CodeWiki — Product Knowledge Base Generator

Scan code, fetch product context, discover user flows, and generate a structured codewiki/ knowledge base.

Output templates are in this skill's templates/ directory. Read each template only when generating that specific file — do NOT load all templates at once.

HARD RULES

Phase 1 is not optional. Product context is the difference between a useful wiki and a mechanical code listing. Complete it before scanning. If user can't answer, use the code-first fallback.
Multi-repo workspace must be verified before scanning. If repos aren't in a shared parent folder, stop and help the user set it up.
Never scan all repos simultaneously. Use Scan → Write to _scratch/ → Carry 30-line summary → Next. This is how you survive large multi-repo projects.
Never skip a service in microservices. User flows that touch 5 services are useless if one is undocumented. Every service gets at least a lightweight scan.
Scan in priority order. Infrastructure/contracts → gateway → frontends → core business services → supporting services.
Report progress between repos. Tell the user what you found and which repo is next. Do not ask for permission between repos — just continue. Only pause on errors or contradictions.
On startup, check if codewiki/_scratch/ exists. If it does, offer to resume from previous scan data.
Always use codewiki/ (no dot prefix). Ensures visibility in file explorers and default git tracking.
Adaptive detection. If hardcoded grep patterns don't match anything for a component, read the entry point file and trace outward. Don't report "no endpoints found" without trying alternative approaches.
WebFetch failures. Report the specific error (401, 404, timeout) and continue. Never silently skip a URL.