All jobs

Set Up Dependabot Auto-merge with GitHub Actions

Yes. An agent adds a workflow that approves and auto-merges Dependabot's minor and patch PRs once your checks pass, scoped to the groups you define.

Run this on RushOpens the console with this prompt ready to send, in a Project that has your repository.

What the agent does

The agent first reads how Dependabot is already configured, because auto-merge should follow the grouping the repository has chosen rather than invent one. In a repository like actions/typescript-action, .github/dependabot.yml has two ecosystems on a weekly schedule: github-actions with an actions-minor group for minor and patch updates, and npm with an npm-development group for minor and patch dev dependencies and an npm-production group for patch updates of production dependencies. There is no auto-merge workflow and nothing in the README mentions one. The agent also reads .github/workflows/ci.yml, which runs npm ci, npm run format:check, npm run lint and npm run ci-test on Node from .node-version (24.4.0) and then runs the action itself with uses: ./, and .github/workflows/check-dist.yml, which rebuilds dist/ with npm run bundle and fails if the checked-in bundle differs.

That second workflow decides the design. A TypeScript action ships its compiled dist/index.js in git, so a Dependabot bump of a production dependency such as @actions/core changes package-lock.json but not dist/, and check-dist fails on that pull request until someone rebuilds. Auto-merging it would be wrong, and so would merging a dev-dependency bump that changes the Rollup or TypeScript output. The agent therefore writes .github/workflows/dependabot-auto-merge.yml on pull_request with if: github.actor == 'dependabot[bot]', reads the update with dependabot/fetch-metadata to get update-type and package-ecosystem, and enables auto-merge only for the github-actions ecosystem and for npm updates whose type is version-update:semver-minor or version-update:semver-patch. Production npm dependencies are left for a person, matching the narrower npm-production group the repository already chose.

The merge itself is gh pr merge --auto --squash "$PR_URL" with GH_TOKEN set to secrets.GITHUB_TOKEN, preceded by gh pr review --approve so a branch protection rule that needs one approval is satisfied. The job declares permissions: contents: write and pull-requests: write and nothing else. Because --auto only queues the merge, GitHub still waits for the required checks, ci.yml, check-dist.yml, linter.yml, codeql-analysis.yml and licensed.yml, to pass, and merges only then. The agent pins the fetch-metadata action to a release tag the same way the repository's other workflows pin theirs, and keeps the file free of a pull_request_target trigger, since the job runs no code from the pull request.

Before it pushes, the agent runs actionlint on the new file (the repository already has actionlint.yml for its linter workflow) and checks the YAML against the .yaml-lint.yml rules. It cannot open a Dependabot pull request from the sandbox, so the pull request description says exactly which conditions enable the merge, which repository settings the reader must turn on, namely "Allow auto-merge" in the repository's general settings and a branch protection or ruleset on main that requires the existing checks, and what a Dependabot update of a production npm package will still require by hand.

Start it with one API call

POST https://api.phoenix.so/o/{org}/sessions with this body:

{
  "agent": "claude",
  "input": "Add Dependabot auto-merge to this repository and open a pull request. Read .github/dependabot.yml and every workflow under .github/workflows first, and note that check-dist.yml fails when a dependency bump changes the compiled dist/ output. Write .github/workflows/dependabot-auto-merge.yml that runs only for pull requests by dependabot[bot], uses dependabot/fetch-metadata pinned to a release to read the update type and ecosystem, approves and enables squash auto-merge with gh for github-actions updates and for npm minor and patch updates of development dependencies, and leaves production dependency bumps for a person. Declare only the permissions the job needs. Do not use pull_request_target. Run actionlint on the new file. In the description list the repository settings the reader must enable for auto-merge to work and the cases the workflow deliberately does not merge.",
  "resources": [
    {
      "url": "https://github.com/your-org/your-app"
    }
  ]
}
  1. Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
  2. Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
  3. Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.

What comes back

What comes back is a pull request that adds one file, .github/workflows/dependabot-auto-merge.yml. It triggers on pull_request, gates on the Dependabot actor, reads the update with fetch-metadata, and approves and queues a squash merge for the cases the prompt allows. dependabot.yml, ci.yml and check-dist.yml are unchanged, and no secret beyond the job's GITHUB_TOKEN is needed.

The description quotes the actionlint output from the sandbox, lists the two repository settings to turn on and names the checks the merge waits for. It states plainly which updates still need a person: production npm dependencies, and any update whose checks fail, because --auto never merges a red pull request.

Before merging, a reviewer confirms that "Allow auto-merge" is enabled on the repository, that the ruleset on main requires the five existing workflows so a merge cannot happen before check-dist runs, and that an approval from github-actions[bot] is acceptable under their review policy. If it is not, they remove the approve step and keep the auto-merge, which then waits for a human approval instead.

Run it on every repository with Dependabot and no auto-merge

An organization that maintains many actions or libraries starts one POST /o/{org}/sessions per repository with this prompt and the repository in resources. GET /o/{org}/github lists every repository the installation reaches; a short script clones each, checks for .github/dependabot.yml without a matching auto-merge workflow, and starts a run for those. Each run reads that repository's own groups and checks, so a repository without a check-dist step gets a workflow that also merges production patch updates, and one whose CI is a single job gets a workflow that waits for that job.

Set a webhook URL on the API key and Rush POSTs run.completed or run.failed when each turn ends, with session_url to read the pull request link. Send a fresh UUID as request_id with each request so a retried call returns the run already started.

The use case this job follows, with its automation and trigger: Hand off a long task.

Why run it on Rush?

The hard part of auto-merge is not the YAML, it is knowing which updates are safe to merge in this repository, and that comes from reading its other workflows. In its own sandbox the agent reads dependabot.yml, every workflow and the lockfile, notices that a compiled dist/ makes production bumps unsafe to merge unattended, and encodes exactly that into the conditions. The reasoning is in the pull request, and actionlint output from the sandbox is kept in the run's record.

Done by hand this is a file copied from a blog post and adjusted after the first bad merge. Done on Rush it is one request per repository, each read on its own terms, each ending in a pull request with the settings checklist a reviewer needs.

Frequently asked questions

Why not auto-merge every Dependabot pull request?

Because in a TypeScript action a production dependency bump changes package-lock.json without rebuilding dist/, and check-dist.yml will fail until someone runs npm run bundle and commits the result. Merging those unattended is not possible anyway, and the agent would rather say so than enable a merge that never fires. Repositories without a committed build output can include production patches.

Does this need a personal access token?

No. gh pr review --approve and gh pr merge --auto work with the job's GITHUB_TOKEN when the job has pull-requests: write and contents: write. One caveat: a merge performed with GITHUB_TOKEN does not trigger other workflows on the resulting push, which is usually what you want for a dependency bump.

Will it merge before CI finishes?

No. --auto enables GitHub's auto-merge, which waits for every required check on the base branch's ruleset and merges only when they pass. That is why the description tells you to require ci.yml, check-dist.yml and the other existing workflows on main; without a ruleset there is nothing to wait for.

What about grouped updates?

The workflow reads fetch-metadata's update-type, which for a grouped pull request is the highest type in the group, so a group mixing minor and patch updates reads as minor and still qualifies. The agent keeps the groups you defined in dependabot.yml and does not add new ones.

More GitHub Actions jobs

More jobs like this: Hand off a long task

Keep reading