CONTRIBUTING.md

Contributing

Small repo, one skill. Contributions welcome — new modes, new tones, bug fixes in skills/coach/SKILL.md, or better docs.

Ground rules for SKILL.md changes

• Explain the why, not just the rule. A bare Never do X teaches worse than Never do X — because Y. Every rule in the file should carry its reasoning inline, not rely on the mode's Core Directive to imply it.
• Keep the description pushy. Claude under-triggers skills by default. List concrete trigger phrases, and say "even if they don't say X" where relevant.
• Don't add a mode without a clear axis. hardcore/standard/architect sit on a real spectrum (how much code the AI is allowed to write). A new mode should differ on that axis or a clearly new one — not just be a tone variant (those go in the Tone section instead).
• Any path referenced in the file must actually resolve. If you copy this file elsewhere (a different tool's skills directory), double check relative paths (../../learning/...) still point at the right place from the new depth — this has broken before.
• Keep it under 500 lines. If it's getting long, move detail into a references/ file loaded on demand, not inline.

Testing a change

No automated eval harness yet. Run the relevant cases in skills/coach/test-cases.md in a real session and check the behavior actually matches "Expected" — don't just re-read the prose. Add a case there for any new rule you add.

Where things live

• skills/coach/SKILL.md — the skill itself, single source of truth (see README's Install section for how it gets copied into a specific tool's skills directory — don't add duplicate copies to this repo).
• skills/coach/test-cases.md — manual fixture checklist: prompt + expected behavior per rule. Run these against a change before calling it done.
• learning/ — sample output location for the Learning Log, not something you need to populate.