Agent skill
cunzhi
寸止协议 - Pause and wait mechanism using cunzhi MCP for human confirmation at critical decision points. Prevents AI from proceeding autonomously past important checkpoints.
Install this agent skill to your Project
npx add-skill https://github.com/WenJunDuan/Rlues/tree/main/vibeCoding/old version/claude/config-agent_v7.9.1/.claude/skills/cunzhi
SKILL.md
寸止 (Cunzhi) Protocol
Concept
寸止 = "Stop at the inch" - Pause execution at critical points for human review.
Uses cunzhi MCP for interactive confirmation.
Cunzhi Points
| Point | When | Purpose |
|---|---|---|
[REQ_READY] |
需求分析完成 | 确认理解正确 |
[DESIGN_READY] |
方案设计完成 | 确认技术方案 |
[PLAN_READY] |
计划制定完成 | 确认执行计划 |
[PHASE_DONE] |
阶段执行完成 | 确认继续下一阶段 |
[RELEASE_READY] |
准备发布 | 确认发布 |
[TASK_DONE] |
任务完成 | 确认归档 |
MCP 调用方式
确认请求
mcp__cunzhi__confirm({
point: "PLAN_READY",
title: "计划确认",
content: `
## 执行计划摘要
1. 创建 auth 模块
2. 实现 JWT 验证
3. 添加单元测试
预计时间: 2小时
`,
options: [
{ id: "continue", label: "继续执行", default: true },
{ id: "modify", label: "修改计划" },
{ id: "cancel", label: "取消任务" }
]
})
处理响应
switch (response.choice) {
case "continue":
// 继续执行下一阶段
break;
case "modify":
// 回到规划阶段
break;
case "cancel":
// 终止任务,保存状态
break;
}
Fallback (无 MCP)
当 cunzhi MCP 不可用时,使用文本确认:
---
🛑 **[PLAN_READY]** 寸止等待确认
## 计划摘要
[内容]
请输入选择:
- `继续` 或 `c` - 执行计划
- `修改` 或 `m` - 调整后重新规划
- `取消` 或 `x` - 终止任务
---
Rules
- 必须等待 - 不可跳过寸止点
- 明确输出 - 清晰展示等待内容
- 选项明确 - 提供清晰的下一步选项
- 状态保存 - 等待前保存当前状态
集成场景
需求审查
await cunzhi.confirm({
point: "REQ_READY",
content: formatRequirement(req),
options: ["确认理解正确", "需要澄清", "重新描述"]
});
方案审查
await cunzhi.confirm({
point: "DESIGN_READY",
content: formatDesign(design),
options: ["批准方案", "调整设计", "换方向"]
});
发布确认
await cunzhi.confirm({
point: "RELEASE_READY",
content: formatReleaseNotes(release),
options: ["确认发布", "延迟发布", "取消"]
});
与 RIPER 集成
| RIPER Phase | Cunzhi Point | Trigger |
|---|---|---|
| Research → Innovate | REQ_READY | 需求分析完成 |
| Plan → Execute | PLAN_READY | 计划制定完成 |
| Execute → Review | PHASE_DONE | 每阶段完成 |
| Review → Done | TASK_DONE | 任务完成 |
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
agent-teams
Path D 并行编排 — 子代理分工 + Agent Teams 协作。需要 CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
tdd
Path 分级 TDD 策略
context7
E 阶段按需拉取库文档 — 用 ctx7 获取准确 API 信息
knowledge
跨会话经验管理 — 在 .ai_state 内沉淀和复用项目经验
code-quality
Rev 阶段 Plugin 编排顺序
verification
V 阶段 Path 分级验证清单
Didn't find tool you were looking for?