Skip to main content

How checks run

Evaluation flow

  1. Load prodverdict.yml and validate schema (version: 1).
  2. Collect observations — read-only Stripe/Paddle API, read-only DB, repo files, or fixtures.
  3. Apply rules — deterministic comparisons; no LLM.
  4. Emit findings{ contract, severity, entity, message, fix? }.
  5. Compute verdictpass | warn | fail from highest severity.

Where secrets run

ContractCredentialsWhere
AccessSTRIPE_SECRET_KEY, DATABASE_URLYour laptop or CI only
ConfigRepo files + optional envCI / local
MigrationSQL files in repoCI / local
Remote MCPGitHub App tokenProdVerdict cloud reads repo files only

ProdVerdict cloud stores verdict metadata — never subscription rows or PII from your database.

Fixture mode

--fixtures loads JSON from fixtures/ or --fixtures-dir instead of live APIs. Use for demos, unit-style runs, and agent onboarding without credentials.

Phase order

When running multiple contracts, evaluate in this order: Access → Config → Migration → Boundary → Webhook → Restore.

npx prodverdict check all