/wiki — read before answering

When this skill is invoked OR when the user is working in a repo that has a wiki built, read the wiki BEFORE answering anything substantial about that repo.

Where the wiki lives

Path: ~/.yashigatakae/state/codebase-wiki/<basename>/ where <basename> is the last path segment of the cwd.

If ~/.yashigatakae/state/codebase-wiki/<basename>/index.md does not exist, run:

yashigatakae graphify "$(pwd)" --pro

This creates the wiki under that directory in 1–2 seconds.

How to read it

1. Read `index.md` first. It contains the infobox (HEAD sha, file count, primary lang, module count, public symbol count, dep count, ADR count) and a navigation TOC.
2. For architectural questions, read architecture.md next (component map + per-module table).
3. For "what does X do / who calls Y" questions, read symbols/<X>.md — every public symbol gets a page with signature, doc, and "referenced by" backlinks.
4. For "why was decision Z made" questions, read DECISIONS.md — auto-extracted from substantive git commits.
5. For "what's in module M", read modules/<M>.md — infobox + grouped symbol list.
6. For domain term lookups, check GLOSSARY.md — proper nouns appearing >3x in docs.
7. For broken refs (something the wiki mentions but doesn't define), check STUB-PAGES.md — those are the gaps to flag.

Wikilink syntax

The wiki uses [[Foo]] markdown wikilinks. They render as relative markdown links. If you see STUB-PAGES.md#foo linked, the page is a stub — the concept exists but isn't documented yet.

Refresh

If the wiki feels stale, re-run yashigatakae graphify <repo> --pro to regenerate. Incremental refresh and CI integration land in v0.13.1.

Why this exists

Without the wiki, every Claude session in a non-trivial repo spends ~10K tokens grepping and reading source to build a mental map. With the wiki, those 10K tokens are pre-computed once and read from a few markdown files. Saves time, saves tokens, keeps Claude grounded in the same shared model the user has.