auto-trigger

自动触发钩子

此技能定义技能之间的触发意图。当某个技能完成其工作流时，宿主运行时或 Agent 可以读取此元数据、记录待处理的后续操作，并且仅当当前会话支持且安全时才执行下一个技能。

钩子定义

PRD 创建链

prd_complete:
  triggers:
    - skill: self-improving-agent
      mode: background
      condition: PRD file exists and is complete
    - skill: session-logger
      mode: auto
      context: "PRD created for {feature_name}"

prd_implemented:
  triggers:
    - skill: session-logger
      mode: auto
      context: "Implemented PRD: {feature_name}"

实现链

implementation_complete:
  triggers:
    - skill: code-reviewer
      mode: ask_first
      message: "Implementation complete. Run code review?"
    - skill: create-pr
      mode: ask_first
      condition: user_requested_submission

会话管理

session_start:
  auto_triggers:
    - skill: session-logger
      action: create_session_file

session_end:
  auto_triggers:
    - skill: session-logger
      action: update_session_file

技能中的钩子格式

若要为技能添加自动触发功能，请在其 front matter 中添加：

---
name: my-skill
description: Skill description
allowed-tools: Read, Write, Edit
hooks:
  before_start:
    - trigger: session-logger
      mode: auto
      context: "Start {skill_name}"
  after_complete:
    - trigger: self-improving-agent
      mode: background
    - trigger: session-logger
      mode: auto
  on_error:
    - trigger: self-improving-agent
      mode: background
---

实现指南

当技能完成其工作流时：

1. 检查其自身 front matter 中的 hooks（before_start、after_complete、on_error、on_progress）
2. 对于每个钩子：
   • 如果 mode: auto，仅在宿主运行时支持时记录或执行低风险的后续操作
   • 如果 mode: background，记录非阻塞的后续操作；不要在未告知的情况下修改持久文件
   • 如果 mode: ask_first，请先询问用户，再触发
   • 如果存在 condition:，请先检查
3. 传递上下文给触发的技能

集成示例

prd-planner 应添加

---
name: prd-planner
description: Creates PRDs using persistent file-based planning...
allowed-tools: Read, Write, Edit, Bash, Grep, Glob, AskUserQuestion, WebSearch
hooks:
  after_complete:
    - trigger: self-improving-agent
      mode: background
      context: "PRD created at {prd_file}"
    - trigger: session-logger
      mode: auto
      context: "PRD creation complete"
---

self-improving-agent 使用显式验证和应用关卡

---
name: self-improving-agent
description: Captures redacted candidates, validates them with evidence, and records only explicitly applied rules...
allowed-tools: Read, Write, Edit, Bash, Grep, Glob
---

其 Claude 失败钩子由 CLI 适配器安装，而非声明为可移植的技能元数据。除非宿主适配器明确实现，否则此技能中的 hooks 示例仅表示声明性意图。其他宿主应使用已文档化的适配器或手动记录。

create-pr 应添加

---
name: create-pr
description: Creates pull requests with bilingual documentation updates...
allowed-tools: Read, Write, Edit, Bash, Grep, AskUserQuestion
hooks:
  after_complete:
    - trigger: session-logger
      mode: auto
      context: "PR created: {pr_title}"
---

链路可视化

┌──────────────┐
│ prd-planner  │
└──────┬───────┘
       │ after_complete
       ├──→ self-improving-agent (background)
       │         └──→ create-pr (ask_first)
       │                  └──→ session-logger (auto)
       └──→ session-logger (auto)

错误纠正链

on_error:
  triggers:
    - skill: self-improving-agent
      mode: background
      context: "Error occurred in {skill_name}"
    - skill: session-logger
      mode: auto
      context: "Error captured for {skill_name}"

重要规则

1. 避免创建无限循环——确保链路会终止
2. 重大操作前先询问——对 PR、部署和持久文件更改使用 mode: ask_first
3. 后台任务——对非阻塞分析或提案产物使用 mode: background
4. 传递上下文——始终向触发的技能提供相关上下文