moai-workflow-docs-claim-check

Supply the README plus evidence files such as manifests, CI logs, or directory listings. The skill extracts every checkable statement, splits composite claims into atomic ones, and labels each as verified, stale-suspected, unsupported, or needs-human with named evidence anchors.

Stale-suspected means the claim was once true but a version, date, count, or name has drifted, making it a maintenance task. Unsupported means the evidence never carried the claim, with reasons of missing-evidence, contradicted, or insufficient-coverage.

No, the skill never executes commands, builds, or tests. When a claim can only be settled by running something, it labels the claim needs-human and names the exact command and target file a maintainer should run.

No, the skill is strictly read-only and produces findings only. It describes what a maintainer would change and where, but never emits patches, diffs, or rewritten passages.

An empty evidence set is valid input. No claim can be labeled verified, so the report lists every claim as unsupported or needs-human and names the specific evidence a maintainer still needs to supply.

Subjective statements cannot be checked against evidence, so they are excluded from labeling and listed under Input Scope Reviewed with a one-line reason. When a subjective adjective wraps a checkable core, the core is labeled and the adjective excluded.