-readable value from an agent — the subagent seam's `SubagentStartRequest.outputSchema` (and, by extension, a workflow's `agent({ schema })`). Unlike `SchemaSpec` (the author-facing DSL for tool parameters), a `StructuredOutputSchema` is an object-rooted **raw JSON Schema subset** as data: it travels verbatim to the model as a forced tool's `parameters`, and the produced value is validated against it.
+
+The subset is deliberately narrow and REJECTS LOUD outside it — accepting a keyword the validator doesn't enforce would validate less than the schema promises (accepted-then-ignored). Supported: single-string `type` (`object`/`array`/`string`/`number`/`integer`/`boolean`/`null`; type arrays rejected), `properties`/`required`/`additionalProperties` (boolean; every `required` key must be declared), `items`, scalar-only `enum`/`const`; annotations (`description`/`title`/`default`/`examples`) are ignored but must still be JSON data. `assertSupportedOutputSchema(schema)` throws `OutputSchemaError` (`code: 'UNSUPPORTED_SCHEMA'`, listing every violation) for anything else; `validateStructuredValue(schema, value)` returns path-qualified violations (empty = valid, total — never throws).
+
 ### Tool-owned UI presentation
 
 A tool owns how ITS calls render in a UI (an editor's tool-call card, a CLI log line) — a UI plugin must NOT special-case tool names. A `ToolDefinition` may declare two optional, pure, display-only methods that return a **`card`-tagged render intent** (a discriminated union — a tool declares its card kind once and a UI bridge switches on `card`):

packages/core/tools/src/index.ts

@@ -28,6 +28,16 @@ export {
   type JsonSchemaObject,
 } from './schema.ts'
 
+export {
+  assertSupportedOutputSchema,
+  validateStructuredValue,
+  OutputSchemaError,
+  type StructuredOutputSchema,
+  type StructuredSchemaNode,
+  type StructuredSchemaType,
+  type StructuredScalar,
+} from './json-schema.ts'
+
 // The render-intent vocabulary a tool declares via `presentCall`/`presentResult`
 // lives in its own UI-facing module; re-export it so `@deepseek-ai/dsh-tools`
 // stays the single public surface for consumers (producers + the ACP bridge).

packages/core/tools/src/json-schema.ts

@@ -0,0 +1,345 @@
+/**
+ * Structured-output JSON Schema subset: the vocabulary a caller uses to demand
+ * a machine-readable result from a subagent (`SubagentStartRequest.outputSchema`)
+ * or a workflow `agent()` call.
+ *
+ * This is deliberately NOT full JSON Schema. The schema travels verbatim to the
+ * model as a forced tool's `parameters`, and the value the model produces is
+ * validated here — so every accepted keyword must be one this module actually
+ * enforces. Accepting a keyword we don't enforce would validate less than the
+ * schema promises (accepted-then-ignored), so anything outside the subset is
+ * REJECTED LOUD by {@link assertSupportedOutputSchema} instead. The subset:
+ *
+ * - `type` — a single string (`object`/`array`/`string`/`number`/`integer`/
+ *   `boolean`/`null`); type ARRAYS (`["string","null"]`) are rejected.
+ * - `properties`/`required`/`additionalProperties` (boolean) on objects; every
+ *   `required` key must be declared in `properties`. `additionalProperties`
+ *   absent keeps standard JSON Schema semantics (extra keys allowed).
+ * - `items` on arrays (absent ⇒ any JSON items).
+ * - `enum` (non-empty, scalars only) and `const` (scalar) on scalar types.
+ * - Annotations `description`/`title`/`default`/`examples` are allowed and
+ *   ignored (they constrain nothing), except that they must still be JSON data
+ *   — the schema is serialized onto the wire, so a non-JSON annotation would be
+ *   silently mangled.
+ *
+ * Values checked by {@link validateStructuredValue} are expected to be plain
+ * host-realm JSON data (model tool-call arguments are parsed wire JSON; a
+ * caller holding foreign-realm data materializes it first).
+ *
+ * @module dsh-tools/json-schema
+ */
+
+import { assertNever, HarnessError } from '@deepseek-ai/dsh-llm'
+
+/** The scalar values `enum`/`const` may carry (finite numbers only). */
+export type StructuredScalar = string | number | boolean | null
+
+/** The `type` keywords the subset accepts. */
+export type StructuredSchemaType = 'object' | 'array' | 'string' | 'number' | 'integer' | 'boolean' | 'null'
+
+/**
+ * One node of the structured-output schema subset. Recursive via `properties`
+ * and `items`; see the module doc for the exact keyword semantics.
+ */
+export interface StructuredSchemaNode {
+  type: StructuredSchemaType
+  /** Nested property schemas (`type: 'object'` only). */
+  properties?: Record<string, StructuredSchemaNode>
+  /** Required property names; each must appear in `properties`. */
+  required?: string[]
+  /** `false` rejects undeclared keys; absent/`true` allows them (JSON Schema default). */
+  additionalProperties?: boolean
+  /** Item schema (`type: 'array'` only); absent ⇒ any JSON items. */
+  items?: StructuredSchemaNode
+  /** Allowed values (scalar types only). */
+  enum?: StructuredScalar[]
+  /** The single allowed value (scalar types only). */
+  const?: StructuredScalar
+  /** Annotation, ignored for validation. */
+  description?: string
+  /** Annotation, ignored for validation. */
+  title?: string
+  /** Annotation, ignored for validation (must still be JSON data). */
+  default?: unknown
+  /** Annotation, ignored for validation (must still be JSON data). */
+  examples?: unknown
+}
+
+/** A structured-output schema: an OBJECT-rooted {@link StructuredSchemaNode}. */
+export type StructuredOutputSchema = StructuredSchemaNode & { type: 'object' }
+
+/**
+ * Thrown by {@link assertSupportedOutputSchema} when a schema falls outside the
+ * supported subset. Extends {@link HarnessError} (`code: 'UNSUPPORTED_SCHEMA'`)
+ * so seam code and tool results can route on it; `violations` lists every
+ * offending path, not just the first.
+ */
+export class OutputSchemaError extends HarnessError {
+  /** The individual violation messages, in walk order. */
+  readonly violations: string[]
+
+  constructor(violations: string[]) {
+    super(`unsupported output schema: ${violations.join('; ')}`, 'UNSUPPORTED_SCHEMA')
+    this.name = 'OutputSchemaError'
+    this.violations = violations
+  }
+}
+
+/** The keywords the subset accepts, checked (`constraint`) or ignored (`annotation`). */
+const CONSTRAINT_KEYWORDS = new Set(['type', 'properties', 'required', 'additionalProperties', 'items', 'enum', 'const'])
+const ANNOTATION_KEYWORDS = new Set(['description', 'title', 'default', 'examples'])
+
+const SCHEMA_TYPES: readonly StructuredSchemaType[] = ['object', 'array', 'string', 'number', 'integer', 'boolean', 'null']
+
+/**
+ * Whether a value is a PLAIN JSON object — non-null, non-array, and with a
+ * prototype chain of at most one link (`null`-proto, or any realm's
+ * `Object.prototype`, whose own prototype is `null`). Realm-agnostic on
+ * purpose: a schema materialized in another realm carries THAT realm's
+ * `Object.prototype`, which an identity check would wrongly reject. Exotic
+ * hosts (`Date`, `Map`, class instances) have longer chains and are rejected —
+ * they would serialize lossily (`Date` → string, `Map` → `{}`) instead of
+ * failing loud.
+ */
+function isObjectLike(value: unknown): value is Record<string, unknown> {
+  if (typeof value !== 'object' || value === null || Array.isArray(value)) return false
+  const proto: unknown = Object.getPrototypeOf(value)
+  return proto === null || Object.getPrototypeOf(proto) === null
+}
+
+/** Whether a value is a supported scalar (`enum`/`const` member): string, finite number, boolean, or null. */
+function isStructuredScalar(value: unknown): value is StructuredScalar {
+  return value === null || typeof value === 'string' || typeof value === 'boolean'
+    || (typeof value === 'number' && Number.isFinite(value))
+}
+
+/**
+ * Whether a value is JSON data (annotation payloads only): scalars, arrays, and
+ * object-likes of such values. Realm-agnostic on purpose (no prototype check) —
+ * the schema may have been materialized from another realm; structural JSON-ness
+ * is what the wire needs. Cycles are rejected via `seen`.
+ */
+function isJsonData(value: unknown, seen: Set<object>): boolean {
+  if (isStructuredScalar(value)) return true
+  // The scalar check above already returned for null, so `object` here is a real object.
+  if (typeof value !== 'object') return false
+  if (seen.has(value)) return false
+  seen.add(value)
+  try {
+    if (Array.isArray(value)) return value.every(entry => isJsonData(entry, seen))
+    // A non-plain object (Date, Map, class instance) is NOT JSON data even when
+    // it has no enumerable values — it would serialize lossily, not loudly.
+    if (!isObjectLike(value)) return false
+    return Object.values(value).every(entry => isJsonData(entry, seen))
+  } finally {
+    seen.delete(value)
+  }
+}
+
+/** Collect subset violations for one schema node (recursive walk). */
+function checkSchemaNode(node: unknown, path: string, violations: string[], seen: Set<object>): void {
+  if (!isObjectLike(node)) {
+    violations.push(`${path} must be a schema object`)
+    return
+  }
+  if (seen.has(node)) {
+    violations.push(`${path} is circular`)
+    return
+  }
+  seen.add(node)
+
+  for (const key of Object.keys(node)) {
+    if (CONSTRAINT_KEYWORDS.has(key)) continue
+    if (ANNOTATION_KEYWORDS.has(key)) {
+      if (!isJsonData(node[key], new Set())) violations.push(`${path}.${key} annotation must be JSON data`)
+      continue
+    }
+    violations.push(`${path}.${key} is not a supported keyword (subset: type/properties/required/additionalProperties/items/enum/const + annotations)`)
+  }
+  if (typeof node.description !== 'undefined' && typeof node.description !== 'string') {
+    violations.push(`${path}.description must be a string`)
+  }
+  if (typeof node.title !== 'undefined' && typeof node.title !== 'string') {
+    violations.push(`${path}.title must be a string`)
+  }
+
+  const type = node.type
+  if (typeof type !== 'string' || !(SCHEMA_TYPES as readonly unknown[]).includes(type)) {
+    violations.push(Array.isArray(type)
+      ? `${path}.type must be a single type string (type arrays are not supported)`
+      : `${path}.type must be one of ${SCHEMA_TYPES.join('/')}`)
+    seen.delete(node)
+    return
+  }
+  const schemaType = type as StructuredSchemaType
+
+  // Keywords that only make sense on one type are rejected elsewhere — an
+  // `items` on an object (or `properties` on a string) is a schema-author bug
+  // the subset surfaces rather than ignores.
+  const allowedFor: Record<string, StructuredSchemaType[]> = {
+    properties: ['object'],
+    required: ['object'],
+    additionalProperties: ['object'],
+    items: ['array'],
+    enum: ['string', 'number', 'integer', 'boolean', 'null'],
+    const: ['string', 'number', 'integer', 'boolean', 'null'],
+  }
+  for (const [key, types] of Object.entries(allowedFor)) {
+    if (key in node && !types.includes(schemaType)) {
+      violations.push(`${path}.${key} is not supported on type "${schemaType}"`)
+    }
+  }
+
+  switch (schemaType) {
+    case 'object': {
+      const properties = node.properties
+      if (properties !== undefined) {
+        if (!isObjectLike(properties)) {
+          violations.push(`${path}.properties must be an object of schemas`)
+        } else {
+          for (const [key, child] of Object.entries(properties)) {
+            checkSchemaNode(child, `${path}.properties.${key}`, violations, seen)
+          }
+        }
+      }
+      const required = node.required
+      if (required !== undefined) {
+        if (!Array.isArray(required) || required.some(entry => typeof entry !== 'string')) {
+          violations.push(`${path}.required must be an array of strings`)
+        } else {
+          const declared = isObjectLike(properties) ? properties : {}
+          // The guard above proved every entry is a string.
+          for (const key of required as string[]) {
+            // Own-property check: `in` would let inherited names (`toString`)
+            // satisfy the declared-in-properties contract via the prototype.
+            if (!Object.hasOwn(declared, key)) violations.push(`${path}.required names "${key}" which is not in properties`)
+          }
+        }
+      }
+      if (node.additionalProperties !== undefined && typeof node.additionalProperties !== 'boolean') {
+        violations.push(`${path}.additionalProperties must be a boolean`)
+      }
+      break
+    }
+    case 'array': {
+      if (node.items !== undefined) checkSchemaNode(node.items, `${path}.items`, violations, seen)
+      break
+    }
+    case 'string':
+    case 'number':
+    case 'integer':
+    case 'boolean':
+    case 'null': {
+      const allowed = node.enum
+      if (allowed !== undefined) {
+        if (!Array.isArray(allowed) || allowed.length === 0 || !allowed.every(entry => isStructuredScalar(entry))) {
+          violations.push(`${path}.enum must be a non-empty array of scalars`)
+        }
+      }
+      if ('const' in node && !isStructuredScalar(node.const)) {
+        violations.push(`${path}.const must be a scalar`)
+      }
+      break
+    }
+    /* v8 ignore start -- defensive: schemaType was membership-checked against SCHEMA_TYPES above, so no runtime value reaches here */
+    default:
+      assertNever(schemaType, 'assertSupportedOutputSchema')
+    /* v8 ignore stop */
+  }
+
+  seen.delete(node)
+}
+
+/**
+ * Assert `schema` is a supported {@link StructuredOutputSchema} — object-rooted
+ * and entirely within the enforced subset. Throws {@link OutputSchemaError}
+ * (`UNSUPPORTED_SCHEMA`) listing EVERY violation; returns (and narrows) on
+ * success. Call this at the seam boundary, before any child is created.
+ * @param schema - the caller-supplied schema (unknown until asserted).
+ * @returns nothing — the assertion signature narrows `schema` to
+ * {@link StructuredOutputSchema} in the caller's scope on normal return.
+ */
+export function assertSupportedOutputSchema(schema: unknown): asserts schema is StructuredOutputSchema {
+  const violations: string[] = []
+  checkSchemaNode(schema, 'schema', violations, new Set())
+  if (violations.length === 0 && (schema as StructuredSchemaNode).type !== 'object') {
+    violations.push('schema.type must be "object" (structured output is object-rooted)')
+  }
+  if (violations.length > 0) throw new OutputSchemaError(violations)
+}
+
+/** Collect violations for one value against an (already asserted) schema node. */
+function checkValue(node: StructuredSchemaNode, value: unknown, path: string): string[] {
+  switch (node.type) {
+    case 'object': {
+      if (!isObjectLike(value)) return [`"${path}" must be an object`]
+      const violations: string[] = []
+      const properties = node.properties ?? {}
+      // Own-property discipline throughout: JSON carries own enumerable
+      // properties only, so an inherited `toString` must not satisfy
+      // `required`, dodge `additionalProperties: false`, or be validated as if
+      // the value carried it.
+      for (const key of node.required ?? []) {
+        if (!Object.hasOwn(value, key) || value[key] === undefined) violations.push(`missing required property "${path}.${key}"`)
+      }
+      for (const [key, child] of Object.entries(properties)) {
+        if (!Object.hasOwn(value, key) || value[key] === undefined) continue
+        violations.push(...checkValue(child, value[key], `${path}.${key}`))
+      }
+      if (node.additionalProperties === false) {
+        for (const key of Object.keys(value)) {
+          if (!Object.hasOwn(properties, key)) violations.push(`"${path}.${key}" is not a declared property (additionalProperties: false)`)
+        }
+      }
+      return violations
+    }
+    case 'array': {
+      if (!Array.isArray(value)) return [`"${path}" must be an array`]
+      if (!node.items) return []
+      const items = node.items
+      return value.flatMap((entry, index) => checkValue(items, entry, `${path}[${index}]`))
+    }
+    case 'string': {
+      if (typeof value !== 'string') return [`"${path}" must be a string`]
+      break
+    }
+    case 'number': {
+      if (typeof value !== 'number' || !Number.isFinite(value)) return [`"${path}" must be a finite number`]
+      break
+    }
+    case 'integer': {
+      if (typeof value !== 'number' || !Number.isInteger(value)) return [`"${path}" must be an integer`]
+      break
+    }
+    case 'boolean': {
+      if (typeof value !== 'boolean') return [`"${path}" must be a boolean`]
+      break
+    }
+    case 'null': {
+      if (value !== null) return [`"${path}" must be null`]
+      break
+    }
+    default:
+      return assertNever(node.type, 'validateStructuredValue')
+  }
+  // Scalar constraint checks, shared by every scalar branch above.
+  if (node.enum && !node.enum.includes(value)) {
+    return [`"${path}" must be one of ${JSON.stringify(node.enum)}`]
+  }
+  if ('const' in node && value !== node.const) {
+    return [`"${path}" must be ${JSON.stringify(node.const)}`]
+  }
+  return []
+}
+
+/**
+ * Validate a value against an (already {@link assertSupportedOutputSchema}-
+ * asserted) schema. Returns human-readable, path-qualified violation messages
+ * — empty means valid. Total: never throws, however malformed the value.
+ * @param schema - the asserted schema to check against.
+ * @param value - the candidate value (e.g. parsed tool-call arguments).
+ * @returns every violation found, in walk order (empty = valid).
+ */
+export function validateStructuredValue(schema: StructuredOutputSchema, value: unknown): string[] {
+  return checkValue(schema, value, 'value')
+}

