Reliably run tests after implementation

[max-tet]

I have seen some instances of pull requests you did where the tests were failing. That should not happen, you are capable of running tests yourself and make sure they pass. Make the required changes to the prompt(s).

[ClaydeCode]

Preliminary Plan

The implement.j2 prompt currently instructs Claude to run tests only "if a test command is discoverable" (step 5), making it advisory rather than mandatory. This explains why PRs occasionally have failing tests.

Approach: Update implement.j2 to make test execution mandatory and blocking when a test suite exists — Claude must run the tests and ensure they pass before committing. Specifically:

• If no test runner is discoverable, skip the step.
• If a test suite exists: run it, ensure it passes (pre-existing failures are exempt), and add new tests where applicable.
• Any pre-existing failures must be noted in the PR summary comment.

I'll also check address_review.j2 to apply the same treatment for review-addressing commits.

---

This looks like a small issue. React with 👍 to approve this preliminary plan and proceed directly to implementation.

[max-tet]

• Pre-existing failures are ok, no need to fix them, but they should be mentioned in the summary of the PR.
• If no tests are set up, sure, skip them. But if a test suite exists, it must pass (minus pre-existing failures). And new tests should be added if applicable.

[ClaydeCode]

Plan updated. Answered both clarifying questions based on @max-tet's feedback: (1) Pre-existing test failures are acceptable but must be mentioned in the PR summary. (2) If no test runner is discoverable, skip the step — but if a test suite exists, it must pass (excluding pre-existing failures). Also added the requirement to write new tests where applicable.

[ClaydeCode]

Implementation complete — PR opened: #59