fix(subagent): keep output past an empty terminal message with one selection rule

A max-tokens step that assembled only tool-call blocks appends an EMPTY-content assistant/message (the usage host). Three consumers each hand-rolled output selection and all let it erase the child's real answer: the in-process readResult and the Activation subagent/end capture took the last message unfiltered, and the SDK backend let any message beat its streamed-text fallback; the in-process driver also had no streamed-text fallback for cancelled turns. dsh-subagent now owns the canonical rule in src/assistant-output.ts (last non-empty assistant message, else the accumulated text-delta stream) and all three consumers apply it. Regression tests in all three packages fail under the previous selections. Closes #1514
2026-08-10 11:15:01 +08:00
parent abaf8f5061
commit 1db1cda464
29 changed files with 344 additions and 66 deletions
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write packages/subagent/subagent-inprocess/README.md
-README.md: 67f0cf5dd1ecb18542af56953a0eaa40988aca0d
-README.zh.md: 648a160be5f1c3dcbe66a867a273a3df610dbc0a
+README.md: 2e2a3873843467b0811ccbc0ed1d9bb6a83eb31f
+README.zh.md: 91013164e762bb01e7ad5a51597c6fa559c8d7a3
@@ -14,7 +14,7 @@ The driver follows this sequence:
 2. Call `parent.ctx.agents.create` directly, passing the required request signal into the factory's creation transaction.
 3. During that transaction's unpublished setup window, install the requested persona, tool restriction, and structured-output runtime.
 4. Publish the child, retain the returned `AgentHandle`, and drive one task with `child.followup(prompt)` followed by `child.whenIdle()`.
-5. Read the child's own last assistant message and final durable turn reason from the complete owned child run, excluding any fork seed.
+5. Read the child's own output — its last NON-EMPTY assistant message (an empty-content message hosts only usage and is skipped), else the text it streamed before cancel or truncation cut the turn short — and the final durable turn reason from the complete owned child run, excluding any fork seed.

 The child gets the parent's working-directory/session lineage and inherits the parent provider, model, and output-token cap unless `request.agentOptions` overrides them. It gets a fresh flat registration scope: parent ownership does not import parent tool restrictions or establish an authority subset.

@@ -14,7 +14,7 @@
 2. 直接调用 `parent.ctx.agents.create`，把必需的请求信号传入工厂的创建事务。
 3. 在该事务未发布的设置窗口中，安装请求的 persona、工具限制和结构化输出运行时。
 4. 发布子 agent，保留返回的 `AgentHandle`，并通过先调用 `child.followup(prompt)`、再调用 `child.whenIdle()` 来驱动一项任务。
-5. 从完整的自有子运行中读取子 agent 自身最后一条 assistant 消息和最终持久化的轮次原因，并排除任何 fork 初始内容。
+5. 从完整的自有子运行中读取子 agent 自身的输出——最后一条**非空** assistant 消息（空内容消息仅承载 usage，会被跳过），否则取轮次被取消或截断前已流式的文本——以及最终持久化的轮次原因，并排除任何 fork 初始内容。

 子 agent 会获得父 agent 的工作目录／会话谱系；除非 `request.agentOptions` 覆盖，否则还会继承父 agent 的提供方、模型和输出 token 上限。它获得全新的扁平注册作用域：父级所有权不会导入父 agent 的工具限制，也不会建立权限子集。

