Claude Code prompt: fix a failing test without touching the test

A test is failing and I want the code fixed, not the test.

Failing test: [test file and test name, or "run the suite and find it"]
Run just that test with: [command]
What it checks: [one line on the behavior the test expects]

Ground rules:
- The test file is read-only. Don't edit, skip, or delete the test, and don't loosen its assertions, timeouts, or snapshots.
- Don't add mocks, flags, or special cases that only exist to make this test pass.
- If you think the test itself is wrong, stop and tell me why with file:line evidence.
- Keep the fix inside [allowed paths]. Ask before touching anything else.
- No new dependencies.

Steps:
1. Run the failing test and show me the part of the output that matters.
2. Read the code under test and explain the cause in two or three sentences. Point to the line where what the code does and what the test expects split apart.
3. Make the smallest change that fixes that cause.
4. Run the test again. Then run [wider command, e.g. all tests for this module] to make sure nothing nearby broke.
5. Run git diff --name-only and confirm no test files are in the list.

Report back: the cause, files changed, the before and after test output, and anything you noticed along the way but didn't fix.

[N] tests started failing after [merge, commit, or version bump]. Don't fix anything yet.

Run [test command] and group the failures by likely shared cause. For each group, give me the tests in it, one line on the cause, and the file you'd change. Then wait while I pick which group to fix first. Test files stay read-only for this whole session.

[test name] passes sometimes and fails sometimes. Run it [10] times with [command] and write down the results.

Look for order dependence, shared state between tests, timing and sleeps, real network or clock calls, and random data without a fixed seed. Tell me which one it is, with evidence, before you change anything. If the fix belongs in test setup rather than the assertions, show me the diff and wait for my OK.

[test name] passes on my machine but fails in CI. Here's the CI log: [paste]

Compare the two environments: runtime and dependency versions, env vars, timezone, locale, file paths, and whether tests run in parallel. List the differences that could explain the failure, most likely first, and give me one command that reproduces the CI failure locally. No code changes until we can reproduce it.