packages/core/tools/tests/json-schema.spec.ts

@@ -0,0 +1,304 @@
+import { describe, expect, it } from 'vitest'
+import {
+  assertSupportedOutputSchema,
+  OutputSchemaError,
+  validateStructuredValue,
+  type StructuredOutputSchema,
+} from '../src/json-schema.ts'
+
+/** Assert-and-narrow helper: the asserted schema, typed. */
+function asserted(schema: unknown): StructuredOutputSchema {
+  assertSupportedOutputSchema(schema)
+  return schema
+}
+
+/** The violations OutputSchemaError carries for a bad schema (throws if it passes). */
+function violationsOf(schema: unknown): string[] {
+  try {
+    assertSupportedOutputSchema(schema)
+  } catch (error: unknown) {
+    if (error instanceof OutputSchemaError) return error.violations
+    throw error
+  }
+  throw new Error('expected the schema to be rejected')
+}
+
+describe('assertSupportedOutputSchema', () => {
+  it('accepts a representative subset schema (all supported keywords)', () => {
+    const schema = asserted({
+      type: 'object',
+      description: 'a finding',
+      title: 'Finding',
+      properties: {
+        file: { type: 'string', description: 'path' },
+        line: { type: 'integer' },
+        severity: { type: 'string', enum: ['low', 'high'] },
+        kind: { type: 'string', const: 'bug' },
+        score: { type: 'number' },
+        confirmed: { type: 'boolean' },
+        parent: { type: 'null' },
+        tags: { type: 'array', items: { type: 'string' } },
+        nested: {
+          type: 'object',
+          properties: { x: { type: 'number', default: 3, examples: [1, 2] } },
+          additionalProperties: false,
+        },
+        anything: { type: 'array' },
+      },
+      required: ['file', 'line'],
+      additionalProperties: true,
+    })
+    expect(schema.type).toBe('object')
+  })
+
+  it('rejects a non-object root (scalar/array-rooted schemas)', () => {
+    expect(violationsOf({ type: 'string' })).toEqual(['schema.type must be "object" (structured output is object-rooted)'])
+    expect(violationsOf({ type: 'array', items: { type: 'string' } }))
+      .toContain('schema.type must be "object" (structured output is object-rooted)')
+  })
+
+  it('rejects non-object schema nodes and missing/unknown type', () => {
+    expect(violationsOf('nope')).toEqual(['schema must be a schema object'])
+    expect(violationsOf(null)).toEqual(['schema must be a schema object'])
+    expect(violationsOf([])).toEqual(['schema must be a schema object'])
+    expect(violationsOf({})).toEqual(['schema.type must be one of object/array/string/number/integer/boolean/null'])
+    expect(violationsOf({ type: 'tuple' })[0]).toMatch(/type must be one of/)
+    expect(violationsOf({ type: 'object', properties: { a: 'str' } })).toEqual(['schema.properties.a must be a schema object'])
+  })
+
+  it('rejects type ARRAYS with a dedicated message', () => {
+    expect(violationsOf({ type: ['string', 'null'] }))
+      .toEqual(['schema.type must be a single type string (type arrays are not supported)'])
+  })
+
+  it('rejects unsupported constraint keywords loudly (never accepted-then-ignored)', () => {
+    for (const keyword of ['oneOf', 'anyOf', 'allOf', 'not', 'pattern', 'minimum', 'maxLength', '$ref']) {
+      const bad = violationsOf({ type: 'object', [keyword]: [] })
+      expect(bad.some(v => v.includes(`schema.${keyword} is not a supported keyword`))).toBe(true)
+    }
+  })
+
+  it('reports EVERY violation, not just the first', () => {
+    const bad = violationsOf({
+      type: 'object',
+      pattern: 'x',
+      properties: { a: { type: 'weird' }, b: { type: 'string', minimum: 1 } },
+    })
+    expect(bad.length).toBe(3)
+  })
+
+  it('rejects keywords on the wrong type (items on object, properties on string, enum on object)', () => {
+    expect(violationsOf({ type: 'object', items: { type: 'string' } }))
+      .toEqual(['schema.items is not supported on type "object"'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'string', properties: {} } } }))
+      .toEqual(['schema.properties.a.properties is not supported on type "string"'])
+    expect(violationsOf({ type: 'object', enum: [1] }))
+      .toEqual(['schema.enum is not supported on type "object"'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'array', const: 1 } } }))
+      .toEqual(['schema.properties.a.const is not supported on type "array"'])
+  })
+
+  it('validates required: must be string[] naming declared properties', () => {
+    expect(violationsOf({ type: 'object', required: 'file' }))
+      .toEqual(['schema.required must be an array of strings'])
+    expect(violationsOf({ type: 'object', required: [1] }))
+      .toEqual(['schema.required must be an array of strings'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'string' } }, required: ['b'] }))
+      .toEqual(['schema.required names "b" which is not in properties'])
+    expect(violationsOf({ type: 'object', required: ['a'] }))
+      .toEqual(['schema.required names "a" which is not in properties'])
+  })
+
+  it('validates additionalProperties must be boolean and enum/const must be scalars', () => {
+    expect(violationsOf({ type: 'object', additionalProperties: {} }))
+      .toEqual(['schema.additionalProperties must be a boolean'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'string', enum: [] } } }))
+      .toEqual(['schema.properties.a.enum must be a non-empty array of scalars'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'string', enum: [{}] } } }))
+      .toEqual(['schema.properties.a.enum must be a non-empty array of scalars'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'string', enum: 'x' } } }))
+      .toEqual(['schema.properties.a.enum must be a non-empty array of scalars'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'number', enum: [Number.NaN] } } }))
+      .toEqual(['schema.properties.a.enum must be a non-empty array of scalars'])
+    expect(violationsOf({ type: 'object', properties: { a: { type: 'string', const: {} } } }))
+      .toEqual(['schema.properties.a.const must be a scalar'])
+  })
+
+  it('rejects non-string description/title and non-JSON annotation payloads', () => {
+    expect(violationsOf({ type: 'object', description: 7 }))
+      .toEqual(['schema.description must be a string'])
+    expect(violationsOf({ type: 'object', title: 7 }))
+      .toEqual(['schema.title must be a string'])
+    expect(violationsOf({ type: 'object', default: () => 1 }))
+      .toEqual(['schema.default annotation must be JSON data'])
+    expect(violationsOf({ type: 'object', examples: [undefined] }))
+      .toEqual(['schema.examples annotation must be JSON data'])
+    expect(violationsOf({ type: 'object', examples: [Number.POSITIVE_INFINITY] }))
+      .toEqual(['schema.examples annotation must be JSON data'])
+    // A cyclic annotation payload is caught by the JSON-data walk.
+    const cyclicAnnotation: Record<string, unknown> = {}
+    cyclicAnnotation.self = cyclicAnnotation
+    expect(violationsOf({ type: 'object', default: cyclicAnnotation }))
+      .toEqual(['schema.default annotation must be JSON data'])
+    // Object/array annotations that ARE JSON data pass.
+    asserted({ type: 'object', default: { a: [1, 'x', null, true] } })
+  })
+
+  it('rejects a circular schema instead of recursing forever', () => {
+    const node: Record<string, unknown> = { type: 'object' }
+    node.properties = { self: node }
+    expect(violationsOf(node)).toEqual(['schema.properties.self is circular'])
+  })
+
+  it('accepts the same subschema object reused in two SIBLING positions (a DAG, not a cycle)', () => {
+    const leaf = { type: 'string' }
+    asserted({ type: 'object', properties: { a: leaf, b: leaf } })
+  })
+
+  it('required cannot be satisfied by INHERITED names — `toString` is not a declared property', () => {
+    // `'toString' in {}` is true via Object.prototype; the declared-property
+    // contract must be an own-property check.
+    expect(violationsOf({ type: 'object', properties: {}, required: ['toString'] }))
+      .toEqual(['schema.required names "toString" which is not in properties'])
+  })
+
+  it('rejects exotic host objects where the subset expects plain JSON structure', () => {
+    // A Map as `properties` has no own enumerable entries: structurally it
+    // would read as "no properties" and serialize to {} — lossy, not loud.
+    expect(violationsOf({ type: 'object', properties: new Map() }))
+      .toEqual(['schema.properties must be an object of schemas'])
+    // A Date node is not a schema object even though Object.values(date) is [].
+    expect(violationsOf({ type: 'object', properties: { at: new Date(0) } }))
+      .toEqual(['schema.properties.at must be a schema object'])
+  })
+
+  it('rejects exotic annotation payloads that would serialize lossily', () => {
+    expect(violationsOf({ type: 'object', default: new Date(0) }))
+      .toEqual(['schema.default annotation must be JSON data'])
+    expect(violationsOf({ type: 'object', examples: [new Map()] }))
+      .toEqual(['schema.examples annotation must be JSON data'])
+  })
+})
+
+describe('validateStructuredValue', () => {
+  const schema = asserted({
+    type: 'object',
+    properties: {
+      file: { type: 'string' },
+      line: { type: 'integer' },
+      score: { type: 'number' },
+      confirmed: { type: 'boolean' },
+      parent: { type: 'null' },
+      severity: { type: 'string', enum: ['low', 'high'] },
+      kind: { type: 'string', const: 'bug' },
+      tags: { type: 'array', items: { type: 'string' } },
+      free: { type: 'array' },
+      nested: { type: 'object', properties: { x: { type: 'number' } }, required: ['x'], additionalProperties: false },
+    },
+    required: ['file'],
+  })
+
+  it('accepts a fully valid value (empty violations)', () => {
+    expect(validateStructuredValue(schema, {
+      file: 'a.ts', line: 3, score: 0.5, confirmed: true, parent: null,
+      severity: 'high', kind: 'bug', tags: ['x'], free: [1, { any: true }], nested: { x: 1 },
+    })).toEqual([])
+  })
+
+  it('reports missing required and wrong root type', () => {
+    expect(validateStructuredValue(schema, {})).toEqual(['missing required property "value.file"'])
+    expect(validateStructuredValue(schema, 'nope')).toEqual(['"value" must be an object'])
+    expect(validateStructuredValue(schema, [])).toEqual(['"value" must be an object'])
+  })
+
+  it('type-checks every scalar branch with path-qualified messages', () => {
+    expect(validateStructuredValue(schema, { file: 1 })).toEqual(['"value.file" must be a string'])
+    expect(validateStructuredValue(schema, { file: 'a', line: 1.5 })).toEqual(['"value.line" must be an integer'])
+    expect(validateStructuredValue(schema, { file: 'a', line: 'x' })).toEqual(['"value.line" must be an integer'])
+    expect(validateStructuredValue(schema, { file: 'a', score: 'x' })).toEqual(['"value.score" must be a finite number'])
+    expect(validateStructuredValue(schema, { file: 'a', score: Number.NaN })).toEqual(['"value.score" must be a finite number'])
+    expect(validateStructuredValue(schema, { file: 'a', confirmed: 'yes' })).toEqual(['"value.confirmed" must be a boolean'])
+    expect(validateStructuredValue(schema, { file: 'a', parent: 0 })).toEqual(['"value.parent" must be null'])
+  })
+
+  it('enforces enum membership and const equality', () => {
+    expect(validateStructuredValue(schema, { file: 'a', severity: 'mid' }))
+      .toEqual(['"value.severity" must be one of ["low","high"]'])
+    expect(validateStructuredValue(schema, { file: 'a', kind: 'feature' }))
+      .toEqual(['"value.kind" must be "bug"'])
+  })
+
+  it('checks arrays per index; an items-less array accepts anything', () => {
+    expect(validateStructuredValue(schema, { file: 'a', tags: 'x' })).toEqual(['"value.tags" must be an array'])
+    expect(validateStructuredValue(schema, { file: 'a', tags: ['ok', 2] })).toEqual(['"value.tags[1]" must be a string'])
+    expect(validateStructuredValue(schema, { file: 'a', free: [{ deep: [1] }, null] })).toEqual([])
+  })
+
+  it('recurses into nested objects: required + additionalProperties: false', () => {
+    expect(validateStructuredValue(schema, { file: 'a', nested: {} }))
+      .toEqual(['missing required property "value.nested.x"'])
+    expect(validateStructuredValue(schema, { file: 'a', nested: { x: 1, y: 2 } }))
+      .toEqual(['"value.nested.y" is not a declared property (additionalProperties: false)'])
+    expect(validateStructuredValue(schema, { file: 'a', nested: 3 }))
+      .toEqual(['"value.nested" must be an object'])
+  })
+
+  it('a required key present-but-undefined counts as missing', () => {
+    expect(validateStructuredValue(schema, { file: undefined })).toEqual(['missing required property "value.file"'])
+  })
+
+  it('inherited properties satisfy nothing: required, additionalProperties, and recursion are own-property only', () => {
+    // required: ['toString'] must NOT be satisfied by Object.prototype.toString.
+    expect(validateStructuredValue(
+      asserted({ type: 'object', properties: { toString: { type: 'string' } }, required: ['toString'] }),
+      {},
+    )).toEqual(['missing required property "value.toString"'])
+    // additionalProperties: false must flag an OWN `toString` key even though
+    // `'toString' in properties` is true via the prototype.
+    expect(validateStructuredValue(
+      asserted({ type: 'object', additionalProperties: false }),
+      { toString: 1 },
+    )).toEqual(['"value.toString" is not a declared property (additionalProperties: false)'])
+    // A declared property the value does NOT carry must not be validated
+    // against the value's INHERITED member (constructor is a function on
+    // every plain object's prototype, not a carried property).
+    expect(validateStructuredValue(
+      asserted({ type: 'object', properties: { constructor: { type: 'string' } } }),
+      {},
+    )).toEqual([])
+  })
+
+  it('a non-plain object value is not an object in the JSON sense', () => {
+    expect(validateStructuredValue(asserted({ type: 'object' }), new Date(0)))
+      .toEqual(['"value" must be an object'])
+  })
+
+  it('collects multiple violations across branches in one pass', () => {
+    expect(validateStructuredValue(schema, { line: 'x', severity: 'mid' })).toEqual([
+      'missing required property "value.file"',
+      '"value.line" must be an integer',
+      '"value.severity" must be one of ["low","high"]',
+    ])
+  })
+
+  it('null-typed const/enum work through the scalar path', () => {
+    const nullish = asserted({ type: 'object', properties: { a: { type: 'null', const: null } } })
+    expect(validateStructuredValue(nullish, { a: null })).toEqual([])
+  })
+
+  it('rejects a non-object properties value in the schema walk', () => {
+    expect(violationsOf({ type: 'object', properties: [] }))
+      .toEqual(['schema.properties must be an object of schemas'])
+  })
+
+  it('an object schema without properties/required only type-checks its value', () => {
+    const bare = asserted({ type: 'object' })
+    expect(validateStructuredValue(bare, { any: ['thing'] })).toEqual([])
+    expect(validateStructuredValue(bare, 7)).toEqual(['"value" must be an object'])
+  })
+
+  it('validateStructuredValue throws on a type the assert would never let through (assertNever backstop)', () => {
+    const forged = { type: 'tuple' } as unknown as StructuredOutputSchema
+    expect(() => validateStructuredValue(forged, 1)).toThrow(/tuple/)
+  })
+})

packages/session-persistence/session-persistence-sqlite/README.md

@@ -8,7 +8,7 @@ A SQLite durable session-persistence backend — a second `SessionPersistence` i
 
