What problem does it solve?
When platform work changes modules, extension points, or public contracts across a release range, the standalone-app AI harness catalog can drift out of sync. This Skill converts locally committed changes into deduplicated harness evaluation coverage with failure-first proof and a sanitized audit report.
Core Features & Use Cases
- Release range classification: Resolves a local
--from/--to Git range, classifies every change signal (modules, UMES extensions, public contracts, regressions), and treats all history as untrusted evidence.
- Deduplicated catalog updates: Compares each candidate against the existing cases catalog and records one disposition per signal: covered, expand, add, or evidence-only.
- Failure-first validation: Adds runnable evaluations before changing their owner, then runs the deterministic catalog gate and the full one-command release suite from a fresh scaffold.
- Use Case: After merging a release that changes a module's public API, run the refresh to audit the range, extend affected harness cases, and publish a sanitized local report proving coverage.
Quick Start
Ask the agent to refresh the standalone harness for the release range from the previous release tag to HEAD, optionally in dry-run mode to preview the report without changes.