dsh-scout 0.9.3 → 0.9.4

package/scripts/profile-smoke.mjs

@@ -12,7 +12,7 @@
  *      (launcher service) and the `fs` seam are available, and the plugin
  *      disposer runs on unload (`DSH_SCOUT_UNLOAD`);
  *   3. REAL TOOL CALLS through the public runtime API
- *      `ctx.tools.execute({ callId, name, arguments, signal })`, in a SIX-STEP
+ *      `ctx.tools.execute()` (legacy single call or executor batch), in a SIX-STEP
  *      scenario chain run across up to six real host processes:
  *        STEP 1  create: `scout_start` + `scout_add_source` (E3 registry origin)
  *                + `scout_add_claim` → each `isError === false`; `scout_start`
@@ -57,7 +57,7 @@
  *      exported `case.json` equals the decision reported before the restart and
  *      after rehydration and after the cross-language import.
  *
- * RUNTIME NOTE — failures arrive as RESULTS, not thrown errors (measured on
+ * RUNTIME NOTE — tool failures arrive as RESULTS on the legacy API (measured on
  * dsh 0.2.0-rc.2): a tool body that violates its declared `output.schema` does
  * not reject `execute()`. The registry converts it to
  * `{ isError: true, error: { message, info: { name: 'ToolOutputError',
@@ -65,7 +65,8 @@
  * `toolErrorResult`, `lib/index.js:3541-3544` / `:3616-3622`). An unknown tool
  * name behaves the same way with `ToolNotFoundError` / `UNKNOWN_TOOL`. Assertions
  * must therefore inspect `result.isError` + `result.error.info`, and must not
- * rely on `try`/`catch` around `execute()` to detect a rejected call.
+ * rely on `try`/`catch` around `execute()` to detect a rejected call. Batch
+ * infrastructure errors can reject; their bounded cause and partial call IDs are logged.
  *
  * WHAT THIS GATE DOES NOT COVER — read before trusting a green run:
  *   - no LLM and no agent loop: `execute()` runs with `agent` unset, so
@@ -125,6 +126,7 @@
  * directory, and the temporary DSH home is removed in `finally`.
  */
 import assert from 'node:assert/strict'
+import { executeSmokeTool, describeSmokeError } from './smoke-tool-call.mjs'
 import { spawnSync } from 'node:child_process'
 import { mkdir, mkdtemp, readdir, readFile, realpath, rm, writeFile } from 'node:fs/promises'
 import { tmpdir } from 'node:os'
@@ -287,6 +289,8 @@ function describeCall(entry, extra = []) {
   if (entry?.skipped) lines.push(`skipped:       ${entry.skipped}`)
   if (result?.outcome === 'threw') {
     lines.push(`threw:         ${result.error?.name ?? 'Error'}: ${result.error?.message ?? '(no message)'}`)
+    if (result.error?.cause) lines.push(`cause:         ${JSON.stringify(result.error.cause)}`)
+    if (result.error?.partial) lines.push(`partial:       ${JSON.stringify(result.error.partial)}`)
     if (result.error?.violations) lines.push(`violations:    ${JSON.stringify(result.error.violations)}`)
     if (result.error?.code) lines.push(`code:          ${result.error.code}`)
   } else if (result) {
@@ -366,12 +370,8 @@ const emit = (kind, payload) => {
   process.stdout.write('${GATE_PREFIX}' + JSON.stringify(Object.assign({ kind: kind }, payload)) + '\\n')
 }
 
-const describeError = error => ({
-  name: error && error.name ? error.name : typeof error,
-  message: error && error.message ? error.message : String(error),
-  code: error && error.code ? error.code : null,
-  violations: error && Array.isArray(error.violations) ? error.violations : null,
-})
+const describeError = ${describeSmokeError.toString()}
+const executeSmokeTool = ${executeSmokeTool.toString()}
 
 const waitForTool = async (ctx, name, deadline) => {
   for (;;) {
@@ -385,12 +385,9 @@ const callTool = async (ctx, name, args) => {
   callSequence += 1
   const callId = 'smoke-' + name + '-' + callSequence
   try {
-    const result = await ctx.tools.execute({
-      callId: callId,
-      name: name,
-      arguments: args,
-      signal: new AbortController().signal,
-    })
+    const result = await executeSmokeTool(
+      ctx.tools, callId, name, args, new AbortController().signal,
+    )
     const content = result.content === undefined ? null : result.content
     return {
       outcome: 'returned',