 Each `SessionEvent` maps 1:1 onto a row in an `events` table `(session_id, seq, type, time, data, source_event_seqs, surface_op)` — `data` is the event payload as JSON text, so the row shape is the event verbatim (including `assistant/chunk`, keeping `seq` contiguous). The two `TEXT` columns `source_event_seqs` and `surface_op` are nullable; they store the event's optional surface-metadata fields (see [session surface](../../../docs/rfc/implemented/architecture/2026-06-18-session-surface.md)). Out-of-log metadata (`SessionHeader`) lives in a `sessions` row. A `sessions` row is written only by the first `append` — its existence is the lazy-materialization signal (`list` reports exactly the sessions that have a row), so no separate column is needed.
 
-The repo targets Node ≥ 24 (the root `engines` field), which includes the stable `node:sqlite` module. The database opens with `foreign_keys = ON` (so `ON DELETE CASCADE` drops a session's events with its row) and the configured `journal_mode` (default `wal`; pick a rollback-journal mode like `delete` on filesystems where WAL's shared-memory files do not work, e.g. network mounts). The table-layout version is stored in `PRAGMA user_version` and checked on open: a fresh database is stamped with the current `SCHEMA_VERSION`; a database written by any other, incompatible build (a non-current `user_version`, older or newer) is rejected rather than opened against an unknown layout — there is no migration (unreleased software).
+The repo's `engines.node` is `^22.19.0 || >=24.0.0` (Node 22.19+ or 24+), matching the LTS floor required by the installed Pi adapter dependency; `node:sqlite` itself ships without the `--experimental-sqlite` flag from Node 22.13 (LTS) and 23.4 / 24 (Current) on. The range deliberately excludes Node 23 because that line is non-LTS/EOL and still has flagged runtime features before 23.6. The database opens with `foreign_keys = ON` (so `ON DELETE CASCADE` drops a session's events with its row) and the configured `journal_mode` (default `wal`; pick a rollback-journal mode like `delete` on filesystems where WAL's shared-memory files do not work, e.g. network mounts). The table-layout version is stored in `PRAGMA user_version` and checked on open: a fresh database is stamped with the current `SCHEMA_VERSION`; a database written by any other, incompatible build (a non-current `user_version`, older or newer) is rejected rather than opened against an unknown layout — there is no migration (unreleased software).
 
 ## Contract semantics over rows
 

packages/subagent/subagent-fork/README.md

@@ -12,7 +12,7 @@ The seam this rides on: `CreateAgentOptions.seed` (added on `dsh-agent`, threade
 
 ## Capabilities
 
-`{ outputSchema: false, depthLimit: true, toolFilter: false }` — identical to spawn (the depth/model/output behavior is the shared driver's).
+`{ outputSchema: true, depthLimit: true, toolFilter: false }` — identical to spawn (the depth/model/structured-output behavior is the shared driver's).
 
 ## Config
 

packages/subagent/subagent-fork/src/index.ts

@@ -28,6 +28,10 @@ import type { SubagentCapabilities, SubagentProvider, SubagentStartRequest } fro
 import { startInProcessRun } from '@deepseek-ai/dsh-subagent-inprocess'
 
 export const name = 'subagent-fork'
+// `tools` is deliberately NOT injected — same rationale as subagent-spawn: the
+// per-run structured runtime gates its capture-tool registration on `tools`
+// itself, so this backend's apply timing (and the delegation tool's position
+// in the model-visible tool list) is unchanged by structured output.
 export const inject = ['subagents', 'agents']
 
 /** Config: the registry name to register the provider under. */
@@ -59,11 +63,12 @@ export function completedTurnPrefix(parent: Agent): SessionEvent[] {
 }
 
 /**
- * The fork provider. Supports `depthLimit`; NOT `outputSchema`/`toolFilter` this
- * cut (the service rejects a request needing either before `start` runs).
+ * The fork provider. Supports `depthLimit` and `outputSchema` (via the shared
+ * in-process structured runtime); NOT `toolFilter` this cut (the service
+ * rejects a request needing it before `start` runs).
  */
 class ForkProvider implements SubagentProvider {
-  readonly capabilities: SubagentCapabilities = { outputSchema: false, depthLimit: true, toolFilter: false }
+  readonly capabilities: SubagentCapabilities = { outputSchema: true, depthLimit: true, toolFilter: false }
   // Context contract: a forked child IS seeded with the parent's completed-turn prefix.
   readonly inheritsParentContext = true
 

packages/subagent/subagent-fork/tests/subagent-fork.spec.ts

@@ -9,9 +9,10 @@ import AgentRegistry, { AgentId } from '@deepseek-ai/dsh-agent'
 import AgentLoop from '@deepseek-ai/dsh-agent-loop'
 import * as Invariants from '@deepseek-ai/dsh-invariants'
 import SubagentService from '@deepseek-ai/dsh-subagent'
-import { MockAdapter, textResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
+import { MockAdapter, textResponse, toolCallResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
 import type { StreamChunk } from '@deepseek-ai/dsh-llm'
 import * as fork from '../src/index.ts'
+import { STRUCTURED_OUTPUT_TOOL } from '@deepseek-ai/dsh-subagent-inprocess'
 import { completedTurnPrefix } from '../src/index.ts'
 
 type Script = ConstructorParameters<typeof MockAdapter>[0]
@@ -141,6 +142,26 @@ describe('dsh-subagent-fork', () => {
     await run.dispose()
   })
 
+  it('captures structured output through the shipped plugin (seeded child, driver runtime)', async () => {
+    const { ctx, parent } = await setup([
+      textResponse('parent turn'),
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 9 }),
+    ])
+    parent.send([{ type: 'text', text: 'warm up' }])
+    await parent.whenIdle()
+    const run = ctx.subagents.start('fork', {
+      prompt: [{ type: 'text', text: 'report structured' }],
+      parent,
+      outputSchema: { type: 'object', properties: { answer: { type: 'number' } }, required: ['answer'] },
+    })
+    const result = await run.result
+    expect(result.stopReason).toBe('completed')
+    expect(result.structured).toEqual({ answer: 9 })
+    // Run-scoped runtime: nothing stays registered after the settle.
+    expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    await run.dispose()
+  })
+
   it('does NOT return the seeded parent output when the child produces no message of its own', async () => {
     // Regression: readResult must scope to the child's OWN events (after the
     // seed). The parent completes a turn with a distinctive assistant message,
@@ -161,9 +182,9 @@ describe('dsh-subagent-fork', () => {
     await run.dispose()
   })
 
-  it('advertises depthLimit but not outputSchema/toolFilter', async () => {
+  it('advertises depthLimit and outputSchema but not toolFilter', async () => {
     const { ctx } = await setup([])
-    expect(ctx.subagents.getProvider('fork')!.capabilities).toEqual({ outputSchema: false, depthLimit: true, toolFilter: false })
+    expect(ctx.subagents.getProvider('fork')!.capabilities).toEqual({ outputSchema: true, depthLimit: true, toolFilter: false })
   })
 
   it('unregisters the provider when its fiber is disposed (HMR safety)', async () => {

packages/subagent/subagent-inprocess/README.md

@@ -8,10 +8,10 @@ The shared **in-process subagent run driver**. A pure library (no provider, no r
 
 Runs a child as a child [`Agent`](../../core/agent) on the same cordis context (`ctx.agents`):
 
-1. computes child depth = `depthOf(parent) + 1`; if `request.maxDepth` is set and exceeded, throws `SubagentDepthError` (the `depthLimit` capability);
-2. creates a child via `ctx.agents.create` with a fresh `AgentId`/`SessionId`, the parent's `cwd` + `parentSession` lineage, the optional `options.seed` (fork's completed-turn prefix; omitted for a fresh child), and `agentOptions` (the child inherits the **parent's model** by default — a child with no model can't run — overridable via `request.agentOptions.model`; the system prompt is NOT inherited);
-3. drives the one-shot: `child.send(prompt)` then `await child.whenIdle()` (ordering matters — `send` enqueues synchronously, so `whenIdle` observes the queued work and resolves on the child's `running → idle` transition, never before the turn starts);
-4. reads the result, scoped to the child's OWN events (everything at or after `seedLength`, so a seeded child that produced no message of its own never returns the seeded parent's last message): the last `assistant/message` content (deep-cloned — the log is frozen) and the last `turn/end.reason` mapped to a `SubagentStopReason`.
+1. computes child depth = `depthOf(parent) + 1`; if `request.maxDepth` is set and exceeded, throws `SubagentDepthError` (the `depthLimit` capability); a `request.outputSchema` is asserted against the supported subset (`assertSupportedOutputSchema` from [dsh-tools](../../core/tools/README.md)) and then snapshotted with `structuredClone` before any child exists — assertion first so a hostile value fails as `OutputSchemaError` (never a raw clone error), the snapshot so a post-`start()` caller mutation cannot drift the enforced schema;
+2. creates a child via `ctx.agents.create` with a fresh `AgentId`/`SessionId`, the parent's `cwd` + `parentSession` lineage, the optional `options.seed` (fork's completed-turn prefix; omitted for a fresh child), and `agentOptions` (the child inherits the **parent's model** by default — a child with no model can't run — overridable via `request.agentOptions.model`; the deployment persona needs no inheritance — it is a context-wide prompt section);
+3. drives the one-shot: `child.send(prompt)` then `await child.whenIdle()` (ordering matters — `send` enqueues synchronously, so `whenIdle` observes the queued work and resolves on the child's `running → idle` transition, never before the turn starts); there is deliberately NO re-prompt for a structured child that finished cleanly without calling `structured_output` — the shortfall maps to an `error` result for the parent;
+4. reads the result, scoped to the child's OWN events (everything at or after `seedLength`, so a seeded child that produced no message of its own never returns the seeded parent's last message): the last `assistant/message` content (deep-cloned — the log is frozen) and the last `turn/end.reason` mapped to a `SubagentStopReason`. A structured run surfaces the captured value as `result.structured`; a structured child that finished cleanly WITHOUT ever capturing settles `error` (a clean finish without the demanded result is a failure, not a success with a missing field).
 
 `dispose()` delegates to `AgentHandle.dispose()` (stop loop → await quiescence → remove session); `cancel()` cancels the child's in-flight turn. A cancel landing before any `turn/end` (the pre-turn window) still settles `aborted`, honoring the cancel contract rather than the generic no-turn `error`.
 
@@ -19,6 +19,19 @@ Runs a child as a child [`Agent`](../../core/agent) on the same cordis context (
 
 `{ providerName: string; seed?: SessionEvent[] }` — the per-backend inputs: the provider name (for error context) and the optional child-session seed.
 
+### Structured output (package-internal runtime)
+
+The mechanism behind `outputSchema` for in-process children — acquired per structured RUN inside `startInProcessRun` (nothing is registered on a context that never runs a structured child; only the model-facing constants `STRUCTURED_OUTPUT_TOOL`/`STRUCTURED_OUTPUT_INSTRUCTION` are exported). One globally registered `structured_output` capture tool (its registered parameters are a placeholder) plus four listeners:
+
+- a `system-prompt/assemble` waterfall listener registered `prepend: true` that post-processes `await next()` — **final-assembly enforcement**: the assembly the loop renders never carries `structured_output` for an agent without a structured run, and for one that has it always carries the run's OWN schema (as the tool's `parameters`) plus the calling instruction as a trailing prompt section (the demand travels with the tool — `AgentOptions` has no per-agent prompt field to carry it). The loop logs the rendered assembly as the step's `request/header`, so the injection is reconstructable log state, never a wire-only mutation. Per-agent shaping lives here because the tool registry and prompt assembly are context-global while schemas differ per concurrent child (FIXME in the module doc: per-agent/per-session scoping would dissolve this); cooperative mutate-then-`next()` would not survive a downstream listener returning a replacement assembly.
+- a `tools/post-execute` listener (`prepend: true` = outermost, so `await next()` yields the composed final decision) that COMMITS the capture: the tool body only stages the validated value, and it becomes the run's result only when the final decision accepts the call — a downstream block (a PostToolUse hook) turns the logged result into `isError`, and the run must not report `structured` success for a call the model and session log saw fail.
+- a `tools/pre-execute` deny for any call arriving after the agent's capture — terminal means terminal WITHIN the step: a response listing `structured_output` before further tool calls cannot run side effects after the final answer was accepted.
+- an `agent/turn-continuation` listener (also `prepend: true` — an earlier-registered force-continue listener returning without `next()` must not decide the turn before the veto runs) that stops a child's turn once its output is captured, so a successful capture doesn't buy a wasted extra model step.
+
+The capture tool validates each call against the run's schema (`validateStructuredValue`) — violations become an `INVALID_ARGS` isError result the model retries in-turn; a valid call stages the value for the post-execute commit.
+
+Lifetime is refcounted by live structured runs: each acquires at start and releases at settle, so the registrations exist exactly while at least one structured child is live, a backend hot-reload mid-run cannot unregister the capture tool under a live child, and the last settle disposes everything. `release()` is idempotent per acquisition.
+
 ### `depthOf(agent): number`
 
