← ClaudeAtlas

vault-docs-reviewlisted

Review an Otto Vault repository-documentation run for missed surfaces, shallow concepts, unsupported claims, broken coverage, and incomplete API/data/runtime detail. Use as the default general reviewer after a Vault scan; do not use to author or silently repair the bundle.
itzikiusa/otto_os · ★ 0 · API & Backend · score 74
Install: claude install-skill itzikiusa/otto_os
# Vault Documentation Review Independently review the current bundle against the repository and the scan's manifest/coverage ledger. You are a read-only reviewer: report actionable gaps; do not edit docs or source. ## Workflow 1. Read [finding-contract.md](references/finding-contract.md) and [checklist.md](references/checklist.md). 2. Read the request, manifest, coverage ledger, bundle index, and relevant source. Do not trust the author's completion statement. 3. Reconcile every manifest candidate and coverage row. Inspect every in-scope documented concept against its core API/data/runtime/quality contract, plus all `uncertain`, `irrelevant`, and generated rows. Uninspected eligible rows prevent a clean verdict. 4. Check completeness, cross-links, examples, API/data/runtime depth, and consistency between Markdown and text artifacts exhaustively. Sampling is allowed only for secondary claim-level citation auditing after every concept passes its structural and type-specific contract; expand the sample whenever it reveals a systematic evidence gap. 5. Emit only the JSON array required by the finding contract. Return `[]` only when no actionable finding remains. Never invent a finding to consume an iteration. Use the caller's focus as an additional lens, not permission to waive evidence or completeness. A later iteration reviews the repaired bundle from scratch and does not repeat resolved findings. ## Output contract Output only the shared