Repository navigation
Conversation
The `pull_request` trigger in `ci.yml` and `smoke-test.yml` was pinned to `main`, so a pull request opened against any other base branch got no lint, build or test run at all - only the unconditional metadata checks. GitHub resolves a `pull_request` workflow definition from the PR's own tree, never from the default branch, so the patch has to be present in the branch under test. `push` stays pinned to `main`. Co-Authored-By: Wayland <wayland@agents.flopbut.local> Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
Probe PR - its purpose was to prove that removing the |
Probe PR. Do not merge - it exists to prove that the trigger patch makes real build/lint/test jobs run on a pull request whose base is not
main.Base
base/upstream-dev-flo896is a verbatim snapshot ofupstream/dev(80430fd2862bfb43a6b0b1993f57033e45a9b98b); the only diff is the one commit below.What changed
.github/workflows/ci.ymland.github/workflows/smoke-test.yml: drop thebranches:filter underpull_request:.push:stays pinned tomain. 2 files, -2 lines, nothing else.Why it has to live here
GitHub Actions resolves a
pull_requestworkflow definition from the PR's own tree (head / merge commit), never from the default branch. Onlyschedule/workflow_dispatch/workflow_run/repository_dispatchare read from the default branch. So patching the fork'smainalone produces zero runs on a PR into a review base.Expected result
Lint (TS), the build jobs andSmoke Test (Mock Mode)appear with non-zero durations instead of only Labeler / Greetings / Validate PR Title. Red real jobs are an acceptable outcome here: the upstreamci.ymlcarries drift/parity gates that may not pass on a bare snapshot. The point is that the signal exists.