 Delegation depth rides on a merge-extensible `AgentOptions.subagentDepth` field (0 for a top-level agent, parent + 1 for a child), so a nested spawn reads its parent's depth from `parent.options.subagentDepth`. `depthOf` reads it (absent ⇒ 0).

packages/subagent/subagent-inprocess/package.json

@@ -26,6 +26,8 @@
     "@deepseek-ai/dsh-llm": "^0.0.1",
     "@deepseek-ai/dsh-session": "^0.0.1",
     "@deepseek-ai/dsh-subagent": "^0.0.1",
+    "@deepseek-ai/dsh-system-prompt": "^0.0.1",
+    "@deepseek-ai/dsh-tools": "^0.0.1",
     "cordis": "^4.0.0-rc.6"
   },
   "devDependencies": {

packages/subagent/subagent-inprocess/src/index.ts

@@ -18,7 +18,20 @@ import type { Context } from 'cordis'
 import { AgentId, type Agent, type AgentHandle, type AgentOptions } from '@deepseek-ai/dsh-agent'
 import { SessionId, type SessionEvent, type TurnEndReason } from '@deepseek-ai/dsh-session'
 import type { ContentBlock } from '@deepseek-ai/dsh-llm'
+import { assertSupportedOutputSchema } from '@deepseek-ai/dsh-tools'
 import type { SubagentResult, SubagentRun, SubagentStartRequest, SubagentStopReason } from '@deepseek-ai/dsh-subagent'
+import {
+  acquireStructuredRuntime,
+  type StructuredAcquisition,
+} from './structured.ts'
+
+// The runtime itself (acquire/attach/release) is package-internal: runs
+// acquire it inside startInProcessRun, and no other package drives it. Only
+// the model-facing vocabulary is public.
+export {
+  STRUCTURED_OUTPUT_TOOL,
+  STRUCTURED_OUTPUT_INSTRUCTION,
+} from './structured.ts'
 
 declare module '@deepseek-ai/dsh-agent' {
   interface AgentOptions {
@@ -109,6 +122,18 @@ export function startInProcessRun(
   if (request.maxDepth !== undefined && childDepth > request.maxDepth) {
     throw new SubagentDepthError(childDepth, request.maxDepth)
   }
+  // Assert, then snapshot, the schema subset BEFORE any child exists (the
+  // service has already capability-gated; this rejects a schema outside the
+  // enforced subset loud). Assertion comes FIRST so a hostile value fails as
+  // OutputSchemaError, never as structuredClone's raw DataCloneError — the
+  // asserted subset is plain JSON data, which always clones. The snapshot is
+  // load-bearing: the caller keeps its reference, so attaching the ORIGINAL
+  // would let a post-start() mutation drift the enforced schema away from the
+  // asserted one — the clone (taken synchronously with the assertion, no
+  // interleaving possible) pins assertion, the model-visible parameters, and
+  // validateStructuredValue to one isolation-immutable value.
+  if (request.outputSchema !== undefined) assertSupportedOutputSchema(request.outputSchema)
+  const schema = request.outputSchema === undefined ? undefined : structuredClone(request.outputSchema)
 
   const childId = AgentId(randomUUID())
   // The child's OWN events begin after the seed (fork seeds the parent's
@@ -120,13 +145,20 @@ export function startInProcessRun(
   // Inherit the parent's model by default (a child with no model cannot run);
   // an explicit `request.agentOptions.model` overrides it. The persona needs
   // no inheritance: the deployment persona is a context-wide prompt section,
-  // so parent and child render the same one.
+  // so parent and child render the same one. A structured run's
+  // structured_output instruction is NOT prompt state either — the structured
+  // runtime's final-request listener appends it per request (see structured.ts).
   const agentOptions: AgentOptions = {
     ...request.parent.options.model !== undefined ? { model: request.parent.options.model } : {},
     ...request.agentOptions,
     subagentDepth: childDepth,
   }
 
+  // The structured runtime is held for the WHOLE run (acquired before the child
+  // exists, released when the result settles), so a backend hot-reload mid-run
+  // cannot unregister the capture tool out from under this live child.
+  const structured: StructuredAcquisition | undefined = schema !== undefined ? acquireStructuredRuntime(ctx) : undefined
+
   const handle: AgentHandle = ctx.agents.create({
     agentId: childId,
     sessionId: SessionId(randomUUID()),
@@ -141,6 +173,7 @@ export function startInProcessRun(
     agentOptions,
   })
   const child = handle.agent
+  if (structured && schema !== undefined) structured.attach(child, schema)
 
   // Bridge the request's abort signal to the child (the consumer also bridges
   // its own exec.signal, but a backend-level bridge keeps the contract local).
@@ -149,6 +182,10 @@ export function startInProcessRun(
   // `turn/end` is logged — settles as `aborted` (honoring the cancel contract)
   // rather than falling through to the no-turn `error` mapping.
   let cancelled = false
+  // An accessor, not an inline read: `cancelled` mutates from closures (the
+  // abort listener, run.cancel), which control-flow narrowing cannot see — an
+  // inline read at the result mapping would narrow to the initializer.
+  const isCancelled = (): boolean => cancelled
   const requestCancel = (reason: string): void => {
     cancelled = true
     child.cancel(reason)
@@ -165,9 +202,16 @@ export function startInProcessRun(
       if (request.signal?.aborted) return { output: [], stopReason: 'aborted' }
       child.send(request.prompt)
       await child.whenIdle()
-      return readResult(child, seedLength, cancelled)
+      // Deliberately NO re-prompt when a structured child finishes cleanly
+      // without calling structured_output: readResult maps that to `error` —
+      // the shortfall goes to the parent instead of buying extra model turns.
+      return readResult(child, seedLength, isCancelled(), structured ? { captured: structured.captured(child) } : undefined)
     } finally {
       request.signal?.removeEventListener('abort', onAbort)
+      if (structured) {
+        structured.detach(child)
+        structured.release()
+      }
     }
   })()
 
@@ -195,12 +239,32 @@ export function startInProcessRun(
  * logged (a cancel landed in the pre-turn window, before any turn ran), the
  * run settles `aborted` per the {@link SubagentRun.cancel} contract rather than
  * the generic no-turn `error`.
+ *
+ * A structured run (`structured` present) additionally reports the captured
+ * value on {@link SubagentResult.structured}. A structured child that finished
+ * CLEANLY without ever capturing (the nudges ran out) settles `error` — a clean
+ * finish without the demanded structured result is a failure, not a success
+ * with a missing field; a non-`completed` reason keeps its own honest mapping.
  */
-function readResult(child: Agent, seedLength: number, cancelled: boolean): SubagentResult {
+function readResult(
+  child: Agent,
+  seedLength: number,
+  cancelled: boolean,
+  structured?: { captured?: { value: unknown } | undefined },
+): SubagentResult {
   const own = child.session.events.slice(seedLength)
   const lastMessage = own.findLast((e): e is SessionEvent<'assistant/message'> => e.type === 'assistant/message')
   const lastEnd = own.findLast((e): e is SessionEvent<'turn/end'> => e.type === 'turn/end')
   const output: ContentBlock[] = lastMessage ? structuredClone(lastMessage.data.content) : []
-  if (lastEnd === undefined && cancelled) return { output, stopReason: 'aborted' }
-  return { output, stopReason: toStopReason(lastEnd?.data.reason) }
+  const stopReason: SubagentStopReason = lastEnd === undefined && cancelled
+    ? 'aborted'
+    : toStopReason(lastEnd?.data.reason)
+  if (structured) {
+    if (structured.captured) return { output, structured: structured.captured.value, stopReason }
+    // No capture on a cleanly-completed turn: an ERROR when the run was left
+    // to finish (the nudges ran out), but ABORTED when a cancel is why the
+    // nudging stopped — the cancel contract outranks the schema shortfall.
+    if (stopReason === 'completed') return { output, stopReason: cancelled ? 'aborted' : 'error' }
+  }
+  return { output, stopReason }
 }

