What problem does it solve? AI coding agents produce more code than anyone reviews, and without a written quality bar they quietly lower standards — adding suppressions, skipping tests, or weakening thresholds to reach green. This Skill records the project's quality bar as a versioned CONSTRAINTS.md file with concrete numbers and tools, then guards it against silent erosion. ## Core Features & Use Cases - Guided constraint setup: A four-question interview with sane defaults establishes coverage, security, performance, and accessibility thresholds, then writes them into CONSTRAINTS.md with the exact command that checks each one. - Diff-scoped floor guard: A reference implementation detects the five cheap-road-to-green moves — lowered thresholds, skipped tests, new suppression comments, unimplemented stubs, and untracked exceptions — using only git diff. - Lifecycle wiring and ratchets: Maps fast checks to the edit loop and slow checks to CI, and supports ratchet-style "measure today, never get worse" constraints when no target number exists. - Use Case: Before running an autonomous build loop on a TypeScript project, use this Skill to record an 80% changed-line coverage gate, a gitleaks secrets check, and a Lighthouse LCP budget, so every agent-generated commit is verified against the same written bar. ## Quick Start Ask the agent to set up project constraints and quality gates for this repository, writing them into a CONSTRAINTS.md file with enforced thresholds.
