convention-extractionlisted

# Convention Extraction Most team conventions live in people's heads — pattern recognition built from years of reviews, production incidents, and architectural discussions. They transfer slowly through pairing and code review, and walk out the door when someone leaves. AI amplifies this: without explicit conventions, AI output quality varies by who prompts. Same codebase, same AI, completely different quality gates. This skill guides systematic extraction of tacit knowledge into versioned, enforceable artefacts. The approach is informed by Rahul Garg's "Encoding Team Standards" (2026), which frames inconsistent AI output as a systems problem requiring a systems solution. This skill does not cover convention enforcement (see constraint-design and verification-slots), convention maintenance (see context-engineering and garbage-collection), or CI pipeline configuration. For the full interview protocol with worked examples, consult `references/extraction-interview-guide.md`. ## When to Extract | Situation | Signal | | ----------- | -------- | | New project setup | CLAUDE.md and HARNESS.md are empty or boilerplate | | Onboarding AI to existing codebase | AI-generated code keeps violating unwritten rules | | After team composition changes | New members or departures change the tacit knowledge base | | Quality variance | AI output quality depends on who is prompting | | Post-incident | A production incident reveals conventions that were implicit | **Sizing heuristic:** Teams