cc-devflow 4.5.8 → 4.5.10

package/docs/guides/getting-started.zh-CN.md

@@ -11,10 +11,10 @@ CC-DevFlow 现在有两条入口：
 - `cc-devflow init`：把整包 `.claude` 安装到你的项目里
 - `cc-devflow adapt`：生成 Codex、Cursor、Qwen、Antigravity 等平台产物
 
-核心工作流由 6 个可见 Skill 组成，复杂工作可选 `cc-review` 做深度 Review：
+核心工作流可以手动走 PDCA/IDCA Skill，也可以通过 PR harness Skill 自动推进：
 
 ```text
-cc-roadmap
+cc-roadmap -> cc-next -> cc-dev
 
 PDCA: cc-plan -> [cc-review] -> cc-do -> [cc-review] -> cc-check -> cc-act
 IDCA: cc-investigate -> [cc-review] -> cc-do -> [cc-review] -> cc-check -> cc-act
@@ -36,7 +36,7 @@ IDCA: cc-investigate -> [cc-review] -> cc-do -> [cc-review] -> cc-check -> cc-ac
 npx cc-devflow init --dir /path/to/your/project
 ```
 
-整包安装会带上 6 个核心 workflow skill、可选 `cc-review`，以及维护用的 `cc-spec-init` 和 `cc-simplify`。
+整包安装会带上 roadmap、next-work selection、autonomous dev、手动 PDCA/IDCA、可选 `cc-review`、PR review/landing，以及维护用的 `cc-spec-init` 和 `cc-simplify`。
 
 ### 单个 Skill 安装
 
@@ -87,9 +87,9 @@ find .codex/skills -mindepth 2 -maxdepth 2 -name SKILL.md | sort
 
 - `cc-roadmap` 产出可编辑真相 `devflow/roadmap.json`，再生成 `devflow/ROADMAP.md` 和 deprecated `devflow/BACKLOG.md`
 - `cc-spec-init` 产出 `devflow/specs/INDEX.md`、capability spec 和 `change-meta.json`
-- `cc-plan` 产出 `planning/design.md`、`planning/tasks.md`、`task-manifest.json` 和 `change-meta.json`
-- `cc-investigate` 产出 `planning/analysis.md`、`planning/tasks.md`、`task-manifest.json` 和 `change-meta.json`
-- `cc-review` 产出 `cc-review-plan.md`、`cc-review-ledger.jsonl`、`cc-review-report.md`、可选 `cc-review-agent-results.jsonl`，以及可选的结构化深度 Review findings
+- `cc-plan` 产出 `planning/tasks.md#Contract Summary`、CLI 生成的 `task-manifest.json` 和 `change-meta.json`
+- `cc-investigate` 产出 `planning/tasks.md#Root Cause Contract`、CLI 生成的 `task-manifest.json` 和 `change-meta.json`
+- `cc-review` 产出 `review-ledger.jsonl`、可选 `review-findings.json`，Markdown 报告只在需要时按需渲染
 - `cc-check` 产出 `report-card.json`
 - `cc-act` 只产出一个最终 handoff 文件：`handoff/pr-brief.md`、`handoff/resume-index.md` 或 `handoff/release-note.md`
 
@@ -97,7 +97,8 @@ durable truth 分两层：
 
 - `devflow/specs/`：capability 真相，保留 `INDEX.md` 与 `capabilities/*.md`
 - 新 change 目录必须命名为 `REQ-<number>-<description>`（需求）或 `FIX-<number>-<description>`（修复）；`REQ` 和 `FIX` 分别维护自己的递增编号，跨前缀同号不是冲突；并行工作树造成重复编号时，完整 change key 的描述负责区分业务内容，旧小写目录只作为历史兼容读取。
-- `devflow/changes/<change>/`：变更真相，保留 `change-state.json`、`change-meta.json`、planning 文档、`task-manifest.json`、可选 `team-state.json`、任务级 `checkpoint.json`、`report-card.json` 和唯一的最终 handoff 文件。
+- `devflow/changes/<change>/`：变更真相，保留 `change-meta.json`、`planning/tasks.md`、CLI 生成的 `task-manifest.json`、review ledger / findings 记录、debug / failed 的可选 CLI 日志、`report-card.json` 和唯一的最终 handoff 文件。不要生成任务级 `context.md`、`checkpoint.json` 或 AI 手写过程文件。
+- 历史 `planning/design.md`、`planning/analysis.md` 和 `cc-review-*.md` 是旧 change 的可读 fallback，不再是新默认写入。
 - worker prompt、journal、assignment、session log 统一放到 `devflow/workspaces/<change>/`，作为 ephemeral scratch。
 
