Skip to content

fix(skills): verify newly-added CI jobs pass, not just required checks#169

Closed
tend-agent wants to merge 1 commit intomainfrom
hourly/review-24116573856
Closed

fix(skills): verify newly-added CI jobs pass, not just required checks#169
tend-agent wants to merge 1 commit intomainfrom
hourly/review-24116573856

Conversation

@tend-agent
Copy link
Copy Markdown
Collaborator

Summary

  • Add guidance to the CI monitoring section to verify newly-created or modified CI jobs pass, not just pre-existing required checks.

Evidence

Run IDs: 24116374057 (tend-mention), 24116452569 (tend-review)

Target repo: PRQL/prql, PR #5779

What happened: The mention run responded to a maintainer request by adding a test-devcontainer job to tests.yaml. The new job failed (uv not in PATH in the devcontainer CI environment). However, the bot's CI polling loop only checked --required checks (pre-commit.ci), so it reported "CI passed." The subsequent review run also polled only required checks and concluded "no issues found." Both sessions incorrectly reasoned the new job "likely didn't run" due to path filters — it did run and failed.

Root cause: The running-in-ci skill's CI monitoring section instructs bots to poll gh pr checks --required. Newly added CI jobs aren't required, so their failures are invisible to this polling. The bot has no guidance to check its own newly-created jobs.

Fix: Add a paragraph after the required-checks instructions telling bots to also verify any jobs they created or modified, using gh pr checks without --required.

Gate assessment

  • Evidence level: Critical — the bot reported CI success while its own change broke CI
  • Failure type: Structural — the --required flag deterministically excludes new non-required jobs; will recur every time the bot adds a CI job
  • Change type: Targeted fix (6 lines added to existing section)
  • Historical evidence: First observation, but structural failures need only 1 occurrence

🤖 Generated with Claude Code

The CI monitoring loop polls only `--required` checks. When the bot adds a
new CI job (which isn't yet required), it never notices if that job fails —
it reports "CI passed" based on pre-existing required checks alone.

Observed on PRQL/prql PR #5779: the bot added a `test-devcontainer` job
that failed (uv not in PATH), but both the mention and review runs reported
success because only `pre-commit.ci` (the sole required check) was polled.

Add guidance to also verify any jobs the bot created or modified after the
required-checks loop completes.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@tend-agent tend-agent added the claude-behavior Behavioral issues found by review-reviewers label Apr 8, 2026
@max-sixty max-sixty closed this Apr 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

claude-behavior Behavioral issues found by review-reviewers

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants