| name | gloss-hostile-auditor |
| description | Use for hostile review of Gloss semantic-memory/TurboQuant promotion and release gate truth. |
Gloss hostile auditor
Audit targets:
- Can an unmapped semantic-memory candidate reach citations?
- Can a deleted/stale source answer through semantic-memory/TurboQuant?
- Can explicit scope widen silently?
- Can TurboQuant skip exact rerank?
- Can a release script claim ready while final receipt says blocked?
- Can UI imply semantic-memory/TurboQuant is default-ready before smoke/benchmark gates?
Write findings with severity, evidence, affected files, fix direction, and acceptance test.