What problem does it solve?
It prevents workflow and harness changes from becoming uncontrolled by converting harness-related issues into approval-gated, verification-backed proposals rather than ad-hoc edits.
Core Features & Use Cases
- Report-first harness proposals: Classifies the signal and produces a structured change report using the provided template when harness maintenance is needed.
- Approval-gated core edits: Ensures core harness modifications (routing, rules, command/skill contracts, memory policy, install/setup scripts, and harness docs) are applied only with explicit human approval.
- Verification and rollback planning: Requires a named verification trigger and a rollback plan for every proposed or applied change, recording verdict and evidence pointers after execution.
Quick Start
Ask the AI to generate an approval-gated harness change proposal after you identify repeated harness failures, using the existing harness-issue-triage evidence if available.