@@ -20,6 +20,7 @@ import {
  applyChildComposition,
  assertSubagentMaxDepth,
  childSessionMeta,
+  finalAssistantOutput,
  resolveChildAgentOptions,
  resolveChildDepth,
 } from '@deepseek-ai/dsh-subagent'
@@ -218,9 +219,11 @@ function readResult(
  structured?: { captured?: { value: unknown } | undefined },
 ): SubagentResult {
  const own = child.session.events.slice(boundary)
-  const lastMessage = own.findLast((event): event is SessionEvent<'assistant/message'> => event.type === 'assistant/message')
  const lastEnd = findLastMessageTurnEnd(own)
-  const output: ContentBlock[] = lastMessage?.data.message.content ?? []
+  // Canonical selection (`finalAssistantOutput`): the last non-empty assistant
+  // message, else the text streamed before cancel/error/truncation cut the
+  // turn short — an empty usage-only message never erases real output.
+  const output: ContentBlock[] = finalAssistantOutput(own) ?? []
  const recorded = toStopReason(lastEnd?.data.reason)
  // Disposal can tear the owner down before the loop records its ordinary
  // `aborted` end, yielding `disposed` instead.
@@ -1,4 +1,4 @@
-import { createUserMessage } from '@deepseek-ai/dsh-llm'
+import { CallId, createUserMessage } from '@deepseek-ai/dsh-llm'
 import { describe, expect, it } from 'vitest'
 import { Context } from 'cordis'
 import { type Agent, type AgentOptions } from '@deepseek-ai/dsh-agent'
@@ -10,7 +10,8 @@ import * as SessionInvariant from '@deepseek-ai/dsh-session/invariant'
 import * as AgentInvariant from '@deepseek-ai/dsh-agent/invariant'
 import * as AgentLoopInvariant from '@deepseek-ai/dsh-agent-loop/invariant'
 import SubagentService, { snapshotSubagentDescriptor } from '@deepseek-ai/dsh-subagent'
-import { maxTokensResponse, MockAdapter, textResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
+import { defineContentToolFixture } from '@deepseek-ai/dsh-tools'
+import { maxTokensResponse, MockAdapter, textResponse, toolCallResponse } from '../../../core/agent-loop/tests/mock-adapter.ts'
 import { startInProcessRun } from '../src/index.ts'

 type Script = ConstructorParameters<typeof MockAdapter>[0]
@@ -155,6 +156,33 @@ describe('startInProcessRun', () => {
    await run.dispose()
  })

+  it('keeps earlier streamed text when the final step appends an empty usage-only message', async () => {
+    // Step 1 streams "partial one" plus a tool call; step 2 hits max-tokens
+    // having assembled only a tool-call block, so the loop appends an EMPTY
+    // assistant/message to host usage. The empty message is not assistant
+    // output and must not erase step 1's text from the run's output.
+    const { ctx, parent } = await setup([
+      toolCallResponse('t1', 'noop', {}, 'partial one'),
+      [
+        { type: 'block-start', index: 0, blockType: 'tool-call' },
+        { type: 'tool-call-delta', index: 0, id: CallId('t2'), name: 'noop', argumentsDelta: '{}' },
+        { type: 'block-end', index: 0, block: { type: 'tool-call', id: CallId('t2'), name: 'noop', arguments: '{}' } },
+        { type: 'usage', usage: { inputTokens: 20, outputTokens: 5 } },
+        { type: 'finish', reason: { kind: 'max-tokens' } },
+      ],
+    ])
+    const disposeNoop = ctx.tools.register(defineContentToolFixture({
+      name: 'noop', description: 'probe', parameters: {},
+      execute() { return Promise.resolve([{ type: 'text', text: 'noop result' }]) },
+    }))
+    const run = await startInProcessRun(request(parent), {})
+    const result = await run.result
+    expect(result.stopReason).toBe('max-tokens')
+    expect(text(result.output)).toBe('partial one')
+    await run.dispose()
+    disposeNoop()
+  })
+
  it('seeds a forked child but reads only the child-owned output', async () => {
    const { ctx, parent } = await setup([textResponse('parent answer'), textResponse('child answer')])
    parent.followup(createUserMessage({ content: [{ type: 'text', text: 'parent question' }], source: { kind: 'user' } }))
@@ -278,7 +306,12 @@ describe('startInProcessRun', () => {
    const signalled = await startInProcessRun(request(parent, controller.signal), {})
    await new Promise(resolve => setTimeout(resolve, 30))
    controller.abort('stop child')
-    await expect(signalled.result).resolves.toMatchObject({ stopReason: 'aborted' })
+    // No step completed a message, so the text streamed before the abort is
+    // the cancelled run's output.
+    await expect(signalled.result).resolves.toEqual({
+      output: [{ type: 'text', text: 'partial' }],
+      stopReason: 'aborted',
+    })
    expect(adapter.requests[0]?.signal?.reason).toEqual({ kind: 'parent' })
    const child = parent.ctx.agents.get(signalled.id)
    const turnEnd = child?.session.events.findLast(event => event.type === 'turn/end')