Cursor Cloud Agents — launch background coding agents

This skill drives the Cursor Cloud Agents API (v1) programmatically: launch an agent on a connected GitHub repo with a prompt, poll its run, follow up on the same agent, and collect the pull request it opens. It is for delegating well-scoped coding tasks — fixes, docs, tests, small features — to a background agent while you keep working.

The core calls are plain REST against https://api.cursor.com with a Bearer API key from your Cursor account settings, passed via environment variable and never written to files or logs. GET /v1/me is the cheap auth check, GET /v1/repositories lists connected repos, and a launch POST creates the agent and enqueues its first run in one call, returning the agent and run ids every later call needs.

Two rules matter most. The repo must already be connected to the Cursor account — there is no API-side way to connect a new one, so check first. And don't pin a model unless you mean it: omitting it uses the account's default models and included usage pool, while an explicit model can draw from a different, metered pool.

With autoCreatePR, a completed run opens a PR. Agents push branches; humans merge — the skill's rule is to review every PR before it lands and never auto-merge agent output. The API is public beta, so if calls fail in new ways, check the current endpoint docs before blaming the key.