 进入实现前，planning handoff 应该先把证据写实：
@@ -153,7 +154,7 @@ npx cc-devflow adapt --cwd /path/to/your/project --platform codex
 
 如果你的项目没有可选的 `.claude/commands/` 输入目录，这也是正常的；编译器仍然会生成 skills registry，并为 Codex 镜像正式分发 skill 集合。
 
-Codex 现在会把正式分发的 skill 从 `.claude/skills/<skill>/` 镜像到 `.codex/skills/<skill>/`。这套集合包含 6 个核心 workflow skill、可选 `cc-review` 和维护类 skill `cc-spec-init`、`cc-simplify`，并且镜像是纯增量的：项目里已有的自定义 Codex skill 不会被删除。
+Codex 现在会把正式分发的 skill 从 `.claude/skills/<skill>/` 镜像到 `.codex/skills/<skill>/`。这套集合包含公开 workflow skill 和维护类 skill `cc-spec-init`、`cc-simplify`，并且镜像是纯增量的：项目里已有的自定义 Codex skill 不会被删除。
 
 ### 保持 skill 和样例同步
 
@@ -172,6 +173,7 @@ npm run verify:publish
 - [CLI 与 Skill](../commands/README.zh-CN.md)
 - [工作流详解](./workflow-guide.md)
 - [最佳实践](./best-practices.md)
+- [最小 Artifact 合同](./minimize-artifacts.md)
 - [样例入口页](../examples/START-HERE.md)
 - [简版样例列表](../examples/README.md)
 - [项目 README](../../README.zh-CN.md)