packages/subagent/subagent-inprocess/src/structured.ts

@@ -0,0 +1,312 @@
+/**
+ * Structured-output support for the in-process subagent backends: the mechanism
+ * behind `SubagentStartRequest.outputSchema` for children that run as agents on
+ * the same context.
+ *
+ * The model-facing surface is one globally registered `structured_output` tool
+ * whose REGISTERED parameters are a placeholder — the real schema is per run.
+ * Because the tool registry and prompt assembly are context-global while
+ * schemas differ per child (two concurrent structured runs may carry different
+ * schemas), per-agent shaping happens on the `system-prompt/assemble`
+ * waterfall with a `prepend: true` listener that post-processes `await next()`
+ * — FINAL-ASSEMBLY enforcement: whatever downstream listeners mutated or
+ * replaced, the assembly the loop renders never carries `structured_output`
+ * for an agent without a structured run, and for one that has it always
+ * carries the run's OWN schema plus a trailing
+ * {@link STRUCTURED_OUTPUT_INSTRUCTION} section (the demand travels with the
+ * tool). The loop logs what the assembly produced as the request header, so
+ * the injection is a reconstructable fact of the session log, never a
+ * wire-only mutation (the reconstructability RFC).
+ * (Cooperative mutate-then-`next()` would not survive a downstream listener
+ * returning a replacement assembly — see the waterfall composition caveat in
+ * docs/architecture.md.)
+ *
+ * FIXME: the whole enforcement dance above exists because the tool registry
+ * and prompt assembly are context-global. If they become per-agent or
+ * per-session scoped, a structured run just registers its own schema'd tool on
+ * the child's scope and this module reduces to the capture tool plus the
+ * turn-stop — no placeholder, no final-assembly swap, no strip-for-everyone-
+ * else, no global-registration lifetime dance.
+ *
+ * A companion `agent/turn-continuation` listener stops a child's turn once its
+ * output is captured — without it, the loop's default "had tool calls ⇒
+ * continue" buys a wasted extra model step per structured child. It is also
+ * `prepend: true`: the veto must run before any earlier-registered listener
+ * that could short-circuit the chain into a forced continue. A third listener
+ * closes the within-step window the continuation veto cannot: a
+ * `tools/pre-execute` deny for any call arriving after the agent's capture, so
+ * a response that lists `structured_output` before further tool calls cannot
+ * run side effects after the final answer was accepted. A fourth,
+ * `tools/post-execute`, is the capture COMMIT: the tool body only stages the
+ * validated value, and it becomes the run's captured result only when the
+ * final post-execute decision accepts the call — a blocking hook downstream
+ * yields `isError` in the log, and the run must not report success for it.
+ *
+ * Lifetime is refcounted by structured RUNS: each acquires from start to
+ * settle, so the registrations exist exactly while at least one structured
+ * child is live — a plain deployment that never passes `outputSchema` carries
+ * no always-on global state, and a backend hot-reload mid-run cannot
+ * unregister the capture tool out from under a live child (the run holds its
+ * own acquisition). Registrations land on the ROOT context and the refcount
+ * disposes them when the last run settles; the next structured run
+ * re-registers them.
+ *
+ * @module @deepseek-ai/dsh-subagent-inprocess/structured
+ */
+
+import type { Context } from 'cordis'
+import type { Agent } from '@deepseek-ai/dsh-agent'
+import type { ContentBlock, ToolSchema } from '@deepseek-ai/dsh-llm'
+import type { ContinuationDecision } from '@deepseek-ai/dsh-agent'
+import type { AssembleContext, PromptAssembly } from '@deepseek-ai/dsh-system-prompt'
+import type { PostToolDecision, PreToolDecision, ToolExecution, ToolExecutionResult } from '@deepseek-ai/dsh-tools'
+import { ToolArgsError, validateStructuredValue, type StructuredOutputSchema } from '@deepseek-ai/dsh-tools'
+
+/** The model-facing tool name a structured child must call to finish. */
+export const STRUCTURED_OUTPUT_TOOL = 'structured_output'
+
+/**
+ * The instruction the assembly listener appends to a structured child's
+ * system prompt as a trailing section on every assembly. Per-assembly state,
+ * NOT agent prompt state: `AgentOptions` has no prompt field (the persona is
+ * deployment config on the system-prompt plugin), so the same final-assembly
+ * enforcement that injects the schema'd tool carries the instruction that
+ * demands calling it.
+ */
+export const STRUCTURED_OUTPUT_INSTRUCTION
+  = 'When you have your final answer, you MUST report it by calling the '
+    + `\`${STRUCTURED_OUTPUT_TOOL}\` tool with arguments matching its parameter schema exactly. `
+    + 'Do not finish with a plain text answer: only the tool call counts as your result.'
+
+/** One structured run's state: the schema to enforce and the captured value, once recorded. */
+interface RunState {
+  readonly schema: StructuredOutputSchema
+  /**
+   * A validated value awaiting the post-execute verdict on ITS OWN call. Set
+   * by the capture tool's body, promoted to {@link RunState.captured} only
+   * when the final `tools/post-execute` decision accepts the call — a
+   * downstream block turns the logged result into `isError`, and a value
+   * committed at body time would let the run report success for a call the
+   * model saw fail.
+   */
+  pending?: { value: unknown }
+  captured?: { value: unknown }
+}
+
+/** The per-root-context runtime: run states plus the shared registrations. */
+interface StructuredRuntime {
+  refs: number
+  readonly states: WeakMap<Agent, RunState>
+  readonly disposers: (() => void)[]
+}
+
+/** One root context ⇒ one runtime (multi-app test isolation). */
+const runtimes = new WeakMap<Context, StructuredRuntime>()
+
+/**
+ * One holder's handle on the shared structured runtime. `release()` is
+ * idempotent per acquisition; the runtime's registrations are disposed when the
+ * LAST holder (backend plugin or live run) releases.
+ */
+export interface StructuredAcquisition {
+  /** Enforce `schema` on `agent`'s requests and start capturing its `structured_output` call. */
+  attach(agent: Agent, schema: StructuredOutputSchema): void
+  /** The captured value, once the child called the tool with valid arguments. */
+  captured(agent: Agent): { value: unknown } | undefined
+  /** Stop enforcing/capturing for `agent` (WeakMap-backed; safe to call twice). */
+  detach(agent: Agent): void
+  /** Drop this holder's reference (idempotent); the last release unregisters everything. */
+  release(): void
+}
+
+/**
+ * Acquire the per-root-context structured runtime, registering the capture tool
+ * and the runtime's listeners on the FIRST acquisition. See the module doc
+ * for the enforcement and lifetime design.
+ * @param ctx - any context of the app; the runtime keys off `ctx.root`.
+ * @returns this holder's handle (attach/captured/detach + idempotent release).
+ */
+export function acquireStructuredRuntime(ctx: Context): StructuredAcquisition {
+  const root: Context = ctx.root
+  let runtime = runtimes.get(root)
+  if (!runtime) {
+    runtime = { refs: 0, states: new WeakMap(), disposers: [] }
+    runtimes.set(root, runtime)
+    registerRuntime(root, runtime)
+  }
+  runtime.refs += 1
+
+  let released = false
+  return {
+    attach(agent: Agent, schema: StructuredOutputSchema): void {
+      runtime.states.set(agent, { schema })
+    },
+    captured(agent: Agent): { value: unknown } | undefined {
+      return runtime.states.get(agent)?.captured
+    },
+    detach(agent: Agent): void {
+      runtime.states.delete(agent)
+    },
+    release(): void {
+      if (released) return
+      released = true
+      runtime.refs -= 1
+      if (runtime.refs > 0) return
+      runtimes.delete(root)
+      for (const dispose of runtime.disposers.splice(0)) dispose()
+    },
+  }
+}
+
+/** Register the capture tool + the two listeners on the root context (first acquire). */
+function registerRuntime(root: Context, runtime: StructuredRuntime): void {
+  // The registered parameters are a PLACEHOLDER: the request listener below
+  // swaps in the run's real schema per child, and strips the tool entirely for
+  // every agent without a structured run — so this shape is never model-visible.
+  //
+  // Registration does NOT ride on the acquiring backend's plugin-level
+  // `inject`: a backend that waited on `tools` would apply later than it did
+  // before this module existed, shifting when its PROVIDER registers — and the
+  // delegation tool mirrors provider lifecycle, so that shift would reorder
+  // the model-visible tool list of every existing prompt. Instead the capture
+  // tool registers synchronously when `tools` is already live (the common
+  // case), and through a scoped inject fiber when the Loader happens to start
+  // the backend first. Either way the registration lands on root and is
+  // disposed by the runtime's refcount; disposing the fiber also covers the
+  // never-activated case.
+  let disposeTool: (() => void) | undefined
+  const registerCapture = (tools: Context['tools']): void => {
+    disposeTool = tools.register({
+      name: STRUCTURED_OUTPUT_TOOL,
+      description:
+        'Report your final structured result. Call this exactly once, when your answer is complete; '
+        + 'the arguments must match this tool\'s parameter schema exactly.',
+      parameters: { type: 'object', properties: {} },
+      execute(args: unknown, exec: ToolExecution): Promise<ContentBlock[]> {
+        const state = exec.agent ? runtime.states.get(exec.agent) : undefined
+        if (!state) {
+          // Reachable only if a non-structured agent somehow calls the tool (the
+          // request listener strips it, so the model never sees it) — fail loud
+          // rather than capture into nowhere.
+          throw new Error(`${STRUCTURED_OUTPUT_TOOL} is only available to subagents started with an output schema`)
+        }
+        const violations = validateStructuredValue(state.schema, args)
+        // ToolArgsError → isError result with INVALID_ARGS: the model retries
+        // within the same turn, exactly like a schema-validated defineTool call.
+        if (violations.length > 0) throw new ToolArgsError(violations)
+        // Two-phase commit: the body only STAGES the value; the post-execute
+        // listener below promotes it once the final decision accepts the call.
+        state.pending = { value: args }
+        return Promise.resolve([{ type: 'text', text: 'Structured output recorded.' }])
+      },
+    })
+  }
+  const liveTools = root.get('tools')
+  const toolsFiber = liveTools ? undefined : root.inject(['tools'], (childCtx: Context) => {
+    registerCapture(childCtx.root.tools)
+  })
+  if (liveTools) registerCapture(liveTools)
+  runtime.disposers.push(() => {
+    disposeTool?.()
+    void toolsFiber?.dispose()
+  })
+
+  // FINAL-ASSEMBLY enforcement (prepend: true = first registered = OUTERMOST
+  // wrapper): post-process whatever the downstream listeners and the registry
+  // produced, so a downstream listener returning a replacement assembly cannot
+  // leak the tool to other agents or erase the child's schema. The loop logs
+  // the rendered assembly as the step's request header, so the swap is
+  // reconstructable log state, never a wire-only mutation.
+  runtime.disposers.push(root.on('system-prompt/assemble', async function (
+    this: unknown, _assembly: PromptAssembly, context: AssembleContext, next: () => Promise<PromptAssembly>,
+  ): Promise<PromptAssembly> {
+    const final = await next()
+    const state = context.agent ? runtime.states.get(context.agent) : undefined
+    if (state) {
+      const schemaEntry: ToolSchema = {
+        name: STRUCTURED_OUTPUT_TOOL,
+        description:
+          'Report your final structured result. Call this exactly once, when your answer is complete; '
+          + 'the arguments must match this tool\'s parameter schema exactly.',
+        // ToolSchema.parameters is the wire-level JSON Schema object; the
+        // asserted subset type is structurally exactly that.
+        parameters: state.schema as unknown as Record<string, unknown>,
+      }
+      final.tools = [...final.tools.filter(tool => tool.name !== STRUCTURED_OUTPUT_TOOL), schemaEntry]
+      // The demand travels WITH the tool: a trailing section in the
+      // tool-guidance order band, appended after next() so it renders last
+      // (renderPrompt joins in array order).
+      final.sections = [...final.sections, { name: `tool:${STRUCTURED_OUTPUT_TOOL}`, order: 190, text: STRUCTURED_OUTPUT_INSTRUCTION }]
+      return final
+    }
+    // No structured run: strip the placeholder so it is never model-visible.
+    // An empty tools array canonicalizes to an absent header/wire field
+    // (canonicalHeader pins empty ≡ absent), so no re-shaping is needed here.
+    final.tools = final.tools.filter(tool => tool.name !== STRUCTURED_OUTPUT_TOOL)
+    return final
+  }, { prepend: true }))
+
+  // Stop a structured child's turn once its output is captured: the default
+  // "had tool calls ⇒ continue" would otherwise buy a wasted extra model step
+  // after every successful capture. `prepend: true` puts the veto OUTERMOST —
+  // an earlier-registered listener that short-circuits the chain (a goal-style
+  // force-continue returning without `next()`) would otherwise decide the turn
+  // before this listener ever ran, and no downstream decision may resurrect a
+  // structured turn that is already finished.
+  runtime.disposers.push(root.on('agent/turn-continuation', function (
+    this: unknown, agent: Agent, _turn: number, _decision: ContinuationDecision, next: () => Promise<ContinuationDecision>,
+  ): Promise<ContinuationDecision> {
+    if (runtime.states.get(agent)?.captured) return Promise.resolve({ action: 'stop' })
+    return next()
+  }, { prepend: true }))
+
+  // The capture COMMIT: promote the staged value only when the final
+  // post-execute decision accepts the call. The capture tool's body cannot
+  // decide — `tools/post-execute` runs after it, and a blocking listener (a
+  // PostToolUse hook) turns the logged result into `isError` feedback; a value
+  // committed at body time would make readResult report `structured` success
+  // for a call whose result the model and session log saw fail. `prepend:
+  // true` = outermost at registration time, so `await next()` returns the
+  // COMPOSED downstream decision — the same final verdict the registry maps
+  // onto the result. (A later-registered outer listener that blocks without
+  // delegating skips this commit entirely: the staged value is dropped and the
+  // run errors — failure-safe in the same direction.) The staging slot clears
+  // on every path, including a rejecting downstream listener.
+  runtime.disposers.push(root.on('tools/post-execute', async function (
+    this: unknown, exec: ToolExecution, _result: ToolExecutionResult, next: () => Promise<PostToolDecision>,
+  ): Promise<PostToolDecision> {
+    const state = exec.agent ? runtime.states.get(exec.agent) : undefined
+    if (!state || exec.name !== STRUCTURED_OUTPUT_TOOL || state.pending === undefined) return next()
+    const pending = state.pending
+    try {
+      const decision = await next()
+      if (decision.kind === 'accept') state.captured = pending
+      return decision
+    } finally {
+      delete state.pending
+    }
+  }, { prepend: true }))
+
+  // Terminal means terminal WITHIN the step, not only at its end: the
+  // turn-continuation veto above runs after every call in the current model
+  // response has executed, so a response that puts `structured_output` before
+  // further tool calls would still perform those side effects after the final
+  // answer was accepted. Deny every later call for a captured agent at the
+  // allow/deny gate — dispatch is skipped and the model sees an `isError`
+  // result naming the contract. Calls that PRECEDE the capture in the same
+  // response ran before `captured` was set and are untouched; a second
+  // `structured_output` is denied like any other call. `prepend: true` for the
+  // same reason as the continuation veto: no earlier-registered allow may
+  // short-circuit past the terminal contract.
+  runtime.disposers.push(root.on('tools/pre-execute', function (
+    this: unknown, exec: ToolExecution, next: () => Promise<PreToolDecision>,
+  ): Promise<PreToolDecision> {
+    if (exec.agent && runtime.states.get(exec.agent)?.captured) {
+      return Promise.resolve({
+        kind: 'deny',
+        reason: `structured output already recorded: the run is complete, so \`${exec.name}\` is not executed`,
+      })
+    }
+    return next()
+  }, { prepend: true }))
+}

