make-ci-green

Make CI Green

Re-run failed GitHub Actions CI jobs for a PR. Automatically detects the failure type and reruns only the failed jobs within a workflow run.

---

When to Use

• Flaky test failures on CI that just need a retry
• Build/infra failures that may recover on a clean re-run
• After pushing new commits to retrigger only the failing checks
• Batch retriggering when multiple workflow runs are failing

---

Prerequisites

• gh CLI must be authenticated (gh auth status)
• The current directory must be inside the repository (or specify --repo)

---

The Job

Step 1: Parse Arguments

Extract the PR number from the user's input. The PR number is required. Check for --dry-run flag.

User says	PR number
/make-ci-green 482	482
"retry ci for 482"	482
"make ci green on PR 482"	482
"re-run failed jobs 482 --dry-run"	482 (dry run)

Step 2: Detect Repository

Auto-detect the current repository from the git remote:

gh repo view --json nameWithOwner --jq '.nameWithOwner'

This gives you the OWNER/REPO needed for all subsequent commands.

Step 3: Get PR Branch and Failed Runs

Fetch the PR's head branch and find failed workflow runs:

# Get the PR branch name
PR_BRANCH=$(gh pr view <pr-number> --json headRefName --jq '.headRefName')

# List workflow runs for that branch, showing failed ones
gh run list --branch "$PR_BRANCH" --status failure --json databaseId,name,conclusion,event,headBranch,createdAt --limit 20

Also check for runs still in progress:

gh run list --branch "$PR_BRANCH" --status in_progress --json databaseId,name,conclusion,event,headBranch,createdAt --limit 10

Step 4: Analyze Failures

For each failed run, get the job-level details to classify the failure:

gh run view <run-id> --json jobs --jq '.jobs[] | select(.conclusion == "failure") | {name, conclusion, steps: [.steps[] | select(.conclusion == "failure") | {name, conclusion}]}'

Classify each failure:

Failed Step Pattern	Type	Rationale
checkout, setup, install, build, compile, configure, bootstrap, deps, cache	Build	Infrastructure/build failure
test, spec, check, lint, audit, storybook, e2e	Test	Test failure, likely flaky
Unknown	Unknown	Safe to retry

Step matching is case-insensitive substring matching.

Step 5: Present Findings

Show the user what was found:

• Failed runs: run name, failed jobs, failed steps, failure type
• In-progress runs: listed for awareness
• No failures: report that CI is already green

Example output:

PR 482: 2 failed workflow run(s)

  [FAILED] CI Tests (run 12345678)
       Job: test-linux
       Failed step: Run tests
       Type: test failure (likely flaky)

  [FAILED] Build (run 12345679)
       Job: build-macos
       Failed step: Compile project
       Type: build failure

  [IN PROGRESS] Lint (run 12345680)
       Still running, skipping.

Step 6: Analyze Test Failures (for test-type failures)

For test failures, pull the logs to identify specific failing tests:

gh run view <run-id> --log-failed 2>/dev/null | tail -200

From the logs, try to extract:

1. Specific test names that failed (look for common patterns: FAILED, FAIL:, Error in, AssertionError, etc.)
2. Whether the failure looks flaky (timing-related, network-related, intermittent keywords)
3. Whether the failure could be PR-related (check if PR changes touch files in the same area as the failing test)

Present per-test analysis to the user when available.

Step 7: Confirm and Trigger

If --dry-run, stop here after presenting findings.

Otherwise, ask the user to confirm, then rerun only the failed jobs:

gh run rerun <run-id> --failed

This reruns only the failed jobs within the run, not the entire workflow.

Report the results: which runs were retriggered, and provide links:

# Get the URL for the rerun
gh run view <run-id> --json url --jq '.url'

---

Failure Classification Summary

Failed Step Keywords	Action	Rationale
checkout, setup, install, build, compile, configure, deps, cache, fetch	Rerun	Build/infra failure; retry may fix
test, spec, lint, audit, e2e, check	Rerun	Likely flaky; retry may pass
Unknown	Rerun	Safe default

---

Usage Examples

# Dry run: analyze without triggering
/make-ci-green 482 --dry-run

# Trigger reruns for all failing checks
/make-ci-green 482

# Natural language
"retry ci for PR 482"
"make ci green on 482"

---

Exit Behavior

• If no failed runs are found, report that CI is already green and stop.
• If --dry-run, present analysis and stop without triggering anything.
• If runs are retriggered, report which ones and provide URLs.

---

Limitations

• Only handles GitHub Actions workflow runs
• Cannot retrigger third-party CI systems (e.g., external services reporting status checks)
• gh run rerun --failed requires the run to have completed (not in progress)
• Log analysis for test failures depends on output format varying by test framework
• Does not auto-file issues for flaky tests (only suggests)