package/docs/guides/minimize-artifacts.md

@@ -0,0 +1,123 @@
+# Minimized Workflow Artifacts
+
+This guide describes the default artifact contract for new cc-devflow changes.
+The goal is simple: keep durable workflow truth readable, small, and measurable.
+
+## Default Shape
+
+Each new change keeps durable truth under `devflow/changes/<change-key>/`.
+
+Default human-authored Markdown:
+
+- `planning/tasks.md`
+
+Default machine-owned records:
+
+- `change-meta.json`
+- `planning/task-manifest.json`
+- `review/review-ledger.jsonl`
+- `review/review-findings.json` when findings exist
+- `execution/tasks/<task-id>/checkpoint.json`
+- `review/report-card.json`
+- one final handoff file under `handoff/`
+
+Runtime scratch, worker prompts, journals, assignments, and session logs belong
+under `devflow/workspaces/<change-key>/`, not beside durable change truth.
+
+## Feature Plans
+
+Feature and scope changes use:
+
+- `planning/tasks.md#Contract Summary`
+- `planning/task-manifest.json`
+- `change-meta.json`
+
+`Contract Summary` owns the frozen human-readable plan: user story, non-negotiable
+constraints, decisions that must not be reopened, task slices, and verification
+expectations. The task manifest is generated or validated by CLI tooling and owns
+machine-readable task status.
+
+## Bug Investigations
+
+Bug, regression, and unexpected-behavior work uses:
+
+- `planning/tasks.md#Root Cause Contract`
+- `planning/task-manifest.json`
+- `change-meta.json`
+
+`Root Cause Contract` owns the symptom, reproduction evidence, confirmed cause,
+rejected near-causes, repair boundary, and task handoff. `cc-do` should implement
+from that frozen contract instead of reopening investigation during execution.
+
+## Review Records
+
+`cc-review` writes structured lifecycle events first:
+
+- `review/review-ledger.jsonl`
+- optional `review/review-findings.json`
+- optional rendered Markdown from `cc-devflow review render`
+
+Markdown review reports are for human reading when needed. They are not the
+default durable review source.
+
+Useful commands:
+
+```bash
+npx cc-devflow review start --change REQ-001 --change-key REQ-001-copy-invite-link --base-sha abc123 --head-sha def456
+npx cc-devflow review record-node --change REQ-001 --change-key REQ-001-copy-invite-link --review-id <review-id> --node-id R001 --target planning/tasks.md --status checked --coverage contract --evidence-ref "cmd:npm run verify"
+npx cc-devflow review add-finding --change REQ-001 --change-key REQ-001-copy-invite-link --review-id <review-id> --finding-id F001 --severity important --confidence 8 --display-tier blocking --fingerprint sha256:<hash> --scope "current change" --path planning/tasks.md --evidence "finding evidence" --recommendation "repair action" --route cc-do
+npx cc-devflow review close --change REQ-001 --change-key REQ-001-copy-invite-link --review-id <review-id> --status clean --blocking-count 0 --warning-count 0 --next cc-check
+npx cc-devflow review render --change REQ-001 --change-key REQ-001-copy-invite-link --review-id <review-id> --output review/review-report.md
+```
+
+## Legacy Fallback
+
+Older changes may still contain:
+
+- `planning/design.md`
+- `planning/analysis.md`
+- `review/cc-review-plan.md`
+- `review/cc-review-report.md`
+- `review/cc-review-agent-results.jsonl`
+
+Those files remain readable compatibility inputs. New changes should not write
+them by default. When migrating old work, fold feature-plan truth into
+`planning/tasks.md#Contract Summary` and bug-investigation truth into
+`planning/tasks.md#Root Cause Contract`.
+
+## Validation Gates
+
+Validate one change:
+
+```bash
+npx cc-devflow task-contract validate --change REQ-001 --change-key REQ-001-copy-invite-link
+```
+
+Validate the repository artifact contract:
+
+```bash
+npm run verify:artifacts
+```
+
+Measure the contract:
+
+```bash
+npm run benchmark:artifacts
+```
+
+The package-level verification command also includes artifact validation:
+
+```bash
+npm run verify
+```
+