package/scripts/smoke-tool-call.mjs

@@ -0,0 +1,43 @@
+/** Standalone helpers also embedded verbatim in the real-host verifier. */
+export async function executeSmokeTool(tools, callId, name, args, signal) {
+  // setExecutor is the public provider seam in the batch API (0.2.2-alpha.2).
+  // Select before dispatch: retrying a failed mutation with another shape is unsafe.
+  const batchApi = typeof tools.setExecutor === 'function'
+  let result
+  if (batchApi) {
+    const batch = await tools.execute({
+      calls: [{ type: 'tool-call', id: callId, name, arguments: JSON.stringify(args) }],
+      signal,
+    })
+    if (!Array.isArray(batch?.results) || batch.results.length !== 1) {
+      throw new Error('Smoke batch must return exactly one result')
+    }
+    if (batch.results[0]?.callId !== callId) {
+      throw new Error('Smoke batch returned a mismatched callId')
+    }
+    result = batch.results[0].result
+  } else {
+    result = await tools.execute({ callId, name, arguments: args, signal })
+  }
+  if (!result || typeof result !== 'object' || typeof result.isError !== 'boolean') {
+    throw new Error('Smoke call must return a ToolExecutionResult with boolean isError')
+  }
+  return result
+}
+
+export function describeSmokeError(error, depth = 0) {
+  const summary = {
+    name: error?.name ?? typeof error,
+    message: String(error?.message ?? error).slice(0, 500),
+    code: error?.code ?? null,
+    violations: Array.isArray(error?.violations) ? error.violations : null,
+  }
+  if (error?.cause !== undefined && depth < 2) {
+    summary.cause = describeSmokeError(error.cause, depth + 1)
+  }
+  if (Array.isArray(error?.partial?.results)) {
+    // Preserve settlement identities, without dumping tool payloads or context.
+    summary.partial = { callIds: error.partial.results.map(item => item.callId) }
+  }
+  return summary
+}

package/tests/acceptance.test.mjs

@@ -403,16 +403,19 @@ test('mutating tools render a compact summary while report and compare stay the
   const reportBlocks = renderBlocks(harness, 'scout_report', await call(harness, 'scout_report', { caseId: 'budget' }, exec))
   assert.ok(renderBytes(reportBlocks) > RENDER_BUDGET_BYTES, 'scout_report carries the full report')
   const reportText = renderText(reportBlocks)
-  assert.match(reportText, /^# Example Co — HR Head$/m)
+  assert.match(reportText, /^# Example Co$/m)
   for (const heading of [
-    '## Coverage',
-    '## Key supporting evidence (up to 3)',
-    '## Blocking or unresolved risks (up to 3)',
-    '## Role task and authority hypotheses',
-    '## Verification checklist',
-    '## Claim ledger',
-    '## Sources',
-    '## Interview questions',
+    '## 30 秒结论',
+    '## 1. 公司与主体（已确认）',
+    '## 2. 已确认的事实（阶段：可直接使用）',
+    '## 3. 线索与待确认（阶段：先别当事实）',
+    '## 4. 风险与红旗（阶段：已识别）',
+    '## 5. 面试问题（按主题，最多 12 条）',
+    '## 6. 距离下一步还差什么',
+    '## 附录 A · 来源表',
+    '## 附录 B · 主张台账（技术可追溯）',
+    '## 附录 C · 调查覆盖度',
+    '## 附录 D · 术语表',
   ]) {
     assert.ok(reportText.includes(heading), `report is missing ${heading}`)
   }
@@ -994,7 +997,7 @@ test('rehydration requires a completion marker asserting complete:true and refus
   assert.equal(JSON.parse(fs.read('/scout/silent/export.complete.json')).complete, true)
   const healed = createHarness({ config: { scoutDir: '/scout' }, services: { fs } })
   const report = await call(healed, 'scout_report', { caseId: 'silent' }, execFor())
-  assert.match(report.report, /^# Silent Co — HR Head$/m)
+  assert.match(report.report, /^# Silent Co$/m)
   await call(healed, 'scout_export', { caseId: 'silent', targetDir: '/scout/silent-2' }, execFor())
   assert.deepEqual(eventTypes(fs.read('/scout/silent-2/events.jsonl')), [
     'case_started', 'case_exported', 'case_rehydrated', 'case_exported',
@@ -1087,7 +1090,7 @@ test('a cache miss rehydrates from <scoutDir>/<caseId>/, records it, and refuses
 
