Remove reviewer final prompt reinjection (#137)

* add gated codex goals support * sync README SDK versions * bump claude agent sdk * add codex goals settings toggle * reuse status dashboard message and simplify settings actions * refine settings page UX * fix settings nav hash routing * add dashboard UX verification * refine dashboard settings and inbox UX * remove inbox top-level navigation * remove dashboard health top-level navigation * fix: allow runtime image attachment paths * feat: configure attachment allowlist dirs * fix: clean duplicate dashboard status messages * fix: poll dashboard duplicate cleanup between status updates * fix: delete dashboard duplicates on create * Refine settings IA with tabbed sections * Add read-only runtime inventory settings * Fix runtime inventory MCP JSON parsing * Add room skill settings inventory * Add room skill setting mutations * Apply room skill overrides to runner spawn * Refine scheduled task dashboard UX * Remove reviewer final prompt reinjection * Stabilize pnpm verification fixture
2026-05-16 17:59:59 +08:00
parent 8c24082b19
commit 55592b34b2
5 changed files with 105 additions and 67 deletions
--- a/prompts/claude-paired-room.md
+++ b/prompts/claude-paired-room.md
@@ -42,4 +42,5 @@ If you see a materially better design, debugging path, or scoping choice, propos
 - Stagnation: **Spinning** (same error 3+), **Oscillation** (alternating approaches), **Diminishing returns** (shrinking improvement), **No progress** (discussion without change) — name the pattern and report: **Status**, **Attempted**, **Recommendation**
 - Implementation, commits, and pushes require agreement from both sides. Either can veto
 - Keep reviews concise — approve quickly when there is nothing to critique, and keep alternative proposals short and actionable
+- Do not carry over old ledgers. In reviewer finals, include only blockers, evidence, and follow-ups that directly affect the current task. Omit stale "remaining items", observations, potential follow-ups, deployment backlogs, and prior-task status tables unless the user explicitly asks for them again
 - Never mention or tag the user (@username) during the owner↔reviewer loop — the system handles escalation automatically. User is only notified when all resolution paths (including arbiter) are exhausted
--- a/src/message-runtime-prompts.test.ts
+++ b/src/message-runtime-prompts.test.ts
@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest';

 import {
+  buildFinalizePendingPrompt,
  buildOwnerPendingPrompt,
  buildPairedTurnPrompt,
  buildReviewerPendingPrompt,
@@ -23,12 +24,15 @@ function makeHumanMessage(content: string): NewMessage {
  };
 }

-function makeTurnOutput(outputText: string): PairedTurnOutput {
+function makeTurnOutput(
+  outputText: string,
+  role: PairedTurnOutput['role'] = 'owner',
+): PairedTurnOutput {
  return {
    id: 1,
    task_id: 'task-1',
    turn_number: 0,
-    role: 'owner',
+    role,
    output_text: outputText,
    created_at: '2026-04-20T00:59:00.000Z',
  };
@@ -111,7 +115,7 @@ describe('message-runtime-prompts carry-forward guidance', () => {
    ).toBe(false);
  });

-  it('includes previous owner and reviewer finals in reviewer pending prompts when the current task has no outputs', () => {
+  it('includes previous owner finals in reviewer pending prompts when the current task has no outputs', () => {
    const prompt = buildReviewerPendingPrompt({
      chatJid: 'group@test',
      timezone: 'UTC',
@@ -129,11 +133,11 @@ describe('message-runtime-prompts carry-forward guidance', () => {
    );
    expect(prompt).toContain('Previous task owner final:');
    expect(prompt).toContain('이전 owner 결론');
-    expect(prompt).toContain('Previous task reviewer final:');
-    expect(prompt).toContain('이전 reviewer 결론');
+    expect(prompt).not.toContain('Previous task reviewer final:');
+    expect(prompt).not.toContain('이전 reviewer 결론');
  });

