A collection of reusable instruction modules ("skills") for AI coding agents — each one a self-contained directory with a SKILL.md (SKILL format) plus any supporting code or docs.
Each skill is harness- and LLM-agnostic. Drop the folder into your agent's skills directory (e.g. ~/.claude/skills/ for Claude Code, ~/.agents/skills/ for Codex, or the equivalent for your tool), or paste the body of SKILL.md into your agent's instruction surface (Cursor rules, Aider conventions, custom GPT system prompt, raw API system message, …). See each skill's README for the full per-tool install matrix.
| Skill | What it does |
|---|---|
mutation-analyzer |
Turns the agent into an investigative staff engineer for mutation-oriented code analysis — "if I change X, what breaks, where does it silently corrupt, and where does it loudly fail?" Persists structured, code-grounded knowledge files under a wiki/code-analysis/ directory. |
generate-tasks-json |
Bulk-creates well-structured GitHub issues from a single JSON file, plus the planning methodology (contract-as-seam decomposition, two-level hierarchy, one Feature + two contract sides by default) that the agent applies when drafting that JSON. |
Each skill ships with a generic evals/evals.json probing the rules it enforces — customize the placeholders to refer to real entities in your codebase and run with the eval harness of your choice.
Each skill is its own self-contained directory.
Start with each skill's README.md for the overview, then SKILL.md for the actual prompt body.
Two common modes per skill, regardless of harness:
- Native SKILL install (Claude Code, Codex, other SKILL-aware harnesses): copy the skill folder into the agent's skills directory — per-project (e.g.
<your-repo>/.claude/skills/<skill-name>/,<your-repo>/.agents/skills/<skill-name>/) or user-global (e.g.~/.claude/skills/<skill-name>/,~/.agents/skills/<skill-name>/). - Instruction-surface install (Cursor, Aider, Copilot, Gemini CLI, custom GPTs, raw API): paste the body of
SKILL.mdinto the agent's instruction surface — see each skill's README for the per-tool path.
Most SKILL-aware harnesses auto-load skills on session start; trigger phrases declared in the frontmatter route automatically. In tools without auto-trigger metadata, invoke the skill explicitly (each skill's README documents its trigger phrases and explicit mode prefixes where applicable).
Open an issue or PR. New skills welcome — see mutation-analyzer and generate-tasks-json for the file layout to imitate.
MIT. Each skill also carries its own LICENSE.