packages/subagent/subagent-inprocess/tests/structured.spec.ts

@@ -0,0 +1,606 @@
+import { describe, expect, it } from 'vitest'
+import { Context } from 'cordis'
+import LlmService, { CallId, type ContentBlock, type GenerateOptions } from '@deepseek-ai/dsh-llm'
+import SessionStore from '@deepseek-ai/dsh-session'
+import SystemPrompt from '@deepseek-ai/dsh-system-prompt'
+import ToolRegistry from '@deepseek-ai/dsh-tools'
+import AgentRegistry, { AgentId } from '@deepseek-ai/dsh-agent'
+import type { Agent, ContinuationDecision } from '@deepseek-ai/dsh-agent'
+import AgentLoop from '@deepseek-ai/dsh-agent-loop'
+import * as Invariants from '@deepseek-ai/dsh-invariants'
+import SubagentService, { type SubagentStartRequest } from '@deepseek-ai/dsh-subagent'
+import type { StructuredOutputSchema } from '@deepseek-ai/dsh-tools'
+import { MockAdapter, textResponse, toolCallResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
+import { startInProcessRun } from '../src/index.ts'
+import {
+  acquireStructuredRuntime,
+  STRUCTURED_OUTPUT_INSTRUCTION,
+  STRUCTURED_OUTPUT_TOOL,
+} from '../src/structured.ts'
+
+type Script = ConstructorParameters<typeof MockAdapter>[0]
+
+const SCHEMA: StructuredOutputSchema = {
+  type: 'object',
+  properties: { answer: { type: 'number' }, note: { type: 'string' } },
+  required: ['answer'],
+}
+
+/**
+ * Real loop + scripted mock model + an INLINE spawn-shaped provider over the
+ * shared driver. The concrete backend plugins are deliberately NOT loaded —
+ * they would devDep-cycle this package (spawn/fork already depend on the
+ * driver), and the runtime under test is the driver's; plugin-level structured
+ * coverage lives in the spawn/fork specs. The mock model script drives the
+ * child's structured_output calls.
+ */
+async function setup(script: Script) {
+  const ctx = new Context()
+  const adapter = new MockAdapter(script)
+  await ctx.plugin(LlmService)
+  await ctx.plugin(SessionStore)
+  await ctx.plugin(SystemPrompt)
+  await ctx.plugin(ToolRegistry)
+  await ctx.plugin(AgentRegistry)
+  await ctx.plugin(Invariants)
+  await ctx.plugin(AgentLoop, { agents: [] })
+  await ctx.plugin(SubagentService)
+  const disposeProvider = ctx.subagents.registerProvider({
+    name: 'spawn',
+    capabilities: { outputSchema: true, depthLimit: true, toolFilter: false },
+    inheritsParentContext: false,
+    start: (request: SubagentStartRequest) => startInProcessRun(ctx, request, { providerName: 'spawn' }),
+  })
+  ctx.llm.registerAdapter(['mock'], adapter)
+  const parent = ctx.agentLoop.create(AgentId('parent'), { model: 'mock' })
+  return { ctx, parent, adapter, disposeProvider }
+}
+
+function structuredRequest(parent: SubagentStartRequest['parent'], extra?: Partial<SubagentStartRequest>): SubagentStartRequest {
+  return { prompt: [{ type: 'text', text: 'produce the answer' }], parent, outputSchema: SCHEMA, ...extra }
+}
+
+/** The tool names of one recorded model request. */
+function toolNames(request: GenerateOptions): string[] {
+  return (request.tools ?? []).map(tool => tool.name)
+}
+
+describe('in-process structured output', () => {
+  it('captures a valid structured_output call and surfaces result.structured', async () => {
+    const { ctx, parent } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 42, note: 'done' }),
+    ])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    expect(result.stopReason).toBe('completed')
+    expect(result.structured).toEqual({ answer: 42, note: 'done' })
+    await run.dispose()
+  })
+
+  it('stops the turn after a successful capture — no extra model step is spent', async () => {
+    const { ctx, parent, adapter } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 1 }),
+      textResponse('MUST NOT BE CONSUMED'),
+    ])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    await run.result
+    // Default continuation would run a second step after the tool call; the
+    // structured runtime's turn-continuation veto stops the turn instead.
+    expect(adapter.requests.length).toBe(1)
+    await run.dispose()
+  })
+
+  it('denies tool calls that FOLLOW the capture in the same response — terminal means terminal', async () => {
+    // One model response carrying structured_output FIRST and a side-effecting
+    // call after it: the continuation veto only fires at step end, so without
+    // the pre-execute deny the trailing call would still run after the final
+    // answer was accepted.
+    const response = [
+      ...toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 5 }).slice(0, -2),
+      { type: 'block-start', index: 1, blockType: 'tool-call' },
+      { type: 'block-end', index: 1, block: { type: 'tool-call', id: CallId('c2'), name: 'side_effect', arguments: '{}' } },
+      { type: 'usage', usage: { inputTokens: 10, outputTokens: 5 } },
+      { type: 'finish', reason: { kind: 'tool-calls' } },
+    ] as Script[number]
+    const { ctx, parent } = await setup([response])
+    let sideEffectRan = false
+    ctx.tools.register({
+      name: 'side_effect',
+      description: 'probe',
+      parameters: { type: 'object', properties: {} },
+      execute(): Promise<ContentBlock[]> {
+        sideEffectRan = true
+        return Promise.resolve([{ type: 'text', text: 'ran' }])
+      },
+    })
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    expect(result.stopReason).toBe('completed')
+    expect(result.structured).toEqual({ answer: 5 })
+    // The deny skipped dispatch entirely: the probe body never ran.
+    expect(sideEffectRan).toBe(false)
+    await run.dispose()
+  })
+
+  it('leaves tool calls that PRECEDE the capture in the same response untouched', async () => {
+    const response = [
+      { type: 'block-start', index: 0, blockType: 'tool-call' },
+      { type: 'block-end', index: 0, block: { type: 'tool-call', id: CallId('c1'), name: 'side_effect', arguments: '{}' } },
+      ...toolCallResponse('c2', STRUCTURED_OUTPUT_TOOL, { answer: 6 }).map(chunk =>
+        'index' in chunk ? { ...chunk, index: 1 } : chunk),
+    ] as Script[number]
+    const { ctx, parent } = await setup([response])
+    let sideEffectRan = false
+    ctx.tools.register({
+      name: 'side_effect',
+      description: 'probe',
+      parameters: { type: 'object', properties: {} },
+      execute(): Promise<ContentBlock[]> {
+        sideEffectRan = true
+        return Promise.resolve([{ type: 'text', text: 'ran' }])
+      },
+    })
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    // The call ran BEFORE captured was set: the deny gate only guards the
+    // window after the terminal answer landed.
+    expect(sideEffectRan).toBe(true)
+    expect(result.structured).toEqual({ answer: 6 })
+    await run.dispose()
+  })
+
+  it('snapshots the schema at start(): caller mutation after start cannot drift enforcement', async () => {
+    const mutable: StructuredOutputSchema = {
+      type: 'object',
+      properties: { answer: { type: 'number' } },
+      required: ['answer'],
+      additionalProperties: false,
+    }
+    const pristine = structuredClone(mutable)
+    const { ctx, parent, adapter } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 3 }),
+    ])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent, { outputSchema: mutable }))
+    // Mutate the caller's object AFTER start() returned but before the child's
+    // first request assembles: with a live reference this would reach both the
+    // model-visible parameters and validateStructuredValue.
+    ;(mutable.properties as Record<string, unknown>).answer = { type: 'string' }
+    const result = await run.result
+    expect(result.structured).toEqual({ answer: 3 })
+    // The child's request carried the PRISTINE schema, not the mutated one.
+    const childRequest = adapter.requests.at(-1)
+    const captureTool = (childRequest?.tools ?? []).find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)
+    expect(captureTool?.parameters).toEqual(pristine)
+    await run.dispose()
+  })
+
+  it('the captured-turn veto is prepend: an EARLIER force-continue listener cannot short-circuit it', async () => {
+    const ctx = new Context()
+    await ctx.plugin(SystemPrompt)
+    await ctx.plugin(ToolRegistry)
+    // Registered BEFORE the structured runtime exists — without prepend, this
+    // goal-style listener would decide the turn first (returning WITHOUT
+    // calling next()) and the veto would never run.
+    ctx.on('agent/turn-continuation', () => Promise.resolve<ContinuationDecision>({ action: 'continue' }))
+    const acquisition = acquireStructuredRuntime(ctx)
+    const agent = { id: AgentId('structured-child') } as unknown as Agent
+    acquisition.attach(agent, SCHEMA)
+    const captured = await ctx.tools.execute({
+      callId: 'call-1' as never,
+      name: STRUCTURED_OUTPUT_TOOL,
+      arguments: { answer: 1 },
+      agent,
+    })
+    expect(captured.isError).toBeFalsy()
+    const decision = await ctx.waterfall(
+      'agent/turn-continuation', agent, 1,
+      { action: 'continue' },
+      () => Promise.resolve<ContinuationDecision>({ action: 'continue' }),
+    )
+    expect(decision).toEqual({ action: 'stop' })
+    acquisition.detach(agent)
+    acquisition.release()
+  })
+
+  it('an invalid call gets an INVALID_ARGS isError result and the model retries in-turn', async () => {
+    const { ctx, parent } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 'not-a-number' }),
+      toolCallResponse('c2', STRUCTURED_OUTPUT_TOOL, { answer: 7 }),
+    ])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    expect(result.structured).toEqual({ answer: 7 })
+    expect(result.stopReason).toBe('completed')
+    // The child's log carries the isError tool/result for the invalid call.
+    const child = ctx.agents.get(run.id)!
+    const results = child.session.events.filter(e => e.type === 'tool/result')
+    expect(results.length).toBe(2)
+    expect((results[0]!.data as { isError?: boolean }).isError).toBe(true)
+    await run.dispose()
+  })
+
+  it('a clean finish without a capture is an immediate error to the parent — deliberately NO re-prompt', async () => {
+    const { ctx, parent, adapter } = await setup([
+      textResponse('here is my answer in prose'),
+      textResponse('MUST NOT BE CONSUMED'),
+    ])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    expect(result.stopReason).toBe('error')
+    expect(result.structured).toBeUndefined()
+    // Exactly one model request and one user message: no nudge turn exists.
+    expect(adapter.requests.length).toBe(1)
+    const child = ctx.agents.get(run.id)!
+    expect(child.session.events.filter(e => e.type === 'user/message').length).toBe(1)
+    await run.dispose()
+  })
+
+  it('an errored child keeps its honest error result (no capture expected)', async () => {
+    // Script exhaustion on the first call → the child turn errors.
+    const { ctx, parent, adapter } = await setup([])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    expect(result.stopReason).toBe('error')
+    expect(adapter.requests.length).toBe(1)
+    await run.dispose()
+  })
+
+  it('a cancel landing after a clean capture-less turn settles aborted, not error', async () => {
+    const { ctx, parent } = await setup([textResponse('prose, no capture')])
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const child = ctx.agents.get(run.id)!
+    // Cancel synchronously inside the turn's end recording: the cancel
+    // contract outranks the schema shortfall, so the result maps to aborted.
+    ctx.on('session/event', (session, event) => {
+      if (session === child.session && event.type === 'turn/end') run.cancel('cancelled at turn end')
+    })
+    const result = await run.result
+    expect(result.stopReason).toBe('aborted')
+    await run.dispose()
+  })
+
+  it('rejects a schema outside the subset loud, before any child exists', async () => {
+    const { ctx, parent } = await setup([])
+    expect(() => ctx.subagents.start('spawn', structuredRequest(parent, {
+      outputSchema: { type: 'object', oneOf: [] } as unknown as StructuredOutputSchema,
+    }))).toThrow(/unsupported output schema/)
+    expect(ctx.agents.get(AgentId('parent'))).toBeDefined()
+  })
+
+  it('a schema carrying non-JSON values fails as OutputSchemaError, never as a raw clone error', async () => {
+    const { ctx, parent } = await setup([])
+    // Assertion runs BEFORE the defensive structuredClone: a function-valued
+    // annotation must surface as the subset violation it is, not escape as
+    // structuredClone's DataCloneError.
+    expect(() => ctx.subagents.start('spawn', structuredRequest(parent, {
+      outputSchema: { type: 'object', default: () => {} } as unknown as StructuredOutputSchema,
+    }))).toThrow(/unsupported output schema.*annotation must be JSON data/)
+  })
+
+  it('a post-execute BLOCK on the capture call denies the capture: log and result agree on failure', async () => {
+    const { ctx, parent, adapter } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 7 }),
+      textResponse('continues after the blocked capture'),
+    ])
+    // A PostToolUse-style hook, registered AFTER the runtime (so the runtime's
+    // prepend commit listener stays outermost and composes this verdict).
+    ctx.on('tools/post-execute', (exec, _result, next) => {
+      if (exec.name === STRUCTURED_OUTPUT_TOOL) {
+        return Promise.resolve({ kind: 'block' as const, feedback: [{ type: 'text' as const, text: 'capture rejected by hook' }] })
+      }
+      return next()
+    })
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    // No capture was committed: the run reports the schema shortfall...
+    expect(result.structured).toBeUndefined()
+    expect(result.stopReason).toBe('error')
+    // ...the logged tool result is the blocked isError with the feedback...
+    const child = ctx.agents.get(run.id)!
+    const results = child.session.events.filter(e => e.type === 'tool/result')
+    expect((results[0]!.data as { isError?: boolean }).isError).toBe(true)
+    expect(JSON.stringify((results[0]!.data as { content: unknown }).content)).toContain('capture rejected by hook')
+    // ...and the turn CONTINUED past the blocked call (no captured veto):
+    // the model got to react to the failure with a second step.
+    expect(adapter.requests.length).toBe(2)
+    await run.dispose()
+  })
+
+  it('a post-execute accept-with-replacement still commits the capture', async () => {
+    const { ctx, parent } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 8 }),
+    ])
+    ctx.on('tools/post-execute', (exec, _result, next) => {
+      if (exec.name === STRUCTURED_OUTPUT_TOOL) {
+        return Promise.resolve({ kind: 'accept' as const, content: [{ type: 'text' as const, text: 'recorded (rewritten)' }] })
+      }
+      return next()
+    })
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    const result = await run.result
+    expect(result.stopReason).toBe('completed')
+    expect(result.structured).toEqual({ answer: 8 })
+    await run.dispose()
+  })
+
+  it('appends the structured instruction to the child REQUEST\'s system text (base prompt preserved)', async () => {
+    const { ctx, parent, adapter } = await setup([toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 1 })])
+    // A context-wide section stands in for the deployment persona: the
+    // instruction must APPEND to whatever the prompt pipeline assembled, not
+    // replace it (AgentOptions has no prompt field — the instruction is
+    // per-request wire state added by the final-request listener).
+    ctx.systemPrompt.section({ name: 'test:persona', order: 10, text: 'You are a counter.' })
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    await run.result
+    const childRequest = adapter.requests.at(-1)!
+    expect(childRequest.system).toContain('You are a counter.')
+    expect(childRequest.system!.endsWith(STRUCTURED_OUTPUT_INSTRUCTION)).toBe(true)
+    expect(childRequest.system!.indexOf(STRUCTURED_OUTPUT_INSTRUCTION)).toBeGreaterThan(0)
+    await run.dispose()
+  })
+
+  it('the instruction rides ONLY structured requests: appended for the child, absent for a plain agent', async () => {
+    const { ctx, parent, adapter } = await setup([
+      textResponse('parent answer'),
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 1 }),
+    ])
+    parent.send([{ type: 'text', text: 'hello' }])
+    await parent.whenIdle()
+    expect(adapter.requests[0]!.system ?? '').not.toContain(STRUCTURED_OUTPUT_INSTRUCTION)
+    const run = ctx.subagents.start('spawn', structuredRequest(parent))
+    await run.result
+    // The loop always assembles a base prompt (the harness identity section),
+    // so the instruction APPENDS — never replaces.
+    const childSystem = adapter.requests.at(-1)!.system!
+    expect(childSystem.endsWith(STRUCTURED_OUTPUT_INSTRUCTION)).toBe(true)
+    expect(childSystem.length).toBeGreaterThan(STRUCTURED_OUTPUT_INSTRUCTION.length)
+    await run.dispose()
+  })
+
+  describe('final-request enforcement (the prepend agent/request listener)', () => {
+    it('a plain agent assembling while the runtime is LIVE gets the placeholder stripped', async () => {
+      // Run-scoped acquisition means a plain deployment never registers the
+      // tool at all; the strip branch exists for the CONCURRENT case — a plain
+      // agent taking a turn while some structured child holds the runtime open.
+      const { ctx, parent, adapter } = await setup([textResponse('parent answer')])
+      const hold = acquireStructuredRuntime(ctx)
+      parent.send([{ type: 'text', text: 'hello' }])
+      await parent.whenIdle()
+      // The placeholder IS in the registry during this turn; the assembly the
+      // loop rendered must not carry it for an agent without a structured run.
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeDefined()
+      expect(toolNames(adapter.requests[0]!)).not.toContain(STRUCTURED_OUTPUT_TOOL)
+      hold.release()
+    })
+
+    it('a structured child sees structured_output with ITS schema; a plain agent never sees the tool', async () => {
+      const { ctx, parent, adapter } = await setup([
+        // Parent turn (a plain agent): must NOT see the tool.
+        textResponse('parent answer'),
+        // Child turn: must see it, with the run's schema.
+        toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 42 }),
+      ])
+      parent.send([{ type: 'text', text: 'hello' }])
+      await parent.whenIdle()
+      expect(toolNames(adapter.requests[0]!)).not.toContain(STRUCTURED_OUTPUT_TOOL)
+
+      const run = ctx.subagents.start('spawn', structuredRequest(parent))
+      await run.result
+      const childRequest = adapter.requests[1]!
+      expect(toolNames(childRequest)).toContain(STRUCTURED_OUTPUT_TOOL)
+      const entry = childRequest.tools!.find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)!
+      expect(entry.parameters).toEqual(SCHEMA)
+      await run.dispose()
+    })
+
+    it('two concurrent structured children each see their OWN schema', async () => {
+      const otherSchema: StructuredOutputSchema = {
+        type: 'object',
+        properties: { verdict: { type: 'string', enum: ['real', 'bogus'] } },
+        required: ['verdict'],
+      }
+      const { ctx, parent, adapter } = await setup([
+        (options: GenerateOptions) => {
+          // Answer with whatever schema this child was given — proves each
+          // request carried the right one regardless of scheduling order.
+          const entry = options.tools!.find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)!
+          const args = 'verdict' in (entry.parameters.properties as Record<string, unknown>)
+            ? { verdict: 'real' }
+            : { answer: 1 }
+          return toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, args)
+        },
+        (options: GenerateOptions) => {
+          const entry = options.tools!.find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)!
+          const args = 'verdict' in (entry.parameters.properties as Record<string, unknown>)
+            ? { verdict: 'real' }
+            : { answer: 1 }
+          return toolCallResponse('c2', STRUCTURED_OUTPUT_TOOL, args)
+        },
+      ])
+      const runA = ctx.subagents.start('spawn', structuredRequest(parent))
+      const runB = ctx.subagents.start('spawn', structuredRequest(parent, { outputSchema: otherSchema }))
+      const [a, b] = await Promise.all([runA.result, runB.result])
+      expect(a.structured).toEqual({ answer: 1 })
+      expect(b.structured).toEqual({ verdict: 'real' })
+      const schemas = adapter.requests.map(request =>
+        request.tools!.find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)!.parameters)
+      expect(schemas).toContainEqual(SCHEMA)
+      expect(schemas).toContainEqual(otherSchema)
+      await runA.dispose()
+      await runB.dispose()
+    })
+
+    it('wins against a downstream listener that REPLACES the assembly object', async () => {
+      const { ctx, parent, adapter } = await setup([
+        toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 5 }),
+      ])
+      // A downstream (non-prepend) listener that returns a brand-new assembly —
+      // the composition caveat that erases cooperative mutations. Registered
+      // AFTER the runtime's prepend listener, so it runs INSIDE it.
+      ctx.on('system-prompt/assemble', async (_assembly, _context, next) => {
+        const replaced = await next()
+        return { sections: [...replaced.sections], tools: [...replaced.tools], variables: { ...replaced.variables } }
+      })
+      const run = ctx.subagents.start('spawn', structuredRequest(parent))
+      const result = await run.result
+      expect(result.structured).toEqual({ answer: 5 })
+      const entry = adapter.requests[0]!.tools!.find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)
+      expect(entry).toBeDefined()
+      expect(entry!.parameters).toEqual(SCHEMA)
+      await run.dispose()
+    })
+
+    it('a non-structured agent request keeps tools ABSENT when it had none (no tools: [] materialized)', async () => {
+      const { parent, adapter } = await setup([
+        // The registry contributes the placeholder via prompt assembly, so
+        // tools is an array in the raw request — but after stripping the
+        // placeholder (its ONLY entry), the field must not be re-added as a
+        // different shape.
+        textResponse('plain'),
+      ])
+      parent.send([{ type: 'text', text: 'q' }])
+      await parent.whenIdle()
+      const request = adapter.requests[0]!
+      expect(toolNames(request)).not.toContain(STRUCTURED_OUTPUT_TOOL)
+      await new Promise(resolve => setTimeout(resolve, 0))
+    })
+
+    it('shapes a bare assembly on the waterfall: no-agent context strips the placeholder; a structured agent gains schema + trailing instruction section', async () => {
+      // Drive ctx.systemPrompt.assemble directly — the enforcement listener
+      // must tolerate a context with NO agent (a bare diagnostic assemble)
+      // and shape a structured agent's assembly on the same path the loop
+      // renders and logs as the request header.
+      const { ctx, parent } = await setup([])
+      const acquisition = acquireStructuredRuntime(ctx)
+      // Bare assemble WHILE the runtime is live: the no-agent branch must
+      // strip the registered placeholder (before the acquisition there is
+      // nothing to strip — run-scoped registration).
+      const bare = await ctx.systemPrompt.assemble({})
+      expect(bare.tools.map(tool => tool.name)).not.toContain(STRUCTURED_OUTPUT_TOOL)
+
+      acquisition.attach(parent, SCHEMA)
+      const shaped = await ctx.systemPrompt.assemble({ agent: parent })
+      expect(shaped.tools.map(tool => tool.name)).toContain(STRUCTURED_OUTPUT_TOOL)
+      expect(shaped.tools.find(tool => tool.name === STRUCTURED_OUTPUT_TOOL)!.parameters).toEqual(SCHEMA)
+      // The demand travels with the tool: the instruction renders LAST
+      // (appended post-next(); renderPrompt joins in array order).
+      expect(shaped.sections.at(-1)).toMatchObject({ name: `tool:${STRUCTURED_OUTPUT_TOOL}`, text: STRUCTURED_OUTPUT_INSTRUCTION })
+      acquisition.detach(parent)
+      acquisition.release()
+    })
+  })
+
+  describe('runtime lifetime (refcount: live structured runs)', () => {
+    it('the runtime exists exactly while structured runs are live: nothing before, nothing after', async () => {
+      const { ctx, parent } = await setup([
+        toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 4 }),
+      ])
+      // No always-on global state: a context that has run no structured child
+      // carries no capture tool.
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+      const run = ctx.subagents.start('spawn', structuredRequest(parent))
+      const result = await run.result
+      // The capture succeeded — the registrations existed while the run lived.
+      expect(result.structured).toEqual({ answer: 4 })
+      // The run's settle released the last acquisition.
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+      await run.dispose()
+    })
+
+    it('concurrent structured runs share one runtime; the last settle disposes it', async () => {
+      const { ctx, parent } = await setup([
+        toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 1 }),
+        toolCallResponse('c2', STRUCTURED_OUTPUT_TOOL, { answer: 2 }),
+      ])
+      const first = ctx.subagents.start('spawn', structuredRequest(parent))
+      const second = ctx.subagents.start('spawn', structuredRequest(parent))
+      const [a, b] = await Promise.all([first.result, second.result])
+      expect([a.structured, b.structured].sort()).toEqual([{ answer: 1 }, { answer: 2 }].sort())
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+      await first.dispose()
+      await second.dispose()
+    })
+
+    it('acquisition release is idempotent (double release cannot underflow the refcount)', async () => {
+      const ctx = new Context()
+      await ctx.plugin(SystemPrompt)
+      await ctx.plugin(ToolRegistry)
+      const first = acquireStructuredRuntime(ctx)
+      const second = acquireStructuredRuntime(ctx)
+      first.release()
+      first.release()
+      // The second holder still keeps the tool registered.
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeDefined()
+      second.release()
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    })
+
+    it('registers the capture tool through the scoped fiber when tools loads after the acquisition', async () => {
+      // The Loader starts sibling plugins concurrently, so a backend can
+      // acquire the runtime before dsh-tools has applied. The capture tool
+      // must then register as soon as `tools` exists — via the inject fiber,
+      // not by deferring the backend (which would reorder the prompt's tools).
+      const ctx = new Context()
+      const acquisition = acquireStructuredRuntime(ctx)
+      await ctx.plugin(SystemPrompt)
+      await ctx.plugin(ToolRegistry)
+      // Fiber activation completes asynchronously after the service appears.
+      await new Promise(resolve => setImmediate(resolve))
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeDefined()
+      acquisition.release()
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    })
+
+    it('releasing before tools ever loads disposes the pending fiber without registering', async () => {
+      const ctx = new Context()
+      const acquisition = acquireStructuredRuntime(ctx)
+      acquisition.release()
+      await ctx.plugin(SystemPrompt)
+      await ctx.plugin(ToolRegistry)
+      await new Promise(resolve => setImmediate(resolve))
+      // The disposed fiber never fires: nothing registers after the fact.
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    })
+
+    it('attach/captured/detach manage per-agent state through the acquisition surface', async () => {
+      const { ctx, parent } = await setup([])
+      const acquisition = acquireStructuredRuntime(ctx)
+      expect(acquisition.captured(parent)).toBeUndefined()
+      acquisition.attach(parent, SCHEMA)
+      expect(acquisition.captured(parent)).toBeUndefined()
+      acquisition.detach(parent)
+      acquisition.detach(parent)
+      acquisition.release()
+      // That manual acquisition was the ONLY holder - release disposes.
+      expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    })
+  })
+
+  it('a direct structured_output call from an agent WITHOUT a structured run is an isError', async () => {
+    const { ctx, parent } = await setup([])
+    // Hold the runtime open (run-scoped: nothing is registered otherwise) so
+    // the call reaches the capture tool's own fail-loud guard, not UNKNOWN_TOOL.
+    const hold = acquireStructuredRuntime(ctx)
+    const result = await ctx.tools.execute({
+      callId: 'x' as never,
+      name: STRUCTURED_OUTPUT_TOOL,
+      arguments: { answer: 1 },
+      agent: parent,
+    })
+    expect(result.isError).toBe(true)
+    expect(JSON.stringify(result.content)).toContain('only available to subagents')
+    hold.release()
+  })
+
+  it('a structured_output call with NO calling agent at all is an isError', async () => {
+    const { ctx } = await setup([])
+    const hold = acquireStructuredRuntime(ctx)
+    const result = await ctx.tools.execute({
+      callId: 'x' as never,
+      name: STRUCTURED_OUTPUT_TOOL,
+      arguments: { answer: 1 },
+    })
+    expect(result.isError).toBe(true)
+    hold.release()
+  })
+})

