← ClaudeAtlas

coverloop-contractlisted

The Coverloop operating contract — how much verification a change must earn before it ships. Load this BEFORE writing or reviewing code in a repo that uses Coverloop (has a .coverloop/ directory or a CLAUDE.md carrying PROTOCOL_VERSION), and whenever deciding a risk tier, choosing reviewers, or about to push/merge/migrate/deploy.
danielsimisi-coder/coverloop · ★ 4 · Code & Development · score 77
Install: claude install-skill danielsimisi-coder/coverloop
# The Coverloop contract **One model reviewing its own work is a model grading its own exam.** This contract says how much independent verification a change has to earn before it ships — scaled to what the change can actually break. Two rules override everything below: 1. **No model is an authority.** Every finding — yours or a reviewer's — is a *claim* to be checked against real code, tests, and runtime. When a test can settle it, run the test instead of adding another opinion. 2. **Never send secrets or PII to any model.** No `.env`, keys, tokens, credentials, customer data. Reading your own DB is PII-bound: select non-PII columns only. ## 1. Classify first — the tier is derived, not declared ```bash coverloop classify # explains the floor and why ``` Take the **MAX** of what it reports and your own judgement. It reads the paths that actually changed: | Change touches | Floor | |---|---| | migrations · `.sql` · schema · auth/RLS · billing/payments · secrets/`.env` · CI-deploy config · workers/cron | **L3** | | API routes · handlers · middleware · shared state · dependency manifests | **L2** | | docs · stylesheets · repo metadata | **L0** | | **anything unrecognised** | **L1 — never L0** | | 10+ files changed | **at least L2** | You may **raise** a tier. **You may never lower a deterministic floor** — not to save tokens, not because it "looks fine". Unsure between two tiers? Take the heavier one. ## 2. What each tier must earn | | Build | Tests (the decidi