CI Monitor
Watch GitHub Actions CI runs after a push. Auto-detects repo and branch, deduplicates concurrent monitors, reports results.
Interpreting skill arguments
Anything on the ARGUMENTS: line is a set of monitoring parameters, never a task description. Pick out the branch name and/or hex commit SHA and pass them as --branch / --sha; ignore filler words like "push", "commit", or "monitor". For example, ARGUMENTS: master push 0bed725ba7cb means:
uv run ~/.claude/skills/monitoring-ci/ci-monitor.py --branch master --sha 0bed725ba7cb
This skill is strictly read-only: run ci-monitor.py and report its result. Do NOT edit ci-monitor.py, this file, or any other file; do NOT commit, push, or change the script's interface — even if the arguments look like they don't match the script's flags. If you can't extract a branch or SHA from the arguments, run the script with no flags (it auto-detects) and say so in your report.
Usage
Run the script in the background after pushing:
uv run ~/.claude/skills/monitoring-ci/ci-monitor.py
With explicit branch:
uv run ~/.claude/skills/monitoring-ci/ci-monitor.py --branch my-feature
How to Invoke from Claude
After a push, resolve the HEAD SHA before launching the background command, then pass it via --sha:
# Get the SHA that was just pushed (jj or git)
SHA=$(jj log -r '@-' --no-graph -T 'commit_id' 2>/dev/null || git rev-parse HEAD)
# Run in background with SHA pinning
uv run ~/.claude/skills/monitoring-ci/ci-monitor.py --branch <branch-name> --sha "$SHA"
Use run_in_background: true so it doesn't block the conversation. Tell the user: "CI monitor running in background — you'll be notified when it completes."
IMPORTANT: Always pass --sha to avoid watching a stale run from a previous push. The monitor matches runs by SHA, so it will wait for the correct run to appear.
Features
- Auto-detects branch: jj bookmarks first, falls back to git
- Deduplicates per push: sentinel at
/tmp/{repo}-ci-monitor-{sha12}prevents double-watching the same push, while monitors for different pushes coexist; stale sentinels (dead monitor) are taken over automatically - Polls for run: waits up to 180s for a CI run to appear on the branch (configurable via
--timeout); sleeps 5s first so GitHub has time to register the run - Watches until done: polls
gh run view --json status,conclusionevery 10s — a transient API error delays detection by one poll instead of crashing the watcher (which is whatgh run watchdoes on a TCP reset) - Reports failure logs: on failure, fetches
gh run view --log-failed(last 3000 chars) - Cleans up: always removes sentinel, even on error
Exit Codes
0— CI passed (or another monitor already active for this push)1— CI failed (logs printed)2— indeterminate: no run found, watch timed out, or gh API kept erroring. NOT a pass — report the manual-check command from the output.
Requirements
ghCLI authenticated- GitHub Actions workflows in the repo