packages/subagent/subagent-inprocess/tsconfig.json

@@ -25,6 +25,12 @@
     },
     {
       "path": "../subagent"
+    },
+    {
+      "path": "../../core/system-prompt"
+    },
+    {
+      "path": "../../core/tools"
     }
   ]
 }

packages/subagent/subagent-spawn/README.md

@@ -10,7 +10,7 @@ The run mechanics live in the shared [`@deepseek-ai/dsh-subagent-inprocess`](../
 
 ## Capabilities
 
-`{ outputSchema: false, depthLimit: true, toolFilter: false }`. It constructs the child, so it enforces a recursion cap; structured output and tool-scoping are deferred (the service rejects a request needing either before `start` runs).
+`{ outputSchema: true, depthLimit: true, toolFilter: false }`. It constructs the child, so it enforces a recursion cap, and it supports structured output via the driver's [structured runtime](../subagent-inprocess/README.md) (acquired per structured run inside the driver — this backend registers nothing at apply). Tool-scoping is deferred (the service rejects a request needing it before `start` runs).
 
 ## Config
 

packages/subagent/subagent-spawn/src/index.ts

@@ -9,6 +9,11 @@
  * ({@link startInProcessRun}); this backend just passes NO seed (a fresh
  * child). The fork backend is an independent peer over the same driver.
  *
+ * Structured output (`outputSchema`) is supported via the driver's shared
+ * structured runtime: the backend acquires it for its plugin lifetime (so the
+ * capture tool and request-shaping listeners exist before any run), and each
+ * structured run holds its own acquisition until it settles.
+ *
  * Plugin export shape: named `name`/`inject`/`Config`/`apply`, NO default.
  *
  * @module @deepseek-ai/dsh-subagent-spawn
@@ -20,6 +25,11 @@ import type { SubagentCapabilities, SubagentProvider, SubagentStartRequest } fro
 import { startInProcessRun } from '@deepseek-ai/dsh-subagent-inprocess'
 
 export const name = 'subagent-spawn'
+// `tools` is deliberately NOT injected: the shared driver's structured runtime
+// (acquired per structured RUN, not at apply) gates its own capture-tool
+// registration on `tools` availability, so this backend's apply timing — and
+// with it the provider-mirroring delegation tool's position in the
+// model-visible tool list — stays what it was before structured output existed.
 export const inject = ['subagents', 'agents']
 
 /** Config: the registry name to register the provider under. */
@@ -34,11 +44,12 @@ export const Config: z<Config> = z.object({
 
 /**
  * The spawn provider. Supports `depthLimit` (it constructs the child, so it can
- * enforce a recursion cap) but NOT `outputSchema` or `toolFilter` in this cut —
- * a request that needs either is rejected by the service before `start` runs.
+ * enforce a recursion cap) and `outputSchema` (via the shared in-process
+ * structured runtime); NOT `toolFilter` in this cut — a request that needs it
+ * is rejected by the service before `start` runs.
  */
 class SpawnProvider implements SubagentProvider {
-  readonly capabilities: SubagentCapabilities = { outputSchema: false, depthLimit: true, toolFilter: false }
+  readonly capabilities: SubagentCapabilities = { outputSchema: true, depthLimit: true, toolFilter: false }
   // Context contract: a spawned child starts fresh — it never sees the parent conversation.
   readonly inheritsParentContext = false
 
@@ -46,7 +57,8 @@ class SpawnProvider implements SubagentProvider {
 
   start(request: SubagentStartRequest) {
     // Fresh child: no seed. The shared driver mints ids, stamps cwd/lineage/
-    // depth, drives the one-shot, and maps the result.
+    // depth, drives the one-shot (including the structured capture when the
+    // request carries an outputSchema), and maps the result.
     return startInProcessRun(this.ctx, request, { providerName: this.name })
   }
 }

packages/subagent/subagent-spawn/tests/subagent-spawn.spec.ts

@@ -10,9 +10,9 @@ import { SessionId } from '@deepseek-ai/dsh-session'
 import AgentLoop from '@deepseek-ai/dsh-agent-loop'
 import * as Invariants from '@deepseek-ai/dsh-invariants'
 import SubagentService from '@deepseek-ai/dsh-subagent'
-import { MockAdapter, maxTokensResponse, textResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
+import { MockAdapter, maxTokensResponse, textResponse, toolCallResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
 import * as spawn from '../src/index.ts'
-import { depthOf, SubagentDepthError } from '@deepseek-ai/dsh-subagent-inprocess'
+import { depthOf, STRUCTURED_OUTPUT_TOOL, SubagentDepthError } from '@deepseek-ai/dsh-subagent-inprocess'
 
 type Script = ConstructorParameters<typeof MockAdapter>[0]
 
@@ -241,10 +241,10 @@ describe('dsh-subagent-spawn', () => {
     await parentHandle.dispose()
   })
 
-  it('advertises depthLimit but not outputSchema/toolFilter', async () => {
+  it('advertises depthLimit and outputSchema but not toolFilter', async () => {
     const { ctx } = await setup([])
     const provider = ctx.subagents.getProvider('spawn')!
-    expect(provider.capabilities).toEqual({ outputSchema: false, depthLimit: true, toolFilter: false })
+    expect(provider.capabilities).toEqual({ outputSchema: true, depthLimit: true, toolFilter: false })
   })
 
   it('unregisters the provider when its fiber is disposed (HMR safety)', async () => {
@@ -257,6 +257,54 @@ describe('dsh-subagent-spawn', () => {
     expect(ctx.subagents.list()).toEqual([])
   })
 
+  it('captures structured output through the shipped plugin (driver runtime, plugin wiring)', async () => {
+    const { ctx, parent } = await setup([
+      toolCallResponse('c1', STRUCTURED_OUTPUT_TOOL, { answer: 42 }),
+    ])
+    const run = ctx.subagents.start('spawn', {
+      prompt: [{ type: 'text', text: 'produce the answer' }],
+      parent,
+      outputSchema: { type: 'object', properties: { answer: { type: 'number' } }, required: ['answer'] },
+    })
+    const result = await run.result
+    expect(result.stopReason).toBe('completed')
+    expect(result.structured).toEqual({ answer: 42 })
+    // Run-scoped runtime: the settle released the last acquisition.
+    expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    await run.dispose()
+  })
+
+  it('a backend unload mid-structured-run settles the run and releases the runtime', async () => {
+    // Rebuild the stack by hand so we hold the backend's fiber.
+    const ctx = new Context()
+    const adapter = new MockAdapter(['hang'])
+    await ctx.plugin(LlmService)
+    await ctx.plugin(SessionStore)
+    await ctx.plugin(SystemPrompt)
+    await ctx.plugin(ToolRegistry)
+    await ctx.plugin(AgentRegistry)
+    await ctx.plugin(Invariants)
+    await ctx.plugin(AgentLoop, { agents: [] })
+    await ctx.plugin(SubagentService)
+    const fiber = await ctx.plugin(spawn, { providerName: 'spawn' })
+    ctx.llm.registerAdapter(['mock'], adapter)
+    const parent = ctx.agentLoop.create(AgentId('parent'), { model: 'mock' })
+    const run = ctx.subagents.start('spawn', {
+      prompt: [{ type: 'text', text: 'q' }],
+      parent,
+      outputSchema: { type: 'object', properties: { a: { type: 'number' } } },
+    })
+    // Let the child's step start streaming, then unload the backend. The
+    // backend owns the child agent, so the unload tears the child down and
+    // the run settles — releasing its own runtime acquisition on the way out.
+    await new Promise(resolve => setTimeout(resolve, 30))
+    await fiber.dispose()
+    const result = await run.result
+    expect(result.stopReason).toBe('error')
+    expect(ctx.tools.get(STRUCTURED_OUTPUT_TOOL)).toBeUndefined()
+    await run.dispose()
+  })
+
   it('has the namespace-plugin export shape (no stray default)', () => {
     expect('default' in spawn).toBe(false)
     expect(spawn.name).toBe('subagent-spawn')

packages/subagent/subagent/src/types.ts

@@ -8,7 +8,7 @@
 
 import type { Agent, AgentId, AgentOptions } from '@deepseek-ai/dsh-agent'
 import type { ContentBlock } from '@deepseek-ai/dsh-llm'
-import type { SchemaSpec } from '@deepseek-ai/dsh-tools'
+import type { StructuredOutputSchema } from '@deepseek-ai/dsh-tools'
 
 /**
  * Which START-TIME features a provider supports. Checked by the service
@@ -56,12 +56,16 @@ export interface SubagentStartRequest {
   /** Per-child agent options (model, system prompt). */
   agentOptions?: AgentOptions
   /**
-   * Optional structured-output schema. When set AND the provider's
-   * {@link SubagentCapabilities.outputSchema} is `true`, the child's final
-   * answer is shaped to this schema and surfaced as {@link SubagentResult.structured}.
+   * Optional structured-output schema — an object-rooted JSON Schema within the
+   * enforced subset (see `assertSupportedOutputSchema` in dsh-tools; a schema
+   * outside the subset is rejected loud at start). When set AND the provider's
+   * {@link SubagentCapabilities.outputSchema} is `true`, the child is driven to
+   * report a value matching this schema, surfaced as
+   * {@link SubagentResult.structured}. The schema must be plain host-realm JSON
+   * data — a caller holding foreign-realm data materializes it first.
    * Requesting it against a provider that lacks the capability is rejected at start.
    */
-  outputSchema?: SchemaSpec
+  outputSchema?: StructuredOutputSchema
   /**
    * Optional recursion cap (max delegation depth below this child). Requires
    * {@link SubagentCapabilities.depthLimit}; rejected at start otherwise.