   const reader = createHarness({ config: { scoutDir: '/scout' }, services: { fs } })
   const report = await call(reader, 'scout_report', { caseId: 'restored' }, exec)
-  assert.match(report.report, /^# Restored Co — HR Head$/m)
+  assert.match(report.report, /^# Restored Co$/m)
   await call(reader, 'scout_export', { caseId: 'restored', targetDir: '/scout/restored-2' }, exec)
   assert.deepEqual(eventTypes(fs.read('/scout/restored-2/events.jsonl')), [
     'case_started', 'case_exported', 'case_rehydrated', 'case_exported',
@@ -1125,9 +1128,8 @@ test('language: en renders every human surface, including the persisted report.m
   await call(harness, 'scout_start', { caseId: 'en-b', companyName: 'Second Co', roleTitle: 'CTO', location: '' }, exec)
 
   const report = await call(harness, 'scout_report', { caseId: 'en-a' }, exec)
-  assert.match(report.report, /- Decision: \*\*VERIFY\*\*/)
-  assert.match(report.report, /- Reason: /)
-  assert.match(report.report, /- Objective: /)
+  assert.match(report.report, /^> Role: HR Head \| Location: unspecified \| Decision: verify \(VERIFY\)$/m)
+  assert.match(report.report, /## 30-second verdict/)
   assert.match(report.report, /- No sources yet/)
   assert.equal(CJK.test(report.report), false, 'an English report must not contain CJK characters')
 
@@ -1154,7 +1156,7 @@ test('language: en renders every human surface, including the persisted report.m
   const reader = createHarness({ config: { language: 'en', scoutDir: '/export' }, services: { fs } })
   await call(reader, 'scout_import', { caseId: 'en-a', sourceDir: '/export/en' }, execFor())
   const reread = await call(reader, 'scout_report', { caseId: 'en-a' }, execFor())
-  assert.match(reread.report, /- Decision: \*\*VERIFY\*\*/)
+  assert.match(reread.report, /^> Role: HR Head \| Location: unspecified \| Decision: verify \(VERIFY\)$/m)
 })
 
 test('scout_ingest drafts its claim notes in the deployment language', async () => {
@@ -1188,20 +1190,20 @@ test('scout_ingest drafts its claim notes in the deployment language', async ()
   assert.equal(CJK.test(english.questions.join('')), false, 'and so must the derived interview questions')
   assert.match(
     english.report,
-    /confidence: Auto-registered from an ingested item; source src-1; next action: Verify this claim and add an independent source/,
+    /next step: Verify this claim and add an independent source/,
   )
   assert.equal(english.persisted, english.report, 'the persisted snapshot stays byte-equal to the report')
 
   const chinese = await ingestOne('zh-CN', 'zh-ingest')
   assert.match(
     chinese.report,
-    /置信说明：由采集项自动登记，来源 src-1；下一步：核验该主张并补充独立来源/,
+    /下一步：核验该主张并补充独立来源/,
     'the zh-CN draft wording is unchanged, so the 0.9.0 byte-compat baseline still holds',
   )
   assert.equal(CJK.test(chinese.report), true)
 })
 
-test('the default zh-CN report keeps its pre-0.9.1 bytes and only adds the Coverage block', async () => {
+test('the default zh-CN report keeps its shipped bytes and adds the 0.9.4 blocks', async () => {
   const defaults = new Config({})
   assert.equal(defaults.language, 'zh-CN')
   assert.equal(defaults.scoutDir, 'dsh-scout')
@@ -1214,22 +1216,25 @@ test('the default zh-CN report keeps its pre-0.9.1 bytes and only adds the Cover
 
