[ci-doctor] CI Test Failure — Run #29057896142 (bun-version: latest regression)

Start with .github/workflows/ci.yml and compare the failing run 29057896142 with the previous successful run 29052744071, focusing on the Bun version used by the Test step. Pin the configured Bun version as described, then rerun the workflow and confirm that bun test --coverage ./tests/ completes successfully.

Summary

The Test & Lint job failed at the Test step (bun test --coverage ./tests/) on commit 7c5ea995, which merged PR #410 ("Improve Evergreen CI diagnostics"). Only workflow files were changed — no source or test code was modified.

Failure Details

• Run: 29057896142
• Commit: 7c5ea995df0694314856f08440316b53004183d9
• Trigger: push (merge of PR #410)
• Failed Step: bun test --coverage ./tests/ (ran for ~4 seconds)
• Previous successful run: 29052744071 on commit a7269d37

Root Cause Analysis

No source code or test files were changed between the successful and failing runs — only workflow files (.github/workflows/evergreen.lock.yml, .github/workflows/evergreen.md, .github/workflows/shared/evergreen/repo-policy.md).

The most likely cause is that bun-version: latest in ci.yml pulled a new bun version between the two runs, which introduced a regression in the test runner or coverage reporting.

This is a recurring pattern — an identical failure occurred previously (see #307).

Failed Jobs and Errors

CI logs are not publicly accessible (require admin), so the exact test failure message is unavailable.

Investigation Findings

1. Only .github/workflows/ files changed — code is unchanged
2. The Test step ran for ~4 seconds (23:38:38–23:38:42), which is abnormally short for a full test suite
3. Build and cross-validation steps were skipped (they depend on test passing)
4. This is at least the third occurrence of this pattern (see also #307)

Recommended Actions

• Pin bun version in .github/workflows/ci.yml: change bun-version: latest → bun-version: "1.2.x" (or the last known-good version)
• Retry the failed run — a re-run may succeed if the issue is a transient bun release
• Check bun release notes for any test runner or coverage changes between the last successful and current bun version

Prevention Strategies

1. Pin bun to a specific version in all CI workflows; update deliberately after reviewing release notes
2. Consider adding a continue-on-error: false with explicit bun version check
3. Add a note to AGENTS.md reminding agents that bun-version: latest is not allowed in CI

AI Team Self-Improvement

## CI Stability: Bun Version Pinning (CRITICAL)
- NEVER use `bun-version: latest` in CI workflow files (.github/workflows/*.yml).
  Always pin to a specific version, e.g. `bun-version: "1.2.15"`.
- When modifying any CI workflow, verify that `bun-version` is pinned and not set to `latest`.
- Recurring test failures where only workflow files changed are almost always caused by
  bun-version: latest pulling a new release with a regression. Recommend pinning immediately.

Historical Context

• Issue #307 (May 2026): Same pattern — bun test CI failure after autoloop merge, only workflow or unrelated files changed, tests pass locally. Root cause identified as bun-version: latest.
• This is the third known occurrence of this pattern in this repository.

Generated by CI Failure Doctor · 326.4 AIC · ⌖ 11 AIC · ⊞ 28.1K · ◷

gh aw add githubnext/agentics/workflows/ci-doctor.md@c7d030cd6d4607b90d9ac3ffc8b24aff4f251632