Engineering skill pack for AI coding agents. Battle-tested patterns distilled from real-world software engineering practices.
Disciplined 6-phase debugging loop. Build feedback loop → reproduce → hypothesise → instrument → fix → post-mortem. Build the right loop and the bug is 90% fixed.
Relentless interviewer that stress-tests plans and designs. Provides recommended answers, explores codebase instead of asking when possible.
Issue triage state machine. Categorize, prioritize, generate durable agent briefs. No file paths or line numbers that go stale.
Break plans into vertical slices. HITL vs AFK classification. Creates GitHub issues with dependency ordering.
Ultra-compressed mode. ~75% fewer tokens. Auto-reverts for safety.
cp -r skills/* ~/.hermes/skills/ln -sfn $(pwd)/skills/diagnose ~/.claude/skills/diagnose
ln -sfn $(pwd)/skills/grill-me ~/.claude/skills/grill-me
ln -sfn $(pwd)/skills/triage ~/.claude/skills/triage
ln -sfn $(pwd)/skills/to-issues ~/.claude/skills/to-issues
ln -sfn $(pwd)/skills/caveman ~/.claude/skills/caveman- Vertical slices over horizontal layers
- Deep modules (Ousterhout) — small interface, deep implementation
- HITL vs AFK — separate human work from agent work
- Durable outputs — no file paths or line numbers
- Feedback loops first
MIT