Install
$ agentstack add skill-jerrylalala-compound-engineering-ce-work-integration ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
ce:work 集成 Overlay — Task Bundle + FSM
> 基础层:上游 ce:work 负责任务执行、质量检查、提交工作流。 > > 本 Overlay 增量:Task Bundle 持久化(state.md 读写)+ Failure FSM 状态转换。
何时使用本 Overlay
当满足以下条件时,在运行 ce:work 前/后应用本 overlay:
- 任务有对应的 Task Bundle:
docs/tasks//目录存在 - 或者任务预期超过 1 小时(值得持久化上下文)
- 或者任务涉及多个子代理
Phase 0: 任务恢复(新增,在 ce:work Phase 1 之前)
当重新进入一个任务时,在执行 ce:work Phase 1 之前:
1. 检查 docs/tasks//state.md 是否存在
2. 如果存在,读取:
- current_step / total_steps
- status (active/blocked/debugging/replanned/resumed)
- blocked_by(阻塞原因)
- last_checkpoint 时间戳
3. 从 current_step 继续,不重新执行已完成步骤
4. 更新 state.md: status → active, updated_at → 当前时间
如果 state.md 不存在(新任务),从 plugins/compound-engineering/skills-custom/task-bundle/templates/state.md.tpl 创建:
cp plugins/compound-engineering/skills-custom/task-bundle/templates/state.md.tpl \
docs/tasks//state.md
# 填写 task_id、total_steps、started_at
FSM 状态转换(替换 ce:work 的隐式 STOP 处理)
原 ce:work 行为(隐式)
ce:work 遇到阻塞时会停下来询问用户,没有状态记录。
本 Overlay 增强(显式状态机)
active → blocked: 遇到任何阻塞(依赖缺失、CI 失败、需求不清)
blocked → debugging: 开始排查阻塞原因
debugging → replanned: 需要修改原计划
debugging → active: 阻塞自行解除(如 CI 通过)
replanned → resumed: 新计划就绪,等待用户确认重启(⚠️ 必须等用户确认)
resumed → active: 用户确认后,按新计划重新开始(AI 执行)
active → reviewed: 实现完成,进入 ce:review
reviewed → compounded: 经验沉淀完成
任何状态 → cancelled: 任务被主动取消
replanned → resumed 用户确认门(强制):
当状态进入 resumed 时,必须展示新计划摘要并等待用户确认,禁止自动进入 active:
📋 计划已更新,重启前需要你确认:
原阻塞原因:[blocked_by 值]
新计划摘要:[更新后的关键步骤]
确认后将从步骤 [N] 重新开始。继续吗?(y/n)
用户确认后才将 state.md 的 status 更新为 active。
每次状态转换写入 state.md:
## 状态历史
| 时间 | 从 | 到 | 原因 |
|------|-----|-----|------|
| 2026-04-08 10:00 | created | active | 开始执行 |
| 2026-04-08 14:30 | active | blocked | 等待 upstream merge 完成 |
阻塞触发条件(替换隐式 STOP)
以下情况触发 active → blocked,写入 state.md 并告知用户:
| 触发条件 | blocked_by 值 | |---------|--------------| | 依赖任务未完成 | depends_on: | | CI/测试持续失败(≥3 次) | ci_failure: | | 需求不清无法继续 | unclear_requirement: | | Rate limit / 工具不可用 | tool_unavailable: | | 需要人工决策 | human_decision_required: |
Phase 1 增强: state.md 写入时机
在 ce:work 原有 Phase 1-3 的基础上,在以下时机写入/更新 state.md:
| 时机 | state.md 更新内容 | |------|-----------------| | 开始执行 | status: active, currentstep: 1 | | 每个任务完成 | currentstep +1, lastcheckpoint | | 遇到阻塞 | status: blocked, blockedby | | 开始排查 | status: debugging | | 提交代码 | 记录 commit hash(可选)| | 完成实现 | status: reviewed(等待 ce:review)|
Phase 4 增强: compounded 终态
ce:work Phase 4(完成阶段)后,如果运行了 ce:compound:
- 将 state.md 的 status 更新为
compounded - 在 docs/tasks//review.md 记录经验沉淀结果
- 可选:将任务目录归档到
docs/tasks/archive/
Ralph-lite Repair Loop(有界微循环)
当任务有明确 verifier(测试/lint/build)且遇到小错误时,在触发 STOP FSM 之前:
Execute → Verify → Fail → Ralph-lite(≤2次) → Verify →
├── 通过 → 继续
└── 仍失败 → 触发 active → blocked
Ralph-lite 规则:
- 最多修复 2 次循环
- 每次修复必须改变假设或策略,不能机械重试
- 在 state.md 中记录失败原因和修复依据
Task Bundle 文件结构参考
docs/tasks//
├── plan.md # 目标、约束、验收标准(链接到 docs/plans/)
├── state.md # 当前步骤、阻塞点、状态历史(本 overlay 写入)
└── review.md # findings、证据、结论(ce:review 写入)
模板位置:plugins/compound-engineering/skills-custom/task-bundle/templates/
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: Jerrylalala
- Source: Jerrylalala/compound-engineering
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.