Usage rules

Copy Markdown View Source
  1. Treat verifier success as a cryptographic result, not an authorization decision.
  2. Supply only already-trusted public keys selected outside this library.
  3. Derive method, URI, operation, and cast arguments on the server; never trust caller-selected expected context.
  4. Enforce live revocation, replay reservation, execution claims, host policy, and effect authorization in a stateful authority runtime.
  5. Use the published bounds and reject unknown versions or extensions.
  6. Never pass private keys, secrets, raw user values, or production credentials into fixtures, logs, errors, or telemetry.
  7. Pin released major versions and run the conformance vectors before accepting a new release.
  8. Treat untrusted_key_locator/2 output only as a case-sensitive lookup hint with trust: :not_evaluated; it does not parse claims, verify bytes, select trust, or authorize.
  9. Pass only raw compact credentials to verification boundaries. Decoded values and verified facts are evidence outputs, never reusable credentials.
  10. Supply chain/archive verification with the intended predecessor/head, both expected anchors, the complete ordered historical-key path, raw archive digest, and exact object-store version. Do not infer those expectations from the archive being verified.
  11. Treat chain consistency as consistency only. It cannot by itself prove that a validly shortened or relinked history omitted nothing.
  12. Keep commitment preimages private. The public row carries only a fixed-width commitment.
  13. Consume the published Hex release ({:bounded_authority_protocol, "~> 0.1.0"}). The v1 verification surface is implemented and the published archive is the exact reviewed candidate.
  14. Run the verifier CLI with an explicit --corpus DIR pointing at the packaged corpus (deps/bounded_authority_protocol/priv/conformance/v1/corpus from a consumer). Never rely on a default corpus path. Treat exit 0 as conformance evidence only — it does not authorize.