← ClaudeAtlas

bootstrap-glossarylisted

Seed GLOSSARY.md with the domain terms the code already uses, each term carrying the file it was read from and marked unreviewed until the owner confirms it. Invoke by name at bootstrap on an existing codebase, instead of a fifty-question domain interview. Extracts vocabulary from code; it does not invent definitions.
thientranhung/astragentic · ★ 4 · AI & Automation · score 70
Install: claude install-skill thientranhung/astragentic
# Bootstrap the glossary from the terms the code uses Upstream's warning is the whole design constraint here: *an unreviewed, agent-authored glossary is worse than none — it becomes confident-sounding lore that later sessions treat as truth.* So this skill seeds a glossary it explicitly marks as unconfirmed, with the evidence attached to every entry, and hands the owner a short review rather than a long interview. **Extract, never invent.** Every term comes from the code. Every definition is derived from how the code uses the term, and says so. ## 1. Harvest the vocabulary Read the names the codebase has already committed to, in roughly this order of authority: 1. **Type, class, and table names** — the nouns the system is built out of. 2. **Module and directory names** — the areas the domain is divided into. 3. **Function names on public surfaces** — the verbs the domain performs. 4. **Enum and constant values** — the states and categories the domain distinguishes. 5. **Recurring words in comments and commit messages** — where a term is explained in prose. Rank by frequency and by position: a term appearing in a type name, a table name and a module name is core vocabulary. A term appearing once in a private helper is not. ## 2. Derive each definition from usage For each term, write what the code shows it to be — its shape, what it relates to, what changes it. Then attach the evidence: **Write the entry in the reader's format, not a format of our own.** `GLOSSARY.md`