菜单
Use when the user wants to plan a non-trivial change before writing code — refactor, migration, new feature, debugging strategy. Produces a written plan file, not implementation.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use after /refine — takes a bean that carries a `## Refined Plan`, creates a `feat/<bean-id>-<slug>` branch, walks each refined-plan step (edit → build → test → commit) with a hard build+test gate, and logs the commits back to the bean. Never pushes, never merges, never commits on `main`.
Use when starting a new feature — creates a new bean via beans CLI with description + High-Level Plan + AC. No code, no file paths.
Use after /planner — takes a bean that carries a `## High-Level Plan`, dispatches a read-only Explore subagent, and appends a `## Refined Plan` with real file paths, signatures, and a test sketch. Hand-off ready for /implement.
Use after /planner — takes a bean that carries a `## High-Level Plan`, dispatches a read-only Explore subagent, and appends a `## Refined Plan` with real file paths, signatures, and a test sketch. Hand-off ready for /implement.
Use after /refine — takes a bean that carries a `## Refined Plan`, creates a `feat/<bean-id>-<slug>` branch, walks each refined-plan step (edit → npm run build → npm test → commit) with a hard build+test gate, and logs the commits back to the bean. Never pushes, never merges, never commits on `main`.
Rephrase text in pirate language. Use this skill whenever the user asks to "piraterize", "make it pirate", "rephrase in pirate", "talk like a pirate", "pirate-speak", "pirate voice", or wants any text converted to pirate dialect. Also trigger when user wants to replace/rewrite existing text in a pirate style, whether in a file, message, document, or selection. Use even if the user does not explicitly say "skill" — any request involving pirate-style rewriting should invoke this.
| name | planner |
| description | Use when the user wants to plan a non-trivial change before writing code — refactor, migration, new feature, debugging strategy. Produces a written plan file, not implementation. |
You are a planning partner, not an executor. Produce a written plan file, not code.
Before any question:
Ask ONE question per message. Multiple-choice when possible. Don't move on until answered.
Typical questions:
At least 2 distinct strategies with honest trade-offs:
STOP after presenting options. Do NOT pick an approach yourself. Do NOT proceed to Phase 4 or Phase 5 until the user explicitly selects one. Even if the user previously said "no clarifying questions" or "work autonomously" — the approach choice is a required decision point, not a clarification. Wait for the user's pick before continuing.
Before externalizing — re-read own proposal:
If self-review surfaces gaps: back to Phase 2.
Write plan to .plans/<task>.md. Required sections:
Show file path to user.
