test: restore stale registry key consistency check

[popwatt]

Summary

The bundled Cookiecutter template registry (templates/index.json) must stay in sync with the actual template directories under templates/. This PR restores one of the registry consistency checks that #92 originally asked for: every key in templates/index.json must point to an existing templates/<type>/<name>/ directory containing cookiecutter.json (a "stale key" check).

Regression context

The original check shipped in #100 (commit 920c1cb) as test_registered_templates_point_to_cookiecutter_directories. During the lifecycle-toolkit refactor in #114 (commit e87460d), it was removed and replaced by test_registry_contains_expected_template_keys, which only asserts that a fixed set (EXPECTED_TEMPLATE_KEYS) is a subset of the registry. That check is one-directional: it cannot detect extra or unknown keys.

As a result, since #114 the repo has had no test that fails when templates/index.json contains a stale key (e.g. a key pointing to a directory that was deleted, renamed, or never merged). Such a stale key would make the CLI advertise a template that cannot be rendered, and CI would pass silently — exactly the failure mode #92's acceptance criteria ("Tests fail when templates/index.json contains a stale key") were meant to prevent.

Relation to #92 / #100

• Add template registry consistency tests #92 requested registry consistency tests (no new CLI command; checks focused on the current registry/template layout).
• test: add template registry consistency checks #100 implemented three checks, including the stale-key check.
• feat: streamline AgentSeek around lifecycle toolkit workflow #114's refactor unintentionally dropped the stale-key check.
• This PR restores it so Add template registry consistency tests #92's acceptance criteria are fully covered again. The quarantined bub/contextseek is unaffected: it is deliberately not registered, so it can never be reported as stale.

Solution Description

Add test_registered_templates_point_to_cookiecutter_directories to tests/cli_commands/test_templates_registry.py:

def test_registered_templates_point_to_cookiecutter_directories() -> None:
    """Every registry key points to a template directory with cookiecutter.json."""
    stale = sorted(_registered_templates() - _template_dirs())

    assert not stale, f"stale templates/index.json key(s): {stale}"

_template_dirs() walks templates/ and collects every directory containing a cookiecutter.json; _registered_templates() reads the keys of templates/index.json. The test fails exactly when the registry advertises a template that has no backing directory. The test name and assertion match the version originally shipped in #100, so the diff is minimal — one test, no production code, no new CLI command.

Validation

$ pytest tests/cli_commands/test_templates_registry.py -q
6 passed in 0.02s

$ pytest tests/cli_commands/test_templates_render.py -q   # existing render tests
50 passed in 12.64s

$ ruff check tests/cli_commands/test_templates_registry.py
All checks passed!

Negative case — verifying the test actually catches a stale key: temporarily added "langchain/nonexistent": "..." to templates/index.json and re-ran:

$ pytest tests/cli_commands/test_templates_registry.py -q
1 failed, 5 passed
>       assert not stale, f"stale templates/index.json key(s): {stale}"
E       AssertionError: stale templates/index.json key(s): ['langchain/nonexistent']

templates/index.json was restored afterwards; the working tree contains only the test change.

[popwatt]

#92

[popwatt]

#92

[webup]

Thanks for the patch. Could you please add more context to the currently blank Summary and Solution Description sections?

In particular, it would help to explain:

• what regression or real-world failure prompted restoring this stale-key check;
• how this follow-up relates to the already-completed Add template registry consistency tests #92/test: add template registry consistency checks #100 work, and why the check removed in feat: streamline AgentSeek around lifecycle toolkit workflow #114 should now be restored; and
• the exact test command and results, ideally including a negative case showing that an intentionally added stale templates/index.json key fails as expected.

That context will make it easier to confirm that this is still the intended registry invariant.

[popwatt]

Thanks for the review! I have filled in the Summary and Solution Description sections of the PR description with the regression context (check originally from #100, dropped in #114), the relation to the #92/#100 work, and the exact test commands/results including the negative case (injected stale key langchain/nonexistent fails with AssertionError: stale templates/index.json key(s), then restored).