+## Authoring Rule
+
+Before adding a durable file under `devflow/changes/<change-key>/`, answer:
+
+1. Which downstream skill reads it by default?
+2. Which state does it own that no existing artifact owns?
+3. Which command fails if it drifts?
+
+If those answers are unclear, keep the information in `planning/tasks.md`, a
+machine record, or ephemeral workspace scratch instead.

package/docs/guides/project-postmortem.md

@@ -0,0 +1,78 @@
+# Project Postmortem Contract
+
+cc-devflow treats project postmortems as a durable AI memory surface. They are not
+chat summaries. They are repo-owned evidence that future agents can search before
+planning, investigating, or executing work.
+
+## Storage Layout
+
+Project-level postmortems live under `devflow/postmortems/`:
+
+| Path | Owner | Purpose |
+| --- | --- | --- |
+| `INDEX.md` | `cc-act` | Progressive entry point, latest incidents, tags, and search hints |
+| `principles.md` | `cc-act` | Generalized lessons about recurring model, process, and engineering mistakes |
+| `incidents/<date>-<change-key>.md` | `cc-act` | Immutable-ish factual record for one incident, bug, or repeated AI failure |
+
+`cc-act` owns writes because it has verified closeout, Git state, review state, and
+ship facts. Earlier skills only read and project the relevant reminders into their
+own artifacts.
+
+## Progressive Disclosure
+
+- Default layer: `INDEX.md` gives tags, one-line lessons, severity, affected
+  surfaces, and links to deeper incident files.
+- Principle layer: `principles.md` gives reusable rules such as model failure
+  modes, domain-specific judgment traps, and required countermeasures.
+- Incident layer: `incidents/*.md` gives the detailed facts, Git evidence,
+  timeline, root cause, detection gap, repair, follow-ups, and search terms.
+
+Agents should start with keyword search over the default and principle layers, then
+open incident files only when the tags or failure class match the current task.
+
+## Required Incident Evidence
+
+Every incident file should include:
+
+- Symptom and impact.
+- Trigger and timeline.
+- Confirmed root cause and rejected near-causes.
+- Why the failure escaped planning, investigation, execution, review, or ship.
+- Git evidence: branch, base, head SHA, PR if any, relevant commits, review range,
+  and dirty-tree notes when they matter.
+- Verification evidence: commands, exit status, key output, and artifact paths.
+- Follow-up actions: root-cause fixes, detection improvements, and backlog items.
+- AI failure mode: model limitation, pattern-matching trap, missing evidence habit,
+  over-broad abstraction, fake compatibility, test-seam mistake, or other reusable
+  class.
+- Search terms future agents should use before repeating similar work.
+
+## Redaction Guard
+
+Postmortems are durable repo artifacts, so they must never preserve secrets,
+tokens, private customer data, personal machine paths, or raw private logs unless
+the repository already treats that exact artifact as public source truth.
+
+- Record the command, file path, commit, or artifact pointer that proves the fact.
+- Quote only the minimal output needed to prove the incident.
+- Replace sensitive values with `<redacted>` and add a short redaction summary.
+- If the only available proof is sensitive, cite the owner artifact and describe
+  the observed shape instead of copying the raw value.
+
+## Read Gates
+
+`cc-plan`, `cc-investigate`, and `cc-do` must run a quick local search before they
+freeze direction or touch code:
+
+```bash
+rg -n "<capability|module|error|failure-class|model-risk>" devflow/postmortems
+```
+
+If `devflow/postmortems/` does not exist, record `no-project-postmortems-yet`.
+If a match exists, load only `INDEX.md`, `principles.md`, and the one or two
+incident files most relevant to the current work.
+
+## State Ownership
+
+Postmortems do not own task status, roadmap progress, review verdicts, or spec sync
+state. They cite those stronger owners by path, commit, or command output.

