fix: skip step-checker in container jobs where _diag is not accessible

[adityamaru]

Problem

The step-checker feature was causing warnings and preventing sticky disk commits in container jobs:

Warning: Unable to check for previous step failures: _diag directory not found at /__w/web/web/_diag
Warning: Skipping sticky disk commit due to ambiguity in failure detection

Root Cause:

• Container jobs mount the workspace at /__w/ inside the container
• The _diag directory exists on the host at /home/runner/_diag but is not mounted into containers
• The step-checker tried to access _diag and failed, returning an error
• This prevented sticky disk commits from happening
• Modern kernels use cgroup v2 which doesn't include docker/containerd in the path

Impact:

• Affecting ~300-400 runs per day since November 1st, 2025 when step-checker was deployed
• Only affects container jobs (jobs with container: in workflow config)

Solution

Add multi-layered container detection that works with both old and new systems:

1. Check for /.dockerenv file (docker-specific indicator)
2. Check /proc/1/cgroup for docker/containerd (cgroup v1)
3. Check if working directory starts with /__w/ (GitHub Actions container mount - works with cgroup v2)

When any of these conditions are met, skip the step-checker gracefully since _diag is not accessible.

No changes to existing path detection logic - only adds the container check at the beginning.

Testing

• ✅ All tests pass with updated mocks
• ✅ Build successful with no linter errors
• ✅ Verified detection works in actual container (pwd starts with /__w/ ✓)
• ✅ Container jobs will now skip step-checker and allow sticky disk commits
• ✅ Non-container jobs continue to work exactly as before

Verification in VM

Tested in actual container environment:

✓ Has /.dockerenv
✗ Not in cgroup (cgroup v2: "0::/")
✓ PWD starts with /__w/
✓ CONTAINER DETECTED - step-checker will be skipped

Related

Based on ClickHouse analysis, this has been affecting the useblacksmith/web repo's container verification jobs since the step-checker feature was deployed on November 1st.

---

Note

Adds container detection (/.dockerenv, /proc/1/cgroup, cwd prefix) to skip step-checker when _diag isn’t accessible, and updates tests accordingly.

• Step-checker:
  • Add container detection using /.dockerenv, /proc/1/cgroup (docker/containerd), and process.cwd().startsWith('/__w/').
  • When detected, skip checking _diag and return no failures; add debug logs for detected paths.
  • Preserve existing runner root detection and log parsing logic.
• Tests:
  • Update src/step-checker.test.ts to mock container/non-container signals and verify behaviors (no _diag, no logs, JSON/text parsing, error handling, boolean check).
• Build:
  • Regenerate dist/index.js and source map.

^{Written by Cursor Bugbot for commit e1196de. This will update automatically on new commits. Configure here.}