README.md

Standards Verifier

The Verifier evaluates the repository's registered structural checks. Use the Standards Engine public Interface to verify standards and refresh generated inputs. Python dependencies come from requirements.lock.

printf '%s\n' '{"kind":"verify-repository","refresh_verification_inputs":false}' |
  PYTHONPATH=. python3 -P .agents/skills/standards-engine/scripts/invoke.py verify_repository

Inspect verification.passed and its diagnostics. Set refresh_verification_inputs to true after changing declared inputs. The Engine owns that generated update. Neither operation certifies policy completeness.

What the checkpoint checks

Check kind	Actual observation
metadata_graph	Metadata acceptance or rejection for a corpus or fixture
metadata_route	Routing results for declared task facts
markdown_targets	Required navigation destinations, independent of link wording
markdown_links	Declared links and destinations
markdown_link_coverage	Navigation coverage of a declared member inventory
plan_contract	Structured plan lifecycle records
policy_impact	Declared consumer graph and fixture validity
contract_projection	Generated Interface agrees with canonical contract
python_package_contract	Declared Python package boundaries

The complete checkpoint evaluates all registered declarative suites in dependency order. It does not execute the Python unit-test suites, assess prose quality, or prove that downstream code follows a standard. The graph cannot discover undeclared consumers merely by validating its existing registrations.

Testing the implementation

Run functional tests separately when changing the Engine or Verifier:

PYTHONPATH=.:tools/standards_verifier python3 -m unittest discover -s tools/standards_verifier/tests
PYTHONPATH=. python3 -m unittest discover -s tools/standards_engine/tests

These tests exercise execution, navigation, authoring, publication, and failure handling. Use the locked environment for both commands. The low-level diagnostic CLI remains tools/standards_verifier/verify.py with --complete, --suite, --list-suites, and JSON output; Engine callers use verify_repository.

Evidence maintenance

Use Engine maintain_evidence to retire obsolete checks, fixtures, certificates, and registered evidence artifacts. Removing an empty suite also removes its graph node and preserves surviving prerequisites of dependent suites. Consumer review ownership is distinct from suite evidence: review:consumer binds the policy, relationship, and consumer content without importing an unrelated suite's inputs. It supplies no automatic attestation.

Synthetic decision tables, exact-prose checks, historical migration inventories, and arbitrary document budgets have been retired. Git history and captured snapshots preserve historical observations; the current runner no longer executes those retired check kinds. Shared TSV projections remain where navigation and routing checks consume real inventories.

Suite-input authority

The canonical manifest compiler consumes check-owned declarations through a CheckInputContext, separately from filesystem check execution. Directory inputs retain physical containment and symlink checks. Frozen inputs carry exact bytes plus explicit regular-file repository membership; an indexed file whose bytes were not captured is unavailable, rather than empty. Both adapters use the same registry parsing, declarations, hashing and manifest serializer.

Engine logical authoring calls suite_input_projection_bytes_from_content with its frozen candidate and computed path population. It creates no temporary Git repository for that derivation. Full verification and publication continue to execute against the actual candidate checkout, including safe-path package and entrypoint processes. The manifest representation and schema version are unchanged.