-  it('includes previous owner and reviewer finals in owner pending prompts when the current task has no outputs', () => {
+  it('includes previous owner finals in owner pending prompts when the current task has no outputs', () => {
    const prompt = buildOwnerPendingPrompt({
      chatJid: 'group@test',
      timezone: 'UTC',
@@ -150,7 +154,67 @@ describe('message-runtime-prompts carry-forward guidance', () => {
      'Background from the previous completed paired task:',
    );
    expect(prompt).toContain('Previous task owner final:');
-    expect(prompt).toContain('Previous task reviewer final:');
+    expect(prompt).not.toContain('Previous task reviewer final:');
    expect(prompt).toContain('추가 owner 질문');
  });
+
+  it('omits previous reviewer finals from prior task context', () => {
+    const prompt = buildReviewerPendingPrompt({
+      chatJid: 'group@test',
+      timezone: 'UTC',
+      turnOutputs: [],
+      recentHumanMessages: [makeHumanMessage('새 작업')],
+      lastHumanMessage: '새 작업',
+      priorTaskContext: {
+        ownerFinal: null,
+        reviewerFinal: `TASK_DONE
+검증한 것:
+- CI pass
+
+남은 항목 (변동 없음):
+(관찰) 사용자 직접 functional 검증
+(잠재) 다른 enum 영어 라벨 추가 발견 시
+
+현재 시리즈 상태:
+#129 Settings IA
+#130 Runtime Inventory`,
+      },
+    });
+
+    expect(prompt).not.toContain('Previous task reviewer final:');
+    expect(prompt).not.toContain('검증한 것:');
+    expect(prompt).not.toContain('CI pass');
+    expect(prompt).not.toContain('남은 항목');
+    expect(prompt).not.toContain('(관찰)');
+    expect(prompt).not.toContain('(잠재)');
+    expect(prompt).not.toContain('현재 시리즈 상태');
+    expect(prompt).not.toContain('#129');
+  });
+
+  it('does not reinject reviewer output into finalize prompts', () => {
+    const prompt = buildFinalizePendingPrompt({
+      turnOutputs: [
+        makeTurnOutput(
+          `TASK_DONE
+현재 검증:
+- build pass
+
+다음 단계:
+- unrelated cleanup
+
+누적 배포 대기 상태:
+#131 pending`,
+          'reviewer',
+        ),
+      ],
+    });
+
+    expect(prompt).not.toContain("Reviewer's final assessment:");
+    expect(prompt).not.toContain('현재 검증:');
+    expect(prompt).not.toContain('build pass');
+    expect(prompt).not.toContain('다음 단계');
+    expect(prompt).not.toContain('unrelated cleanup');
+    expect(prompt).not.toContain('누적 배포 대기 상태');
+    expect(prompt).not.toContain('#131');
+  });
 });
--- a/src/message-runtime-prompts.ts
+++ b/src/message-runtime-prompts.ts
@@ -77,11 +77,6 @@ function formatPriorTaskPromptContext(
      `Previous task owner final:\n---\n${truncatePriorTaskFinal(priorTaskContext.ownerFinal)}\n---`,
    );
  }
-  if (priorTaskContext.reviewerFinal?.trim()) {
-    sections.push(
-      `Previous task reviewer final:\n---\n${truncatePriorTaskFinal(priorTaskContext.reviewerFinal)}\n---`,
-    );
-  }

  if (sections.length === 0) {
    return '';
@@ -202,18 +197,11 @@ export function buildArbiterPromptForTask(args: {
  });
 }

-export function buildFinalizePendingPrompt(args: {
+export function buildFinalizePendingPrompt(_args: {
  turnOutputs: PairedTurnOutput[];
 }): string {
-  const lastReviewerOutput = [...args.turnOutputs]
-    .reverse()
-    .find((output) => output.role === 'reviewer');
-  const reviewerSummary = lastReviewerOutput?.output_text
-    ? `\n\nReviewer's final assessment:\n${lastReviewerOutput.output_text.slice(0, 2000)}`
-    : '';
-
  return `The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.
 If you intend to close this paired turn now, your first line must be TASK_DONE.
 If the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.
-If your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.${reviewerSummary}`;
+If your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.`;
 }
--- a/src/message-runtime.test.ts
+++ b/src/message-runtime.test.ts
@@ -1646,12 +1646,11 @@ describe('createMessageRuntime', () => {
    );
  });

-  it('includes the latest reviewer summary in merge_ready finalize prompts and truncates it', async () => {
+  it('uses a compact finalize prompt without reinjecting reviewer output', async () => {
    const chatJid = 'group@test';
    const group = makeGroup('codex');
    const channel = makeChannel(chatJid);
    const longReviewerOutput = `검토 요약 ${'a'.repeat(2105)} 끝`;
-    const truncatedReviewerOutput = longReviewerOutput.slice(0, 2000);

    vi.mocked(serviceRouting.hasReviewerLease).mockReturnValue(true);
    vi.mocked(db.getLatestOpenPairedTaskForChat).mockReturnValue({
@@ -1701,11 +1700,12 @@ describe('createMessageRuntime', () => {
    vi.mocked(agentRunner.runAgentProcess).mockImplementation(
      async (_group, input, _onProcess, onOutput) => {
        expect(input.prompt).toBe(
-          `The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.
-If you intend to close this paired turn now, your first line must be TASK_DONE.
-If the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.
-If your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.\n\nReviewer's final assessment:\n${truncatedReviewerOutput}`,
+          'The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.\n' +
+            'If you intend to close this paired turn now, your first line must be TASK_DONE.\n' +
+            'If the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.\n' +
+            'If your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.',
        );
+        expect(input.prompt).not.toContain("Reviewer's final assessment:");
        expect(input.prompt).not.toContain(longReviewerOutput);
        expect(input.prompt).not.toContain('이전 reviewer 요약');
        await onOutput?.({
@@ -1811,8 +1811,9 @@ If your first line is DONE_WITH_CONCERNS, the system will reopen review instead
    vi.mocked(agentRunner.runAgentProcess).mockImplementation(
      async (_group, input, _onProcess, onOutput) => {
        expect(input.prompt).toBe(
-          "The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.\nIf you intend to close this paired turn now, your first line must be TASK_DONE.\nIf the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.\nIf your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.\n\nReviewer's final assessment:\n리뷰 승인 요약",
+          'The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.\nIf you intend to close this paired turn now, your first line must be TASK_DONE.\nIf the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.\nIf your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.',
        );
+        expect(input.prompt).not.toContain('리뷰 승인 요약');
        expect(input.prompt).not.toContain('DONE\n승인합니다.');
        await onOutput?.({
          status: 'success',
@@ -1993,7 +1994,7 @@ If your first line is DONE_WITH_CONCERNS, the system will reopen review instead
    });
  });

