Integrate Claude Code into CI/CD pipelines

Run the same command CI as you do locally.

Interactive mode waits for input that will never arrive, so the job hangs until it times out with nothing to show.

Non-interactive mode is a requirement of the environment, not a preference.

Parse the findings out of the model's prose in the pipeline.

The wording varies between runs, so the parser silently drops findings whenever the phrasing shifts.

Request structured output against a schema and parse that.

Have the session that generated the code review it, since it has the most context.

It carries the assumptions that produced the code, so the review confirms its own reasoning and misses exactly what it got wrong.

Review from an independent instance; the shared context is the liability.

Re-run the review on each commit and post whatever comes back.

Unresolved issues are re-posted every push, so the thread fills with duplicates and reviewers stop reading it.

Include prior findings in context and ask only for what is new or still unaddressed.

Ask for tests and filter the low-value ones by hand.

Without the existing suite and the standards in context it re-covers what is already tested, and the manual filtering becomes the actual work.

Put the existing tests, the fixtures and what counts as a valuable test into project context.