| name | image-result-auditor |
| description | Audit completed image jobs and export frozen comparison bundles from evidence-backed records. Use when you need approval candidates, retry candidates, comparison summaries, or auto-eval outputs without mutating upstream artifacts. |
Image Result Auditor
Overview
Use this skill after worker completion or bundle freeze to turn evidence-backed records into review-ready queues, comparison summaries, and auto-eval outputs.
It owns audit and consumer preparation only. It does not mutate files, regenerate upstream arms, or approve items without evidence.
Use This Skill When
- worker records already exist
- a human review queue must be prepared
- rename or metadata conflicts must be detected before commit
- frozen per-image comparison bundles already exist
- a downstream consumer must compare arms or run proxy auto-eval without regenerating ledgers
Required MCPs
agent-task-managerconport
Optional MCPs
Not Owned Here
- worker execution
- file rename or metadata commit
- human approval
- cross-tool speculative repair
- arm regeneration
- context package mutation
- semantic judge implementation
- iterative reinjection, baseline comparison, pending closure →
multimodal-evidence-refinement-loop
- review surface layout, machine-truth manifest split →
image-text-cot-review
References
references/runtime.mdreferences/troubleshooting.mdreferences/review-queue-contract.mdreferences/comparison-consumer-runtime.mdreferences/frozen-bundle-contract.mdreferences/semantic-judge-waiver.mdchecklists/audit-preflight.mdchecklists/four-mode-consumer-readiness.mdknowledge_bases/image-result-auditor-knowledge-base-at2026-03-28-11-20.mdknowledge_bases/image-result-auditor-comparison-consumer-knowledge-base-at2026-03-29-00-30.mdevals/evals.json
