[bug] Match GitHub required-check conclusions for skipped, neutral, and startup failure

[coygeek]

[bug] Match GitHub required-check conclusions for skipped, neutral, and startup failure

Summary

DeployBot's check-state reducer does not match GitHub's required-status-check semantics. It only treats SUCCESS as passing, treats every other non-failure as pending, and does not include STARTUP_FAILURE in its failure set.

GitHub's required-check documentation says successful required check statuses are success, skipped, and neutral. GitHub's GraphQL schema also exposes STARTUP_FAILURE as a check conclusion, which is a terminal failure state, not an in-progress state.

Evidence

• DeployBot defines FAILED_CHECK_STATES without STARTUP_FAILURE in src/agent_merge_queue/cli.py lines 77-84.
• check_states() returns passed only for SUCCESS, failed only for FAILED_CHECK_STATES, and pending for everything else in src/agent_merge_queue/cli.py lines 154-162.
• The current test suite locks in the mismatch: tests/test_cli.py lines 144-151 asserts a SKIPPED required check remains pending.
• The GitHub Docs source repository is https://github.com/github/docs/; the source file says: "Successful check statuses are: success, skipped, and neutral" in https://github.com/github/docs/blob/main/content/pull-requests/collaborating-with-pull-requests/collaborating-on-repositories-with-code-quality-features/troubleshooting-required-status-checks.md.
• The current public GitHub Docs page has the same rule: https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/collaborating-on-repositories-with-code-quality-features/troubleshooting-required-status-checks
• The GitHub Docs source repository is https://github.com/github/docs/; its GraphQL schema source lists NEUTRAL, SKIPPED, STARTUP_FAILURE, SUCCESS, and TIMED_OUT in CheckConclusionState in https://github.com/github/docs/blob/main/src/graphql/data/fpt/schema.docs.graphql.

Impact

Pull requests can be stuck forever when a required GitHub check legitimately completes as skipped or neutral. Conversely, a check run that ends with STARTUP_FAILURE is reported as pending, which hides a terminal failure and produces misleading queue status.

This makes DeployBot disagree with GitHub's own mergeability model, so the queue can either wait unnecessarily or fail late with unclear diagnostics.

Proposed fix

• Treat SUCCESS, SKIPPED, and NEUTRAL as passing terminal states for required-check evaluation.
• Add STARTUP_FAILURE to the terminal failed-state set.
• Add regression coverage for:
  • SKIPPED required check is passed.
  • NEUTRAL required check is passed.
  • STARTUP_FAILURE required check is failed.
• Update the existing skipped-check test that currently encodes the wrong behavior.

Verification

• python3 -m pytest tests/test_cli.py -q
• python3 -m pytest -q