What problem does it solve? Global harness defaults ship generic thresholds and path globs that do not match any specific repository, causing quality gates to fail constantly or train teams to ignore review lists. This Skill calibrates those defaults against the project's own measurement history so quality gates reflect reality. ## Core Features & Use Cases - Threshold Derivation: Proposes quality thresholds from the distribution of the project's own baseline runs (minimum, median, latest) rather than copying another project's numbers. - High-Risk Path Mapping: Walks the repository tree to match always-review categories (auth, payments, secrets, migrations, CI workflows) against directories that actually exist, dropping irrelevant globs. - Tool Availability Audit: Lists every unavailable gate metric with the concrete cost of enabling it, so gaps are visible instead of hidden. - Blind-Spot Analysis: Reads escaped-defect logs and bug history to propose review lines describing how this system actually breaks. - Use Case: After running harness-bootstrap on a new repo, run this calibration to produce an approval-gated proposal that updates agent-os/quality-thresholds.json and the Quality section of AGENTS.md with repo-specific values. ## Quick Start Ask the agent to calibrate this project's quality thresholds and high-risk paths from its measurement history and produce an approval-gated proposal.