package/lib/compiler/__tests__/skills-registry.test.js

@@ -159,9 +159,9 @@ describe('Skills Registry Generator', () => {
       expect(execute.writes).toEqual(
         expect.arrayContaining([
           expect.objectContaining({
-            path: 'devflow/changes/<change-key>/execution/tasks/<task-id>/checkpoint.json',
+            path: 'devflow/changes/<change-key>/execution/tasks/<task-id>/events.jsonl',
             durability: 'durable',
-            required: true
+            required: false
           })
         ])
       );

package/lib/skill-runtime/CLAUDE.md

@@ -4,7 +4,7 @@
 职责分组
 入口层: `cli.js` 负责命令分发，`index.js` 提供给测试和内部脚本的稳定聚合入口。
 基础层: `schemas.js`、`store.js`、`paths.js` 管住契约、持久化与路径规则，避免执行层重复造轮子。
-状态层: `artifacts.js`、`lifecycle.js`、`query.js`、`review.js`、`team-state.js` 维护运行时真相源与只读查询。
+状态层: `artifacts.js`、`lifecycle.js`、`query.js`、`workflow-context.js`、`review.js`、`team-state.js` 维护运行时真相源与只读查询。
 规划与交接: `planner.js`、`intent.js`、`delegation.js` 把任务解析、handoff 生成和 team/workspace 委派收口成统一语义。
 阶段操作: `operations/` 是唯一 stage 入口目录；具体阶段边界见 `operations/CLAUDE.md`。
 测试布局: `__tests__/` 紧贴模块放置单元、回归与集成测试；顶层 `test/` 不再承载 `skill-runtime` 私有测试。

package/lib/skill-runtime/__tests__/autopilot.test.js

@@ -9,13 +9,13 @@ const {
   getTaskManifestPath,
   getReportCardPath,
   getReleaseNotePath,
-  getRuntimeStatePath,
-  getCheckpointPath
+  getRuntimeStatePath
 } = require('../store');
 const {
   getIntentResumeIndexPath,
   getIntentPrBriefPath
 } = require('../artifacts');
+const { getChangePaths } = require('../paths');
 
 jest.setTimeout(20000);
 
@@ -41,6 +41,41 @@ function markManifestReviewsPassed(repoRoot, changeId) {
   fs.writeFileSync(manifestPath, `${JSON.stringify(manifest, null, 2)}\n`);
 }
 
