project-auditlisted

# project-audit Audit the repo end-to-end with extended thinking: prove the docs match the code, score every feature group /100 against the bar "the most competently executed, robust, performant product of its kind available today" (a product brief that defines its own groups + bar takes precedence), and turn every lost point into a concrete backlog item. Read-only with respect to product code — outputs are docs only (report + backlog + status pointers). Adapter: `.claude/ai-dev-kit.config.json` (`docs` block, `projectType`); missing paths → discover them from the repo and say so. ## 1. Inventory (docs first, then code) - **If a prior audit report exists, bound the surface via git first**: diff the last-audited sha against HEAD. Code byte-identical to an already-verified tree carries the prior pass's per-group findings by identity — spend the pass on what actually changed, plus the checks below that *time alone* can invalidate on an unchanged tree (currency, external gates, the live public surface). State the bounding in the report. - Read the project's status, backlog, and verification docs plus **every** agent context doc (adapter `docs.status` / `docs.backlog` / `docs.contextDir`). Note each doc's checkable claims (file paths, line refs, behaviors, "X is verified/gated/opt-in"). - Sweep the code: root configs (package manager/toolchain, task runner, lint/format), each entry point's source (routes/scenes/commands/modules per the project type), every w