gh-retry

Rerun all failed workflow runs: $ARGUMENTS

Step 1: Determine mode and collect PRs

Parse the argument to determine if this is a single PR or all my PRs in a repo:

• Single PR: A PR number (e.g., 123), PR URL (e.g., https://github.com/owner/repo/pull/123), or PR reference (e.g., owner/repo#123)
• Repo-wide: A repo URL (e.g., https://github.com/owner/repo), or owner/repo without a PR number — list all open PRs by the current user:

  gh pr list --repo <owner/repo> --author @me --json number,title,headRefOid
  

Step 2: For each PR, get the current check status

Use gh pr checks to get the authoritative, current state of all checks for the PR. This correctly reflects rerun results, unlike gh run list which can show stale job data from before a rerun completed.

gh pr checks <pr> --repo <owner/repo>

Step 3: Identify failures

Parse the gh pr checks output for checks with status "fail" or "cancelled". If all checks are still "pending" with no failures, check GitHub Actions status before reporting:

curl -s https://www.githubstatus.com/api/v2/components.json | jq '.components[] | select(.name == "Actions") | {name, status}'

If Actions is degraded or has a major outage, inform the user that GitHub Actions is experiencing issues and their runs may be stuck. Otherwise, report that all checks are passing or pending.

Step 4: Extract run IDs from failed checks

Each failed check URL contains the run ID (e.g., https://github.com/owner/repo/actions/runs/12345/job/67890). Extract the unique run IDs from the URLs of all failed and cancelled checks.

Step 5: Display a tree and rerun failed jobs

Always group by PR:

Workflows to rerun:
PR #123 - Add feature X
├── **Build** (`21610828075`)
│   ├── Build - Ubuntu (failure)
│   ├── Build - Windows (failure)
│   └── Build - macOS (failure)
└── **Coverage** (`21610828076`)
    └── Coverage - Windows (failure)
PR #456 - Fix bug Y
└── **Orchestrator** (`21610828077`)
    └── Coverage - Windows (failure)

• Workflow names should be bold
• Run IDs should be in backticks
• Only show jobs with conclusion "failure" or "cancelled" in the tree (these are the ones that will be rerun)

Immediately rerun without asking for confirmation (rerunning is non-destructive):

gh run rerun <run-id> --repo <owner/repo> --failed

Step 6: Report what was rerun