What problem does it solve? Teams adopting coding agents often blame the model when agents produce broken or messy PRs, when the real cause is a weak engineering system around the agent. This Skill audits a repository against ten evidence-based rails — from runnable verifiers and green CI to scoped tools and human checkpoints — and emits a severity-ranked scorecard with a prioritized fix list, so you know exactly which rail to build before investing elsewhere. ## Core Features & Use Cases - Two-tier readiness audit: Scores six Build rails (R1–R6, author → merge) on every repo, and four Operate rails (R7–R10, merge → production) when the goal is unattended agents against production, reported as separate subtotals. - Deterministic evidence collector: A read-only Python script walks the target repo and emits sorted JSON evidence per rail — test configs, CI workflows, style docs, agent context files, permission settings, CODEOWNERS — which the agent then verifies and scores 0/1/2. - Optional rail installation: On explicit request, implements the highest-leverage fixes (a single verify entrypoint, CI gating, conventions docs, architecture maps) through the repo's own PR flow, then re-verifies the improved scores. - Use Case: Before onboarding Claude Code or a background coding agent into a codebase, run the audit to discover that CI exists but never runs tests (R2 = 1) and there is no discoverable verify command (R1 = 0) — then fix those load-bearing rails first. ## Quick Start Ask the agent to audit whether this repository is ready for coding agents and produce a readiness scorecard with prioritized fixes.
