metaobjects-audit

Run the audit against the repository root; it triages MetaObjects presence, classifies the project as greenfield, partial, or deep, then works an 8-axis surface review. It writes findings to .metaobjects/adoption-audit.json plus a rendered Markdown report without editing any code.

It hunts hand-written validators shadowing generated schemas, field-by-field DTO mappers, camelCase-to-snake_case maps, keep-in-sync comments, runtime schema patching, duplicate shape declarations, and hand-written SQL views that projections or entity read-views could replace.

Yes, it ships per-port reference guides for TypeScript, Java, Kotlin, C#, and Python covering generated-file locations, codegen commands, drift signatures, and calibration gaps. Cross-language version consistency is checked via each port's reported metamodelVersion.

No, the audit is strictly read-only. It never edits code or authors metadata; each finding may include a metadata_sketch proposal for human review, and actual cutovers are routed to separate skills per finding tier.

That pairing generates a String property over a uuid column, forcing String-to-UUID coercion at every boundary, and verify --db cannot catch it because the column type matches. The recommended fix is field.uuid, staged by blast radius.

The requirement.* capability ledger is entirely opt-in, so a project declaring no requirement nodes sees no diagnostics and absence is never scored as a defect. The audit instead scores the hand-written prose, such as a CAPABILITIES.md, that the ledger would replace.