story-013: Move the schema inventory out of tests/ into schemas/manifest.json

[jerodw]

Summary

Adding a file to schemas/ forced the implementer to hand-edit two exact-equality inventories inside tests/ — SHIPPED in tests/test_schema_validator.py and SHIPPED_SCHEMAS in tests/test_story_004_validation.py. Every story since story-008 has forbidden the implementer from touching anything under tests/. Both rules were correct and could not both be satisfied; story-011 is where they collided, and its verifier recorded the deviation rather than closing it.

The inventory's value is that declaring a new schema must be a deliberate, noticed act. Nothing about that requires the declaration to live in a test file. This story moves it beside the schemas it declares: schemas/manifest.json becomes the one declared inventory, and the two tests stop being where the fact is stored and become where it is checked.

The duplication dissolves as a side effect — both test files read one manifest instead of each maintaining a copy.

Changes

• schemas/manifest.json (new) — the single declared inventory, naming the seven schemas the harness ships.
• orchestration/schema_validator.py — shipped_schemas(harness_root) reads the manifest and is what both test files import instead of holding copies. It raises on a missing or malformed manifest rather than degrading to an empty inventory, which would make the parametrized per-schema checks silently pass over nothing.
• tests/test_schema_validator.py, tests/test_story_004_validation.py — the SHIPPED tuple and the SHIPPED_SCHEMAS set deleted, every use repointed at shipped_schemas(). Both still assert exact set equality against schemas/, not a subset. The directory glob narrows to *.schema.json so the manifest is not counted as a schema, with a companion assertion that schemas/ holds nothing but *.schema.json files plus manifest.json — replacing the coverage the old glob("*") gave for free.
• tests/test_story_013_validation.py (new, tester stage) — 55 new tests: the single-inventory search, both failure directions, the stray-file assertion, and the demonstration that adding a throwaway schema now requires edits only under schemas/.

Testing

750 passed, zero failures — and the harness verified that number twice, in two environments:

• test-results.json: 750 run, 750 passed.
• clean-clone-result.json: exit_code: 0, 750 passed, python_version: 3.10.20 — the suite re-run in a fresh clone with the story committed into it, on the oldest Python CI tests.

Verification passed on the first iteration with zero blocking issues and no retry.

Notes for review

• This is a re-run. story-013's first attempt shipped one test that failed once the run committed and two that could not fail at all — it resolved baselines against HEAD, which stops meaning anything the moment _complete commits the working tree. That attempt was reset and archived under .harness/runs-archive/story-013-vacuous-tests/, and it is the reason story-014 and story-015 were written and merged first. This run has neither defect: the clean-clone check story-014 added exercised the committed state directly, and story-015's baseline-honesty check now fails the suite on that idiom mechanically.
• The manifest names seven schemas, not the six the story's first acceptance criterion enumerates. clean-clone-result was added by story-014 after story-013 was planned. The verifier recorded this as a stale criterion rather than a defect — the story adds and removes no schema, which is the constraint that actually matters.
• The implementer's changed-files.json lists paths under tests/, deliberately. Deleting those two inventory definitions is this story's deliverable, and the story was written to say so rather than carrying the usual "touches no file under tests/" sentence it would have contradicted. From here that sentence holds verbatim for every future story: adding a schema now means editing schemas/ and its manifest, both implementer territory.
• The tester modified tests/test_story_015_validation.py — story-015's tests reference the schema inventory, so they moved with it.
• Two limits the verifier stated rather than glossed: the literal-inventory AST scan exempts one shape (a collection under a *.md key, which is test_context_assembler.py's prompt-to-schema map), and it covers tests/*.py and orchestration/*.py only — verified there are no subdirectories under either today, but a future one would fall outside it.

🤖 Generated with Claude Code