-  it('includes previous task owner and reviewer finals in a new reviewer prompt when the current task has no outputs', () => {
+  it('includes previous task owner final but omits reviewer final in a new reviewer prompt when the current task has no outputs', () => {
    const chatJid = 'group@test';
    const group = makeGroup('claude-code');
    const currentTask = {
@@ -2070,8 +2071,8 @@ If your first line is DONE_WITH_CONCERNS, the system will reopen review instead
    );
    expect(pending?.prompt).toContain('Previous task owner final:');
    expect(pending?.prompt).toContain('이전 owner 답변');
-    expect(pending?.prompt).toContain('Previous task reviewer final:');
-    expect(pending?.prompt).toContain('이전 reviewer 피드백');
+    expect(pending?.prompt).not.toContain('Previous task reviewer final:');
+    expect(pending?.prompt).not.toContain('이전 reviewer 피드백');
  });

  it('does not build an owner follow-up pending turn from owner STEP_DONE on an active task', () => {
--- a/src/verification.test.ts
+++ b/src/verification.test.ts
@@ -11,6 +11,7 @@ import {
  runVerificationRequest,
 } from './verification.js';
 import {
+  buildWorkspaceCommandEnvironment,
  ensureWorkspaceDependenciesInstalled,
  hasInstalledNodeModules,
 } from './workspace-package-manager.js';
@@ -106,7 +107,7 @@ describe('verification helpers', () => {
    });
  });

-  it('verifies nested pnpm workspaces under a bun parent without tripping corepack project specs', async () => {
+  it('builds nested pnpm verification commands without tripping corepack project specs', () => {
    const parentDir = fs.mkdtempSync(
      path.join(os.tmpdir(), 'ejclaw-verification-corepack-parent-'),
    );
@@ -122,48 +123,31 @@ describe('verification helpers', () => {
      JSON.stringify({
        name: 'nested-pnpm-workspace',
        scripts: {
-          typecheck:
-            'node -e "process.stdout.write(process.env.COREPACK_ENABLE_PROJECT_SPEC || \'missing\')"',
+          typecheck: 'tsc --noEmit',
        },
      }),
    );
    fs.writeFileSync(
      path.join(repoDir, 'pnpm-lock.yaml'),
-      [
-        "lockfileVersion: '6.0'",
-        'settings:',
-        '  autoInstallPeers: true',
-        '  excludeLinksFromLockfile: false',
-        'importers:',
-        '  .: {}',
-        '',
-      ].join('\n'),
+      'lockfileVersion: 9.0\n',
    );
-    fs.mkdirSync(path.join(repoDir, 'node_modules', '.bin'), {
-      recursive: true,
+    const command = buildVerificationCommand('typecheck', repoDir);
+
+    expect(command).toMatchObject({
+      file: 'corepack',
+      args: ['pnpm', 'run', 'typecheck'],
+      commandText: 'corepack pnpm run typecheck',
+      packageManager: 'pnpm',
+      requiredScript: 'typecheck',
+    });
+    expect(
+      buildWorkspaceCommandEnvironment(repoDir, command.packageManager, {
+        PATH: '/tmp/bin',
+      }),
+    ).toMatchObject({
+      PATH: '/tmp/bin',
+      COREPACK_ENABLE_PROJECT_SPEC: '0',
    });
-    fs.writeFileSync(
-      path.join(repoDir, 'node_modules', '.bin', 'placeholder'),
-      '',
-    );
-
-    expect(hasInstalledNodeModules(repoDir)).toBe(false);
-
-    const expectedSnapshotId = computeVerificationSnapshot(repoDir).snapshotId;
-    const result = await runVerificationRequest(
-      {
-        requestId: 'req-corepack-parent-pnpm',
-        profile: 'typecheck',
-        expectedSnapshotId,
-      },
-      { repoDir },
-    );
-
-    expect(result.ok).toBe(true);
-    expect(result.exitCode).toBe(0);
-    expect(result.command).toBe('corepack pnpm run typecheck');
-    expect(result.stdout).toContain('0');
-    expect(result.snapshotId).toBe(expectedSnapshotId);
  });

  it('computes a stable snapshot over the readable workspace inputs', () => {