Cursor Action runs your coding agent inside the workflows you already have.

review.yml

- name: Run Cursor Agent
  id: cursor
  uses: PunGrumpy/cursor-action@v1
  with:
    api-key: ${{ secrets.CURSOR_API_KEY }}
    prompt: "Review the changes and summarize the risks."

Job summary

Status: Success
Exit code: 0

The response also lands in outputs.summary, ready for the next step to comment, gate, or ignore.

What you get

The answer is a value, not a log line

steps.<id>.outputs.summary holds whatever the agent wrote back. Comment it on the pull request, append it to the job summary, or gate what runs next. The action does not decide for you.

See the examples

comment.yml

- name: Comment the review
  uses: actions/github-script@v8
  with:
    script: |
      github.rest.issues.createComment({
        issue_number: context.issue.number,
        owner: context.repo.owner,
        repo: context.repo.repo,
        body: `${{ steps.cursor.outputs.summary }}`,
      })

One step, one output

Send a prompt, read the response from steps.<id>.outputs.summary, and pipe it wherever the workflow needs it.

Read the quickstart

Ubuntu, Windows, macOS

Every push to main runs the action on all three runners, so the platform you build on is the one it was tested on.

See how it runs

Documented honestly

The reference tables are generated from action.yml, and the inputs that do not work yet say so on the page.

Open the reference

What it does not do yet

permissions is accepted but never enforced. read-only will not stop the agent editing files or running shell commands.
A run that times out, or fails inside the agent, still reports success. Do not gate a merge on it yet.

Read exactly how it behaves