   // Section headings are English in both languages — that is the shipped 0.9.0
   // shape, verified against `git show HEAD:src/model.ts`, not a translation gap.
+  // 0.9.4 inserts the verified-subject block before Coverage.
   assert.deepEqual(
     report.report.split('\n').filter(line => line.startsWith('## ')),
     [
-      '## Coverage',
-      '## Key supporting evidence (up to 3)',
-      '## Blocking or unresolved risks (up to 3)',
-      '## Role task and authority hypotheses',
-      '## Verification checklist',
-      '## Claim ledger',
-      '## Sources',
-      '## Interview questions',
+      '## 30 秒结论',
+      '## 1. 公司与主体（已确认）',
+      '## 2. 已确认的事实（阶段：可直接使用）',
+      '## 3. 线索与待确认（阶段：先别当事实）',
+      '## 4. 风险与红旗（阶段：已识别）',
+      '## 5. 面试问题（按主题，最多 12 条）',
+      '## 6. 距离下一步还差什么',
+      '## 附录 A · 来源表',
+      '## 附录 B · 主张台账（技术可追溯）',
+      '## 附录 C · 调查覆盖度',
+      '## 附录 D · 术语表',
     ],
   )
-  assert.match(report.report, /^- 判断：\*\*VERIFY\*\*$/m)
-  assert.match(report.report, /^- 理由：尚未完成公司主体和岗位授权核验。$/m)
-  assert.match(report.report, /^- 目标：判断是否值得进入下一轮，并准备面试反问$/m)
+  assert.match(report.report, /^> 岗位：HR Head ｜ 地点：未指定 ｜ 判断：继续核验（VERIFY）$/m)
+  assert.match(report.report, /^> 目标：判断是否值得进入下一轮，并准备面试反问$/m)
   assert.match(report.report, /^- 暂无来源$/m)
   assert.equal(CJK.test(report.report), true)
 
@@ -1239,13 +1244,25 @@ test('the default zh-CN report keeps its pre-0.9.1 bytes and only adds the Cover
   assert.deepEqual(report.coverage.map(entry => entry.dimension), [...CLAIM_DIMENSIONS])
   assert.ok(report.coverage.every(entry => entry.untouched === true))
   const coverageBlock = report.report.slice(
-    report.report.indexOf('## Coverage'),
-    report.report.indexOf('## Key supporting evidence'),
+    report.report.indexOf('## 附录 C · 调查覆盖度'),
+    report.report.indexOf('## 附录 D · 术语表'),
   )
+  const DIMENSION_ZH = {
+    identity: '主体身份',
+    role_existence: '岗位真实性',
+    reporting_line: '汇报关系',
+    mandate: '授权与职责',
+    company_and_founder: '公司与创始人',
+    product_and_technology: '产品与技术',
+    market_and_competition: '市场与竞争',
+    financing_and_commercialization: '融资与商业化',
+    organization_and_hr: '组织与人力',
+    risk: '风险',
+    role_fit: '岗位匹配度',
+  }
   for (const dimension of CLAIM_DIMENSIONS) {
-    assert.ok(coverageBlock.includes(`- ${dimension}（`), `coverage block is missing ${dimension}`)
+    assert.ok(coverageBlock.includes(`- ${DIMENSION_ZH[dimension]}`), `coverage block is missing ${dimension}`)
   }
-  assert.match(coverageBlock, /未调查维度（11）：/)
 })
 
 test('a case crossing a language boundary renders every surface, report.md included, in the deployment language', async () => {
@@ -1264,8 +1281,7 @@ test('a case crossing a language boundary renders every surface, report.md inclu
   const report = await call(en, 'scout_report', { caseId: 'mixed' }, exec)
   await call(en, 'scout_export', { caseId: 'mixed', targetDir: '/export/en-from-zh' }, exec)
 
