Thanks for your interest in improving this skill. Contributions are welcome — bug fixes, new examples, domain coverage, and behavioural improvements are all appreciated.
Examples are the most valuable thing you can add. Good examples:
- Cover a real domain with real URLs (no placeholder links)
- Show the full exchange: clarifying questions → user answers → verified response with citations
- Represent an edge case not already covered (verification failure, phrasing ambiguity, referential follow-up, etc.)
- Use primary sources (official docs, datasheets, RFCs, GitHub) rather than aggregators or blog posts
If diligence mode behaves incorrectly — answers without clarifying when it should ask, fails to cite something it should, or gets caveman compression wrong — please open an issue with:
- The exact question asked
- What Claude responded
- What you expected instead
The skill currently covers embedded firmware, electronics, Python, C/C++, GUI frameworks, Linux/Windows desktop, and kernel development. Other technical domains are welcome as long as they include worked examples with real citations.
- Fork the repository
- Create a branch:
git checkout -b my-improvement - Make your changes to
SKILL.md - Verify your citation URLs are live and point to the right section
- Open a pull request with a short description of what you changed and why
- Examples must show the full exchange (question → clarifying questions if needed → answer with citations)
- Every verifiable technical claim in an example must have a real, working citation
- Caveman pair examples must have both a diligence-only and a diligence+caveman version of the same exchange
- Keep prose in examples consistent with the existing tone: direct, no hedging, no filler
- Do not add examples that use placeholder URLs like
example.com— find the real source
After editing SKILL.md, install the skill in Claude.ai and run through a few of your new example questions manually to verify the behaviour matches what the examples show.
Open an issue if you're unsure whether a contribution fits or want feedback before writing a full example.