+function writeCleanReviewLedger(repoRoot, changeId) {
+  const change = getChangePaths(repoRoot, changeId);
+  const ledgerPath = path.join(change.reviewDir, 'review-ledger.jsonl');
+  fs.mkdirSync(path.dirname(ledgerPath), { recursive: true });
+  fs.writeFileSync(ledgerPath, [
+    JSON.stringify({
+      schema: 'review-ledger.v2',
+      change: change.changeKey,
+      reviewId: 'RVW-20260512-001',
+      createdAt: '2026-05-12T00:00:00.000Z',
+      createdBy: 'cc-devflow-cli',
+      event: 'review-started',
+      mode: 'implementation',
+      scope: 'current-diff',
+      baseSha: 'abc123',
+      headSha: 'def456',
+      selectedNodes: [],
+      skippedNodes: [],
+      riskLanes: []
+    }),
+    JSON.stringify({
+      schema: 'review-ledger.v2',
+      change: change.changeKey,
+      reviewId: 'RVW-20260512-001',
+      createdAt: '2026-05-12T00:01:00.000Z',
+      createdBy: 'cc-devflow-cli',
+      event: 'review-closed',
+      status: 'clean',
+      blockingCount: 0,
+      warningCount: 0,
+      next: 'cc-check'
+    })
+  ].join('\n'));
+}
+
 describe('runAutopilot', () => {
   test('stops at the approval gate after planning without writing approval-phase handoff markdown', async () => {
     const repoRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'cc-devflow-autopilot-'));
@@ -81,7 +116,7 @@ describe('runAutopilot', () => {
     expect(fs.existsSync(getIntentResumeIndexPath(repoRoot, 'REQ-123'))).toBe(false);
   });
 
-  test('resumes after approval, executes delegated work, and prepares a PR from checkpoints', async () => {
+  test('resumes after approval, executes delegated work, and prepares a PR from task state', async () => {
     const repoRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'cc-devflow-autopilot-workers-'));
 
     writeJson(path.join(repoRoot, 'package.json'), {
@@ -123,17 +158,16 @@ describe('runAutopilot', () => {
     });
 
     const manifest = JSON.parse(fs.readFileSync(getTaskManifestPath(repoRoot, 'REQ-123'), 'utf8'));
-    const delegatedCheckpoint = JSON.parse(fs.readFileSync(getCheckpointPath(repoRoot, 'REQ-123', 'T002'), 'utf8'));
     const report = JSON.parse(fs.readFileSync(getReportCardPath(repoRoot, 'REQ-123'), 'utf8'));
 
     expect(firstRun.executed).toEqual(expect.arrayContaining(['delegate', 'worker-run', 'dispatch', 'verify']));
     expect(firstRun.currentStage).toBe('verify');
     expect(manifest.tasks.find((task) => task.id === 'T002').status).toBe('passed');
-    expect(delegatedCheckpoint.outputExcerpt).toContain('delegate-ok');
     expect(report.review.status).toBe('blocked');
     expect(fs.existsSync(getIntentPrBriefPath(repoRoot, 'REQ-123'))).toBe(false);
 
     markManifestReviewsPassed(repoRoot, 'REQ-123');
+    writeCleanReviewLedger(repoRoot, 'REQ-123');
 
     const secondRun = await runAutopilot({
       repoRoot,
@@ -146,7 +180,8 @@ describe('runAutopilot', () => {
 
     expect(secondRun.executed).toEqual(expect.arrayContaining(['verify', 'prepare-pr']));
     expect(secondRun.currentStage).toBe('prepare-pr');
-    expect(prBrief).toContain('execution/tasks/T002/checkpoint.json');
+    expect(prBrief).toContain('planning/task-manifest.json');
+    expect(prBrief).not.toContain('checkpoint.json');
   });
 
   test('runs release after prepare-pr when requested for an approved plan', async () => {
@@ -195,6 +230,7 @@ describe('runAutopilot', () => {
     expect(fs.existsSync(getReleaseNotePath(repoRoot, 'REQ-123'))).toBe(false);
 
     markManifestReviewsPassed(repoRoot, 'REQ-123');
+    writeCleanReviewLedger(repoRoot, 'REQ-123');
 