packages/subagent/subagent/tests/service.spec.ts

@@ -178,7 +178,7 @@ describe('SubagentService', () => {
 
   describe('start-time capability validation (fail loud, before any child)', () => {
     it.each([
-      { field: 'outputSchema', request: baseRequest({ outputSchema: { x: { type: 'string' } } }) },
+      { field: 'outputSchema', request: baseRequest({ outputSchema: { type: 'object', properties: { x: { type: 'string' } } } }) },
       { field: 'maxDepth', request: baseRequest({ maxDepth: 2 }) },
       { field: 'toolFilter', request: baseRequest({ toolFilter: { deny: ['bash'] } }) },
     ])('rejects $field against a provider that lacks the capability — before start() runs', ({ request }) => {
@@ -203,7 +203,7 @@ describe('SubagentService', () => {
       await ctx.plugin(SubagentService)
       const provider = new StubProvider('strong', ALL_CAPS)
       ctx.subagents.registerProvider(provider)
-      ctx.subagents.start('strong', baseRequest({ outputSchema: { x: { type: 'string' } }, maxDepth: 1 }))
+      ctx.subagents.start('strong', baseRequest({ outputSchema: { type: 'object', properties: { x: { type: 'string' } } }, maxDepth: 1 }))
       expect(provider.startCount).toBe(1)
     })
   })

packages/support/subagent-mock/tests/subagent-mock.spec.ts

@@ -41,13 +41,13 @@ describe('dsh-subagent-mock', () => {
 
   it('surfaces a structured result when the request carries an outputSchema', async () => {
     const ctx = await mount({ reply: 'r', structured: { answer: 42 } })
-    const run = ctx.subagents.start('mock', baseRequest({ outputSchema: { answer: { type: 'number' } } }))
+    const run = ctx.subagents.start('mock', baseRequest({ outputSchema: { type: 'object', properties: { answer: { type: 'number' } } } }))
     await expect(run.result).resolves.toMatchObject({ structured: { answer: 42 } })
   })
 
   it('defaults structured output to { reply } when outputSchema is requested but no structured value is configured', async () => {
     const ctx = await mount({ reply: 'fallback reply' })
-    const run = ctx.subagents.start('mock', baseRequest({ outputSchema: { answer: { type: 'number' } } }))
+    const run = ctx.subagents.start('mock', baseRequest({ outputSchema: { type: 'object', properties: { answer: { type: 'number' } } } }))
     await expect(run.result).resolves.toMatchObject({ structured: { reply: 'fallback reply' } })
   })
 

packages/ui/acp-agent/README.md

@@ -24,6 +24,7 @@ Because the package wires no logger entry, an ACP leaf has **nothing to get wron
 |---|---|---|
 | `model` | (required) | the per-session agent template the bridge creates agents from |
 | `persona` | — | the deployment persona template (may reference `{{model}}`/`{{cwd}}`), routed to `dsh-system-prompt` |
+| `toolOrder` | — | explicit model-facing tool order (a name list with one `'<unlisted-tools>'` rest entry; absent — lexicographic; an unregistered name fails each turn at prompt assembly), routed to `dsh-system-prompt` |
 | `persistenceRoot` | `./.sessions` | the JSONL backend's root directory |
 
 The leaf supplies the swappable backends: an LLM adapter (`llm-deepseek` for the real model, `llm-replay` for keyless snapshot replay) and a bash executor (`bash-local`).

packages/ui/acp-agent/package.json

@@ -45,6 +45,7 @@
     "@deepseek-ai/dsh-app-boot": "workspace:^",
     "@deepseek-ai/dsh-acp": "workspace:^",
     "@deepseek-ai/dsh-agent-core": "workspace:^",
+    "@deepseek-ai/dsh-system-prompt": "workspace:^",
     "@deepseek-ai/dsh-session-persistence-jsonl": "workspace:^",
     "cordis": "^4.0.0-rc.6",
     "schemastery": "^3.17.0"

packages/ui/acp-agent/src/index.ts

@@ -42,7 +42,8 @@ export const name = 'acp-agent'
  * App config: the swappable per-deployment values. `model` configures the
  * agent template the ACP bridge creates each session's agent from (NOT a
  * pre-created agent — ACP creates agents at `session/new`); `persona` is the
- * deployment persona (forwarded to the system-prompt plugin);
+ * deployment persona (forwarded to the system-prompt plugin); `toolOrder` is
+ * the explicit model-facing tool order (forwarded to the system-prompt plugin);
  * `persistenceRoot` is the JSONL backend's directory.
  */
 export interface Config {
@@ -50,6 +51,8 @@ export interface Config {
   model: string
   /** Deployment persona (the system-prompt plugin's `persona` config). */
   persona?: string
+  /** Explicit model-facing tool order (the system-prompt plugin's `toolOrder` config; see dsh-system-prompt). */
+  toolOrder?: string[]
   /** Directory the JSONL session backend writes under. Defaults to `./.sessions`. */
   persistenceRoot?: string
 }
@@ -57,6 +60,10 @@ export interface Config {
 export const Config: z<Config> = z.object({
   model: z.string().required(),
   persona: z.string(),
+  // The array default is forced to undefined: ABSENT means "lexicographic
+  // order" (the owning dsh-system-prompt schema does the same), while
+  // schemastery's native [] default would read as an invalid configured list.
+  toolOrder: z.array(z.string()).default(undefined as unknown as string[]),
   persistenceRoot: z.string().default('./.sessions'),
 })
 
@@ -70,6 +77,7 @@ export const Config: z<Config> = z.object({
 export function apply(ctx: Context, config: Config): void {
   ctx.plugin(agentCore, {
     ...config.persona !== undefined ? { persona: config.persona } : {},
+    ...config.toolOrder !== undefined ? { toolOrder: config.toolOrder } : {},
   })
   ctx.plugin(SessionPersistenceJsonl, { root: config.persistenceRoot ?? './.sessions' })
   ctx.plugin(acp, { model: config.model })

packages/ui/acp-agent/tests/acp-agent.spec.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest'
 import { Context } from 'cordis'
 import Loader from '@cordisjs/plugin-loader'
+import { TOOL_ORDER_REST } from '@deepseek-ai/dsh-system-prompt'
 import * as acpAgent from '../src/index.ts'
 
 /**
@@ -52,6 +53,27 @@ describe('dsh-acp-agent composition', () => {
     expect(acpAgent.Config).toBeDefined()
   })
 
+  it('forwards toolOrder through agent-core to the system-prompt assembly', async () => {
+    const ctx = await mount({
+      model: 'mock',
+      toolOrder: ['zulu', TOOL_ORDER_REST],
+      persistenceRoot: '/tmp/dsh-acp-agent-test-tool-order',
+    })
+    // The bundle's own bash tools pend on the absent `ctx.bash` executor in
+    // this providerless mount, so register two plain tools to order.
+    for (const name of ['alpha', 'zulu']) {
+      ctx.get('tools')!.register({
+        name,
+        description: name,
+        parameters: {},
+        execute: async () => [],
+      })
+    }
+    const assembly = await ctx.get('systemPrompt')!.assemble()
+    expect(assembly.tools.map(tool => tool.name)).toEqual(['zulu', 'alpha'])
+    await ctx.fiber.dispose()
+  })
+
   it('has the namespace-plugin export shape (no stray default) so the Loader keeps name/Config/apply', () => {
     // Postmortem 0001 guard: a stray `export default apply` makes the Loader's
     // `unwrapExports` (`exports.default ?? exports`) collapse the module to the

packages/ui/stdio-agent/README.md

@@ -25,6 +25,7 @@ The leaf `cordis.yml` supplies only the **swappable backends** — an LLM adapte
 |---|---|---|
 | `model` | (required) | the pre-created `main` agent's model |
 | `persona` | — | the deployment persona template (may reference `{{model}}`), routed to `dsh-system-prompt` |
+| `toolOrder` | — | explicit model-facing tool order (a name list with one `'<unlisted-tools>'` rest entry; absent — lexicographic; an unregistered name fails each turn at prompt assembly), routed to `dsh-system-prompt` |
 | `persistenceRoot` | `./.sessions` | the JSONL backend's root directory |
 | `welcome` | `ready.` | the stdin-chat banner |
 | `resumeSessionId` | — | resume a persisted session id instead of starting fresh (sourced from an env var in the leaf) |

packages/ui/stdio-agent/package.json

@@ -50,6 +50,7 @@
     "@deepseek-ai/dsh-agent": "workspace:^",
     "@deepseek-ai/dsh-llm": "workspace:^",
     "@deepseek-ai/dsh-agent-core": "workspace:^",
+    "@deepseek-ai/dsh-system-prompt": "workspace:^",
     "@deepseek-ai/dsh-session": "workspace:^",
     "@deepseek-ai/dsh-session-persistence-jsonl": "workspace:^",
     "cordis": "^4.0.0-rc.6",

packages/ui/stdio-agent/src/index.ts

@@ -53,7 +53,8 @@ export const name = 'stdio-agent'
  * App config: the swappable per-demo values, each routed to where the app wires
  * it. `model`/`resumeSessionId` configure the pre-created `main` agent (through
  * {@link @deepseek-ai/dsh-agent-core}'s forwarded `agents` list); `persona` is
- * the deployment persona (forwarded to the system-prompt plugin);
+ * the deployment persona (forwarded to the system-prompt plugin); `toolOrder`
+ * is the explicit model-facing tool order (forwarded to the system-prompt plugin);
  * `persistenceRoot` is the JSONL backend's directory; `welcome` is the UI banner.
  */
 export interface Config {
@@ -61,6 +62,8 @@ export interface Config {
   model: string
   /** Deployment persona (the system-prompt plugin's `persona` config). */
   persona?: string
+  /** Explicit model-facing tool order (the system-prompt plugin's `toolOrder` config; see dsh-system-prompt). */
+  toolOrder?: string[]
   /** Directory the JSONL session backend writes under. Defaults to `./.sessions`. */
   persistenceRoot?: string
   /** stdin-chat banner printed once on start. Defaults to `'ready.'`. */
@@ -76,6 +79,10 @@ export interface Config {
 export const Config: z<Config> = z.object({
   model: z.string().required(),
   persona: z.string(),
+  // The array default is forced to undefined: ABSENT means "lexicographic
+  // order" (the owning dsh-system-prompt schema does the same), while
+  // schemastery's native [] default would read as an invalid configured list.
+  toolOrder: z.array(z.string()).default(undefined as unknown as string[]),
   persistenceRoot: z.string().default('./.sessions'),
   welcome: z.string().default('ready.'),
   resumeSessionId: z.string(),
@@ -92,6 +99,7 @@ export function apply(ctx: Context, config: Config): void {
   ctx.plugin(ConsoleExporter)
   ctx.plugin(agentCore, {
     ...config.persona !== undefined ? { persona: config.persona } : {},
+    ...config.toolOrder !== undefined ? { toolOrder: config.toolOrder } : {},
     agents: [{
       id: AgentId('main'),
       model: config.model,

packages/ui/stdio-agent/tests/built-bin.e2e.ts

@@ -76,9 +76,10 @@ async function makeConsumer(welcome: string, disabledBrokenEntry = false): Promi
     await mkdir(dirname(target), { recursive: true })
     await symlink(abs, target)
   }
-  // The example's mock model + echo tool are example-local TS plugins (Node 24+
-  // strips types natively, so plain `node` loads them); they import the workspace
-  // packages the symlinked node_modules now provides.
+  // The example's mock model + echo tool are example-local TS plugins (Node
+  // 22.19+ — the engines floor — strips types natively, so plain `node` loads
+  // them); they import the workspace packages the symlinked node_modules now
+  // provides.
   await cp(join(repoRoot, 'examples/echo-agent/src'), join(dir, 'src'), { recursive: true })
   await writeFile(join(dir, 'cordis.yml'), [
     '- id: mock-llm',

packages/ui/stdio-agent/tests/stdio-agent.spec.ts

@@ -2,6 +2,7 @@ import { describe, it, expect } from 'vitest'
 import { Context } from 'cordis'
 import Loader from '@cordisjs/plugin-loader'
 import { AgentId } from '@deepseek-ai/dsh-agent'
+import { TOOL_ORDER_REST } from '@deepseek-ai/dsh-system-prompt'
 import * as stdioAgent from '../src/index.ts'
 
 /**
@@ -74,6 +75,27 @@ describe('dsh-stdio-agent app', () => {
     expect(stdioAgent.Config).toBeDefined()
   })
 
+  it('forwards toolOrder through agent-core to the system-prompt assembly', async () => {
+    const ctx = await mount({
+      model: 'mock',
+      toolOrder: ['zulu', TOOL_ORDER_REST],
+      persistenceRoot: '/tmp/dsh-stdio-agent-spec-tool-order',
+    })
+    // The bundle's own bash tools pend on the absent `ctx.bash` executor in
+    // this providerless mount, so register two plain tools to order.
+    for (const name of ['alpha', 'zulu']) {
+      ctx.get('tools')!.register({
+        name,
+        description: name,
+        parameters: {},
+        execute: async () => [],
+      })
+    }
+    const assembly = await ctx.get('systemPrompt')!.assemble()
+    expect(assembly.tools.map(tool => tool.name)).toEqual(['zulu', 'alpha'])
+    await ctx.fiber.dispose()
+  })
+
   it('has the namespace-plugin export shape (no stray default) so the Loader keeps name/Config/apply', () => {
     // Postmortem 0001 guard: a stray `export default apply` makes the Loader's
     // `unwrapExports` (`exports.default ?? exports`) collapse the module to the

packages/web/web-fetch-local/src/provider.ts

@@ -1,6 +1,6 @@
 /**
  * `LocalFetchProvider`: a `WebFetchProvider` that retrieves a concrete public
- * HTTP(S) URL with the platform-native `fetch` (Node 24) and returns a status
+ * HTTP(S) URL with platform-native `fetch` at the repo's Node floor and returns a status
  * code plus bounded decoded content. It owns SAFE RESOURCE RETRIEVAL — URL
  * validation, redirect policy, timeout, abort, byte caps, charset decoding,
  * content-type classification, binary rejection — but NOT presentation

packages/web/web-search-deepseek/src/provider.ts

@@ -12,7 +12,7 @@
  * `web_search_tool_result` block (native search did not trigger), it throws
  * `WEB_PROVIDER_ERROR` rather than degrading to prose-scraping.
  *
- * Network requests use platform-native `fetch` (Node 24), mirroring
+ * Network requests use platform-native `fetch` at the repo's Node floor, mirroring
  * `@deepseek-ai/dsh-llm-deepseek`'s adapter — not a cordis HTTP-client service.
  * The Anthropic wire shape is a provider-private detail and does NOT make this
  * provider depend on `ctx.llm`.

packages/web/web-search-exa/src/provider.ts

@@ -6,7 +6,7 @@
  * `title`, the first highlight as `snippet`, and `publishedDate` as
  * `publishedAt`.
  *
- * Network requests use platform-native `fetch` (Node 24), mirroring
+ * Network requests use platform-native `fetch` at the repo's Node floor, mirroring
  * `@deepseek-ai/dsh-llm-deepseek`'s adapter — not a cordis HTTP-client service.
  *
  * @module @deepseek-ai/dsh-web-search-exa/provider

packages/web/web-search-perplexity/src/provider.ts

@@ -5,7 +5,7 @@
  * structured `search_results[]` for `sources[]`, falling back to the URL-only
  * `citations[]` when `search_results` is absent.
  *
- * Network requests use platform-native `fetch` (Node 24), mirroring
+ * Network requests use platform-native `fetch` at the repo's Node floor, mirroring
  * `@deepseek-ai/dsh-llm-deepseek`'s adapter. The OpenAI-compatible request shape
  * is a provider-private detail and does NOT make this provider depend on
  * `ctx.llm`.