菜单
Systematic codebase exploration. Trigger this skill when you need to understand how a specific subsystem works, trace the downstream impact of a feature, find where a function is defined, or onboard to an entirely new area of the codebase.
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
Convert URLs or local files (PDF/DOCX/PPTX/HTML) to Markdown, with optional summarization. Single or batch processing. Uses markitdown for intelligent extraction, falls back to curl+pandoc. Use when fetching web pages as markdown, converting documents for analysis, or producing quick summaries.
Use `uv` instead of pip/python/venv. Run scripts with `uv run script.py`, add deps with `uv add`, use inline script metadata for standalone scripts.
Adversarial code reviewer. Use when you need a quality agent to critically review code, a diff, test output, or a pull request — looking for bugs, edge cases, correctness failures, lazy shortcuts, and pattern violations. Works standalone (against any code/diff/worktree) or with GitHub PR mechanics (inline comments, merge decision) when a PR number is provided. Triggers: "bart", "review this", "adversarial review", "code review", "review PR", "find bugs", "QA", "quality check".
Use this skill to orchestrate multiple parallel sub-agents (Ralph/Bart) across isolated git worktrees to burn down a TODO.md backlog. Trigger this when asked to implement multiple independent features simultaneously, dispatch a "wave of agents", or perform "multi-agent orchestration". This handles complex dependency analysis, model spread, and feedback accumulation, which standard dispatch skills cannot.
Launch Claude Code CLI as a headless sub-agent in a tmux pane, monitor its stream-json JSONL output, and poll for completion. Use when delegating a well-defined coding task (bug fix, feature, refactor) to a Claude agent subprocess with live observability. Triggers include "run claude agent", "delegate to claude", "claude sub-agent", or any request to run claude headlessly and monitor it.
Launch GitHub Copilot CLI as a headless sub-agent in a tmux pane, monitor its JSONL output stream, and poll for completion. Use when delegating a well-defined coding task (bug fix, feature, refactor) to a Copilot agent subprocess with live observability. Triggers include "run copilot agent", "delegate to copilot", "copilot sub-agent", or any request to run copilot headlessly and monitor it.
| name | codemap-explore |
| description | Systematic codebase exploration. Trigger this skill when you need to understand how a specific subsystem works, trace the downstream impact of a feature, find where a function is defined, or onboard to an entirely new area of the codebase. |
codemap . — project tree, file counts, top extensionscodemap --deps . — how packages and files connectcodemap --importers <entry-file> — what calls this?codemap --deps . — follow the import chain downstream| Question | Command |
|---|---|
| Where is X defined? | codemap --deps . → scan the output |
| Who uses X? | codemap --importers <file> |
| What's the architecture? | codemap --deps . → look at hubs |
| What changed recently? | codemap --diff |
| What was I working on? | Check working set in prompt-submit hook output |
.codemap/config.json has routing)Check codemap config show — routing subsystems define logical boundaries with associated docs and keywords. Read subsystem docs before tracing code.