菜单
Use when reviewing architecture, planning refactors, reducing coupling, improving testability, or deciding whether modules should become deeper behind simpler interfaces.
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 when writing technical articles, blog posts, X/Twitter threads, LinkedIn posts, documentation, runbooks, or any text that must sound like a specific person, not a generic assistant. Also use when the user asks to humanize text, remove AI-sounding language, rewrite in their voice, or convert rough notes into publishable content. Trigger even when the user just says "write this up" or "draft a post about X" without explicitly mentioning voice or tone.
Use when the user asks for caveman mode, terse mode, fewer tokens, less prose, or very brief technical communication.
Use for hard bugs, failing checks, pipeline failures, homelab incidents, performance regressions, or reports that something is broken and needs root-cause analysis.
Use before git mutation, branch deletion, resets, cleaning, force pushes, broad restore commands, PR operations, or any git action with destructive or remote side effects.
Use when a plan, design, or implementation request has ambiguity that could change scope, architecture, acceptance criteria, safety, or validation.
Use when work should continue in another session, another lane, or a fresh context window, especially after planning, debugging, or partial implementation.
| name | architecture-deepening |
| description | Use when reviewing architecture, planning refactors, reducing coupling, improving testability, or deciding whether modules should become deeper behind simpler interfaces. |
Prefer deep modules: simple interface, useful behavior behind it, clear test surface.
Return candidates with files, problem, smallest safe improvement, test impact, risk, and confidence. Do not implement broad refactors unless an accepted plan authorizes them.