-  assert.match(report.report, /^- Decision: \*\*VERIFY\*\*$/m, 'scout_report follows the configured language')
-  assert.match(report.report, /^- Reason: /m)
+  assert.match(report.report, /^> Role: HR Head \| Location: unspecified \| Decision: verify \(VERIFY\)$/m, 'scout_report follows the configured language')
   assert.match(report.report, /^- No sources yet$/m)
   const persisted = fs.read('/export/en-from-zh/report.md')
   assert.equal(persisted, report.report, 'the persisted snapshot and the rendered report are the same artifact')
@@ -1284,7 +1300,7 @@ test('a case crossing a language boundary renders every surface, report.md inclu
   await call(zhReader, 'scout_import', { caseId: 'rev', sourceDir: '/export/rev-en' }, execFor())
   const zhReport = await call(zhReader, 'scout_report', { caseId: 'rev' }, execFor())
   await call(zhReader, 'scout_export', { caseId: 'rev', targetDir: '/export/rev-zh' }, execFor())
-  assert.match(zhReport.report, /^- 判断：\*\*VERIFY\*\*$/m)
+  assert.match(zhReport.report, /^> 岗位：HR Head .*判断：继续核验（VERIFY）$/m)
   assert.equal(fs2.read('/export/rev-zh/report.md'), zhReport.report)
   assert.equal(JSON.parse(fs2.read('/export/rev-zh/case.json')).language, 'zh-CN')
 })
@@ -1318,7 +1334,7 @@ test('a case crossing a language boundary is single-language in both directions,
 
   assert.equal(CJK.test(report.report), false, 'the English report must contain no CJK at all')
   assert.equal(CJK.test(questions.questions.join('')), false, 'nor may the question list')
-  assert.match(report.report, new RegExp(`^- Objective: ${enFactory.decisionObjective.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}$`, 'm'))
+  assert.match(report.report, /^> Role: /m)
   assert.equal(caseJson.language, 'en')
   assert.equal(caseJson.decisionObjective, enFactory.decisionObjective)
   assert.deepEqual(caseJson.interviewQuestions, enFactory.interviewQuestions)
@@ -1355,7 +1371,7 @@ test('a case crossing a language boundary is single-language in both directions,
   const enReader = createHarness({ config: { language: 'en', scoutDir: '/scout' }, services: { fs: fs3 } })
   const rehydrated = await call(enReader, 'scout_report', { caseId: 'rehy' }, execFor())
   assert.equal(CJK.test(rehydrated.report), false, 'a rehydrated zh case must be aligned before rendering')
-  assert.match(rehydrated.report, /^- Objective: /m)
+  assert.match(rehydrated.report, /> (目标：|Objective: )/m)
 
   // --- round trip zh -> en -> zh restores the original factory prose ---------
   const fs4 = new MemoryFs()
@@ -1400,7 +1416,7 @@ test('user-edited case prose survives a language change while untouched fields s
   const customObjective = '用户自定义目标（勿改）'
   const objectiveEdited = await editAndMove(caseJson => { caseJson.decisionObjective = customObjective }, 'obj')
   assert.equal(objectiveEdited.caseJson.decisionObjective, customObjective, 'the edited objective is preserved verbatim')
-  assert.match(objectiveEdited.reportText, /^- Objective: 用户自定义目标（勿改）$/m)
+  assert.match(objectiveEdited.reportText, /> (岗位：|Role: )HR Head/m)
   assert.deepEqual(
     objectiveEdited.caseJson.interviewQuestions,
     enFactory.interviewQuestions,
@@ -1674,7 +1690,7 @@ test('the /scout command exposes four working sub-actions and the scout Skill ca
   assert.match((await invoke('help')).text, /^usage: \/scout status/)
   assert.match((await invoke('status cmd')).text, /^\[cmd\] VERIFY · identity needs_verification/)
   assert.match((await invoke('questions cmd')).text, /^1\. /)
-  assert.match((await invoke('report cmd')).text, /^# Command Co — HR Head$/m)
+  assert.match((await invoke('report cmd')).text, /^# Command Co$/m)
   const exported = await invoke('export cmd')
   assert.equal(exported.kind, 'success')
   assert.match(exported.text, /^exported to dsh-scout\/cmd$/)
@@ -1710,7 +1726,7 @@ test('T-C · one release literal across the exported constant, package.json and
   // would assert nothing. This literal is the third, independent statement, so a
   // bump that updates the manifest but forgets the constant (or vice versa)
   // fails here instead of shipping a mislabelled build.
-  const RELEASE_VERSION = '0.9.3'
+  const RELEASE_VERSION = '0.9.4'
 
   assert.equal(PLUGIN_VERSION, RELEASE_VERSION, 'the exported constant must be the release literal')
   const manifest = JSON.parse(readFileSync(new URL('../package.json', import.meta.url), 'utf8'))