[Khushbu677]

What

Added tests/test_turnstate_hook.py with four tests covering ledger field extraction, duplicate-byte skipping, malformed input, and a missing transcript. Updated the README to reflect the new test coverage.

Why

The hook silently handles errors, so a regression in its SQLite write path could cause ledger entries to stop being recorded without failing the session. Tests make these failures detectable.

Closes #15

How to verify

Command:

python -m pytest -q tests/test_turnstate_hook.py

Actual output: 4 passed.

Red-green verification: In an isolated copy with the SQLite insert disabled, the write-path and duplicate-byte tests failed as expected.

Command:

python -m pytest -q

Actual output: 167 passed, 1 failed. The unrelated failure is tests/test_star_prompt.py::test_tilde_in_config_dir_expands, which also fails when run independently.

Command:

git diff --check

Actual output: no output; passed.

What this does NOT include

• No changes to the production turnstate_hook.py implementation.
• No new dependencies.
• The existing test_tilde_in_config_dir_expands failure is not addressed by this PR.

---

• One idea in this PR.
• No secrets, credentials, personal data or real third-party names in the code, tests, examples or pasted output.
• An AI wrote part or all of this. I reviewed the generated code and ran the tests.

[tonydzi]

Mycroft here, Anton's synthetic AI co-founder, disclosing in the first line because the regulation asks. Thanks @Khushbu677, this is the test #15 asked for.

I checked it against the PR head (333803f) instead of the description:

• python -m pytest -q tests/test_turnstate_hook.py gives 4 passed, and the full python -m pytest -q gives 168 passed.
• I broke the hook on purpose, twice, to see whether the tests notice.
  • Turning the INSERT INTO turns call into a no-op makes test_main_writes_extracted_turn_to_ledger and test_main_does_not_write_again_without_new_transcript_bytes fail (ProgrammingError at turnstate_hook.py:229).
  • Changing the final sys.exit(0) to sys.exit(1) makes both test_module_wrapper_* tests fail.
    So each of the four tests goes red on the break it is meant to catch, and none of them passes by accident.
• Standard library plus pytest only, no new dependency, and the README sentence that claimed the hook has no test is gone.

One thing I would not change but want on record: the two wrapper tests only assert exit code 0, so they pin the "never break a session" contract, not that a bad payload writes nothing. That matches case (3) of the issue, so it is fine as scoped.

Merging now.

[tonydzi]

Approving on the evidence in the review above (4 tests, each red on its own break; full suite 168 passed).