     const result = await runAutopilot({
       repoRoot,

package/lib/skill-runtime/__tests__/benchmark-artifacts.test.js

@@ -0,0 +1,165 @@
+/**
+ * [INPUT]: 依赖 scripts/benchmark-artifacts.js 导出的 runBenchmarkArtifacts 和临时 artifact fixture。
+ * [OUTPUT]: 验证 benchmark:artifacts 使用 ceil(len/4) 估算并报告 profile 阈值 savings。
+ * [POS]: REQ-003-minimize-workflow-artifacts T017 的 Red/Green 证据。
+ * [PROTOCOL]: 变更时更新此头部，然后检查 CLAUDE.md
+ */
+
+const fs = require('fs');
+const os = require('os');
+const path = require('path');
+const { spawnSync } = require('child_process');
+
+const { runBenchmarkArtifacts } = require('../../../scripts/benchmark-artifacts');
+
+const REPO_ROOT = path.resolve(__dirname, '../../..');
+const BENCHMARK_SCRIPT = path.join(REPO_ROOT, 'scripts', 'benchmark-artifacts.js');
+
+function writeText(filePath, text) {
+  fs.mkdirSync(path.dirname(filePath), { recursive: true });
+  fs.writeFileSync(filePath, text);
+}
+
+function writeJson(filePath, value) {
+  writeText(filePath, `${JSON.stringify(value, null, 2)}\n`);
+}
+
+function contractTasks({ changeKey, profile = 'standard', filler = '' }) {
+  return [
+    '# Tasks',
+    '',
+    '## Contract Summary',
+    '',
+    `Change: ${changeKey}`,
+    'Mode: plan',
+    `Profile: ${profile}`,
+    'Approval: approved',
+    '',
+    'Goal:',
+    '- Minimize workflow artifacts.',
+    '',
+    'Do Not Do:',
+    '- Do not change token estimator math.',
+    '',
+    'Approved Direction:',
+    '- Use tasks.md plus generated JSON records.',
+    '',
+    'Acceptance:',
+    '- Benchmark savings stay above threshold.',
+    '',
+    'Verification:',
+    '',
+    '```bash',
+    'npm run benchmark:artifacts',
+    '```',
+    '',
+    'Risk / Escalate If:',
+    '- Savings fall below profile threshold.',
+    '',
+    filler,
+    '## Phase 1',
+    '',
+    '- [ ] T001 benchmark minimized artifact surface',
+    '  Vertical slice: Slice 1',
+    ''
+  ].join('\n');
+}
+
+function seedLegacyBaseline(repoRoot, changeKey, size = 6000) {
+  const changeDir = path.join(repoRoot, 'devflow', 'changes', changeKey);
+  writeText(path.join(changeDir, 'planning', 'design.md'), `# Design\n\n${'d'.repeat(size)}\n`);
+  writeText(path.join(changeDir, 'planning', 'analysis.md'), `# Analysis\n\n${'a'.repeat(size / 2)}\n`);
+  writeText(path.join(changeDir, 'planning', 'tasks.md'), `# Tasks\n\n${'t'.repeat(size / 2)}\n`);
+  writeJson(path.join(changeDir, 'planning', 'task-manifest.json'), { changeId: changeKey, tasks: [] });
+  writeJson(path.join(changeDir, 'change-meta.json'), { changeId: changeKey, goal: ['legacy'] });
+  writeJson(path.join(changeDir, 'review', 'report-card.json'), { overall: 'pass' });
+}
+
+function seedMinimizedChange(repoRoot, changeKey, options = {}) {
+  const changeDir = path.join(repoRoot, 'devflow', 'changes', changeKey);
+  writeText(path.join(changeDir, 'planning', 'tasks.md'), contractTasks({ changeKey, ...options }));
+  writeJson(path.join(changeDir, 'planning', 'task-manifest.json'), {
+    changeId: changeKey,
+    metadata: { source: 'tasks.md', generatedBy: 'cc-devflow task-contract', planVersion: 1 },
+    tasks: []
+  });
+  writeJson(path.join(changeDir, 'change-meta.json'), {
+    changeId: changeKey,
+    _meta: { generatedBy: 'cc-devflow task-contract' }
+  });
+  writeJson(path.join(changeDir, 'review', 'review-ledger.jsonl'), { note: 'counted as text by benchmark' });
+  writeJson(path.join(changeDir, 'review', 'review-findings.json'), { findings: [] });
+  writeJson(path.join(changeDir, 'review', 'report-card.json'), { overall: 'pass' });
+}
+
+describe('benchmark:artifacts', () => {
+  let repoRoot;
+
+  beforeEach(() => {
+    repoRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'cc-devflow-benchmark-artifacts-'));
+    seedLegacyBaseline(repoRoot, 'REQ-001-legacy-baseline');
+    seedLegacyBaseline(repoRoot, 'REQ-002-legacy-baseline');
+  });
+
+  afterEach(() => {
+    fs.rmSync(repoRoot, { recursive: true, force: true });
+  });
+
+  test('reports standard savings >= 30% for REQ-003-example', () => {
+    seedMinimizedChange(repoRoot, 'REQ-003-example', { profile: 'standard' });
+
+    const result = runBenchmarkArtifacts(repoRoot);
+    const row = result.rows.find((item) => item.changeKey === 'REQ-003-example');
+
+    expect(result.code).toBe(0);
+    expect(row).toMatchObject({
+      profile: 'standard',
+      threshold_pct: 30,
+      correctness_pass: true
+    });
+    expect(row.savings_vs_baseline_pct).toBeGreaterThanOrEqual(30);
+  });
+
+  test('reports tiny savings >= 60% for tiny fixture', () => {
+    seedMinimizedChange(repoRoot, 'REQ-004-tiny-example', { profile: 'tiny' });
+
+    const result = runBenchmarkArtifacts(repoRoot);
+    const row = result.rows.find((item) => item.changeKey === 'REQ-004-tiny-example');
+
+    expect(result.code).toBe(0);
+    expect(row).toMatchObject({
+      profile: 'tiny',
+      threshold_pct: 60,
+      correctness_pass: true
+    });
+    expect(row.savings_vs_baseline_pct).toBeGreaterThanOrEqual(60);
+  });
+
+  test('exits 1 when savings are below the profile threshold', () => {
+    seedMinimizedChange(repoRoot, 'REQ-005-bloated-example', {
+      profile: 'standard',
+      filler: 'x'.repeat(20000)
+    });
+
+    const result = runBenchmarkArtifacts(repoRoot);
+
+    expect(result.code).toBe(1);
+    expect(result.rows[0]).toMatchObject({ correctness_pass: false });
+  });
+
+  test('CLI prints stdout JSON array', () => {
+    seedMinimizedChange(repoRoot, 'REQ-003-example', { profile: 'standard' });
+
+    const result = spawnSync(process.execPath, [BENCHMARK_SCRIPT, repoRoot], { encoding: 'utf8' });
+    const rows = JSON.parse(result.stdout);
+
+    expect(result.status).toBe(0);
+    expect(Array.isArray(rows)).toBe(true);
+    expect(rows[0]).toHaveProperty('savings_vs_baseline_pct');
+  });
+
+  test('package.json exposes npm run benchmark:artifacts', () => {
+    const pkg = JSON.parse(fs.readFileSync(path.join(REPO_ROOT, 'package.json'), 'utf8'));
+    expect(pkg.scripts['benchmark:artifacts']).toBe('node scripts/benchmark-artifacts.js');
+  });
+});

