← ClaudeAtlas

docdnalisted

Backfill a project's documentation from its codebase. Decides which documents the repo actually owes and which it does not, flags docs that now contradict the code, and writes the ones the code can prove, with citations. Use for documentation backfill, missing or stale docs, "what docs does this project need", handover prep, audit readiness, drifted README or API docs, ADRs, or a doc inventory.
hannsxpeter/docdna · ★ 0 · DevOps & Infrastructure · score 73
Install: claude install-skill hannsxpeter/docdna
# DocDNA Version: 1.0.1 ## What this is Documentation lies, not by malice but by drift, and the documents that would survive a handover were never written because nobody could say which ones this project owes. docdna answers three questions from the code: **which documents does this project owe, which of the ones you have are now false, and what can I write without asking anyone.** It is a selection engine and a drift detector first, a generator third. **The one rule, and everything else follows from it:** nothing is asserted that the repository cannot prove. Every claim carries a citation or a GAP marker. There is no third state. ## Three modes Read the user's intent and pick one. - **Survey** (default): scan, detect drift, decide the document set, write the manifest. Writes no documents. Triggered by "what docs does this need", "survey the documentation", "docdna". - **Backfill**: write selected documents from code evidence. Triggered by "backfill the docs", "write the config reference", "document this repo". - **Check**: drift, lint, GAP rollup, stale exclusions. The CI gate. Triggered by "check the docs against the code", "are our docs still true". **Routing, three rules in order:** 1. **A request naming a single document goes straight to Backfill for that document.** "Write the threat model for this repo" is maximum intent and must not be met with a questionnaire. Write the manifest row as a side effect. 2. **Survey requires nothing.** No prior run,