diff --git a/.agents/skills/pr-review/SKILL.md b/.agents/skills/pr-review/SKILL.md index b76f0a524da..3831c3a3fe1 100644 --- a/.agents/skills/pr-review/SKILL.md +++ b/.agents/skills/pr-review/SKILL.md @@ -186,7 +186,19 @@ on code the change adds or modifies, not on pre-existing issues next to it. another Go version, so run the changed acceptance tests under one (`GOTOOLCHAIN=goX.Y.Z` downloads it): a golden whose values change across toolchains asserts the standard library, not the CLI, and would - fail only on the next `Bump Go toolchain` PR. + fail only on the next `Bump Go toolchain` PR. If `gh pr checks` reports + no checks for the head, say so in Evidence and run the changed tests + locally; missing checks are not a green build. For a change to the + acceptance harness or to CI test selection, a green cell shows the + suite passed, not that the new path ran: grep the job log for the line + the change adds (`gh run view --job --log`). + - The eng-dev-ecosystem-bot "Integration test report" comment lists, per + cloud, which acceptance tests ran, passed or were skipped; use it when + the deco run itself is not readable. In it, flaky means fail-then-pass + on rerun and RECOVERED means the test is listed in `known_failures.txt` + on the `ciconfig` branch. On a flake-fix PR, read that file (and its + commit history) before treating a flaky mark on the changed tests as + evidence the fix did not work; the entry often names a different error. - Does each new or changed test show the effect it claims? The output should include the state after the action (list or print it again after a sync, a deploy or a migration) and the exact value the change is