Local passes, CI fails: what is the first clue you capture?

A concrete debugging question about the first evidence developers should preserve before changing code after a CI-only failure.

When tests pass locally but fail in CI, I keep seeing teams jump straight into code edits. That can work, but it can also erase the useful clue: exact command, runtime version, missing variable, lockfile state, or filesystem assumption. My current rule is to capture the first useful error line and the CI command before editing. If the failure points to a plain broken contract, patch locally. If it points to environment drift, reproduce the CI path first. What is the first CI-only clue you preserve before changing code?