doc-auditlisted

Audit and optimize a project's documentation, agent-context files, persistent memory, and showcase docs for accuracy and token efficiency — code↔doc drift, duplication, archiving, memory slimming, showcase currency, standing-instruction budgets. Use when asked to review/clean up docs, context, or memory — or as a periodic maintenance pass to keep them lean and accurate.
jrittelmeyer/ai-dev-kit · ★ 1 · Data & Documents · score 74

Install: claude install-skill jrittelmeyer/ai-dev-kit

# Documentation / Context / Memory Audit A periodic, sign-off-gated pass that keeps a project's docs, agent-context files, persistent memory, and showcase docs **accurate** and **token-efficient**. Use extended thinking — this is an analysis task before it's an editing task. ## Operating principles (the "why") - **Optimize the hot path first.** The most expensive tokens are in files the agent loads *every turn* (`CLAUDE.md`/`AGENTS.md`, the memory index) or *every resume* (a status/handoff file). Slimming those beats slimming an on-demand reference doc. Triage every file by load frequency: always-loaded → on-resume → on-demand → archival. - **Spend audit tokens the same way.** Read hot-path files fully; spot-check on-demand files against their drift-prone claims; review archives at index level only. Reuse the project's own doc map instead of re-deriving the landscape. Verify with the cheapest sufficient probe — grep for the symbol instead of reading the file; query the registry instead of a changelog. - **Accuracy beats completeness.** A doc that contradicts the code actively misleads the agent. Never trust a doc's claim about code; verify it against source. - **Preserve, don't destroy.** Cull verbose detail by *moving* it to an archive/history file and recording in the archive's index what was preserved where. The hot-path file keeps a compact one-line-per-item record. - **One source of truth.** If two files state the same fact, the hot-path file