What problem does it solve? When asked to "build a harness," AI agents often produce static documents or checklists that lack machine-verifiable pass/fail judgment. This Skill enforces a strict definition of a verification harness and provides fixed procedures to build one that actually detects RED/GREEN states mechanically. ## Core Features & Use Cases - Two-scale classification: Distinguishes unit harnesses (hooks, CI steps, single checks requiring only Red/Green detection) from system harnesses (full REPF structure: Red/Green observation, Progress tracking, Executable artifacts, Flow documentation). - Anti-pattern guardrails: Rejects deliverables like Markdown runbooks, checklists, or dashboards that lack machine judgment, preventing false "harness" claims. - Reference pattern library: Ships a reusable pattern for pre/post state verification using already-authenticated CLIs (cloud, SaaS, hosting, DB) with read-only, idempotent check.sh scripts. - Use Case: Before a multi-resource cloud migration, generate a harness with check.sh, baseline snapshot, PROGRESS.md, and README so any future operator can resume work and verify GREEN completion with one command. ## Quick Start Ask the agent to build a verification harness for your migration or recurring check, and it will first confirm whether you need a unit or system harness before constructing the REPF structure.
