claude-skills A collection of Claude Code skills I use day-to-day.
claude-skills
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
A collection of Claude Code skills I use day-to-day. Open-sourced for anyone who wants to fork, learn from, or adapt them.
What is a skill?
A skill is a folder under ~/.claude/skills/ / containing a SKILL.md file with YAML frontmatter (name, description, optional argument-hint) followed by Markdown instructions. Claude Code surfaces the skill via / and loads its instructions when invoked. See the official skills docs for details.
Install
Clone into your Claude Code skills directory:
Or if ~/.claude/skills already exists with other skills, clone elsewhere and symlink the ones you want.
After install, restart Claude Code (or wait for the next session) and the skills will appear in the available-skills list.
Skills in this repo
| Skill | What it does |
|-------|--------------|
| /ship-it | Deliver a scoped task to its requested local, PR, merge or verified-production endpoint; includes a batch reference for coordinated tickets. |
| /close | Session closing checklist — Git/worktree audit, handoff persistence, cleanup, optional Munin sync, and one deterministic cross-agent report format. |
| /code-friction | Opt-in capture of a few concrete code-changeability observations during a coding task, with explicit privacy bounds and Close handoff. |
| /commit | Standardized git commit workflow: author verification, diff review, security check, conventional message, push. |
| /debate | Adversarial debate against a cross-model reviewer (Codex or Antigravity/agy) to stress-test a draft, plan, or design before finalizing. Multi-round critique with self-review and summary. Backend chosen via --model codex\|agy (defaults to codex). |
| /eli5 | Re-explain the previous assistant turn in plain language — no jargon, no unexplained acronyms, short sentences. |
| /issues | Check GitHub issues for the current repo: list, filter by label, view details, search. |
| /mimir-search | Search and cite source documents or slides in a configured local Mímir collection. |
| /review-pr-codex | Cross-model PR review — invokes Codex headless on the current branch diff, then reads findings and fixes or reports. |
| /security-review | Full security review: threat model, OWASP Top 10 + LLM Top 10 scan, finding validation, patch proposals. |
| /user-test-memory | Spawn three subagents (Opus, Sonnet, Haiku) in parallel to stress-test the Munin memory MCP and produce a synthesized UX report. |
Dependencies
Some skills shell out to external CLIs or MCP servers:
- **debate** — requires the OpenAI Codex CLI (for --model codex) and/or the Google Antigravity CLI agy (for --model agy), installed and authenticated.
- **review-pr-codex** — requires the OpenAI Codex CLI installed and authenticated.
- **user-test-memory** — requires the Munin memory MCP server registered with Claude Code.
- **issues** — requires the GitHub CLI (gh).
- **mimir-search** — requires a separately installed and configured Mímir search CLI; its example invocation and tests use Python 3. The skill does not include credentials or host-specific configuration.