package/lib/skill-runtime/__tests__/cli-bootstrap.integration.test.js

@@ -217,9 +217,9 @@ describe('cc-devflow cli distribution bootstrap', () => {
     expect(codexDoSkill.data.writes).toEqual(
       expect.arrayContaining([
         expect.objectContaining({
-          path: 'devflow/changes/<change-key>/execution/tasks/<task-id>/checkpoint.json',
+          path: 'devflow/changes/<change-key>/execution/tasks/<task-id>/events.jsonl',
           durability: 'durable',
-          required: true
+          required: false
         })
       ])
     );

package/lib/skill-runtime/__tests__/dispatch.test.js

@@ -7,7 +7,6 @@ const { runResume } = require('../operations/resume');
 const {
   getRuntimeStatePath,
   getTaskManifestPath,
-  getCheckpointPath,
   getEventsPath
 } = require('../store');
 
@@ -76,7 +75,7 @@ describe('runDispatch', () => {
     expect(nextManifest.tasks[0].status).toBe('pending');
   });
 
-  test('rejects stale results when planVersion changes during task execution and records it in checkpoint', async () => {
+  test('rejects stale results when planVersion changes during task execution and records it in manifest and events', async () => {
     const repoRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'cc-devflow-dispatch-'));
     const manifestPath = getTaskManifestPath(repoRoot, 'REQ-123');
 
@@ -133,28 +132,25 @@ describe('runDispatch', () => {
     });
 
     const nextManifest = JSON.parse(fs.readFileSync(manifestPath, 'utf8'));
-    const checkpoint = JSON.parse(fs.readFileSync(getCheckpointPath(repoRoot, 'REQ-123', 'T001'), 'utf8'));
     const events = fs.readFileSync(getEventsPath(repoRoot, 'REQ-123', 'T001'), 'utf8');
 
     expect(result.success).toBe(false);
     expect(nextManifest.tasks[0].status).toBe('failed');
     expect(nextManifest.tasks[0].lastError).toContain('Stale result rejected');
-    expect(checkpoint.planVersion).toBe(1);
-    expect(checkpoint.error).toContain('Stale result rejected');
     expect(events).toContain('task_stale_rejected');
   });
 
-  test('restores unresolved work from the latest stable checkpoint on resume without creating handoff markdown', async () => {
+  test('restores unresolved work from the latest stable manifest state on resume without creating handoff markdown', async () => {
     const repoRoot = fs.mkdtempSync(path.join(os.tmpdir(), 'cc-devflow-resume-stable-'));
     const changeId = 'REQ-123';
     const manifestPath = getTaskManifestPath(repoRoot, changeId);
 
     writeJson(getRuntimeStatePath(repoRoot, changeId), {
       changeId,
-      changeKey: 'REQ-123-recover-from-stable-checkpoint',
-      slug: 'recover-from-stable-checkpoint',
+      changeKey: 'REQ-123-recover-from-stable-state',
+      slug: 'recover-from-stable-state',
       createdAt: '2026-04-09T01:00:00.000Z',
-      goal: 'Recover from stable checkpoint',
+      goal: 'Recover from stable state',
       status: 'in_progress',
       initializedAt: '2026-04-09T01:00:00.000Z',
       plannedAt: '2026-04-09T01:01:00.000Z',
@@ -169,13 +165,13 @@ describe('runDispatch', () => {
 
     writeJson(manifestPath, {
       changeId,
-      goal: 'Recover from stable checkpoint',
+      goal: 'Recover from stable state',
       createdAt: '2026-04-09T01:00:00.000Z',
       updatedAt: '2026-04-09T01:02:00.000Z',
       tasks: [
         {
           id: 'T001',
-          title: 'Stable checkpoint task',
+          title: 'Stable completed task',
           type: 'TEST',
           dependsOn: [],
           touches: ['src/a.ts'],
@@ -219,32 +215,6 @@ describe('runDispatch', () => {
       }
     });
 
-    writeJson(getCheckpointPath(repoRoot, changeId, 'T001'), {
-      changeId,
-      taskId: 'T001',
-      sessionId: 'stable-session',
-      planVersion: 1,
-      status: 'passed',
-      summary: 'Task passed after 1 attempt(s)',
-      error: '',
-      outputExcerpt: '',
-      timestamp: '2026-04-09T01:05:00.000Z',
-      attempt: 1
-    });
-
-    writeJson(getCheckpointPath(repoRoot, changeId, 'T002'), {
-      changeId,
-      taskId: 'T002',
-      sessionId: 'failed-session',
-      planVersion: 1,
-      status: 'failed',
-      summary: 'Task failed: Command failed',
-      error: 'Command failed',
-      outputExcerpt: 'Command failed',
-      timestamp: '2026-04-09T01:06:00.000Z',
-      attempt: 2
-    });
-
     const result = await runResume({
       repoRoot,
       changeId,
@@ -255,7 +225,7 @@ describe('runDispatch', () => {
     const nextManifest = JSON.parse(fs.readFileSync(manifestPath, 'utf8'));
 
     expect(result.success).toBe(true);
-    expect(result.restoredCheckpoint).toMatchObject({
+    expect(result.restoredState).toMatchObject({
       taskId: 'T001',
       status: 'passed'
     });