Cursor has released the Agent SDK for software teams that want agents handling recurring work across the development lifecycle, including pull request review, issue triage, compliance checks, and incident investigation. It is the framework Cursor uses to build its own managed agents, including Bugbot, Security Reviewer, and Rollouts. Prompts, tools, outside connections, and tests sit in one project, so a team can put agent changes through the same version control and review process as the rest of its software. An agent is a folder of code. The prompt is instructions.md, each file under tools/ is a typed TypeScript action the model can call, skills hold procedures the model loads on demand, and channels define where the agent receives and answers messages. Starter templates cover a security reviewer, an owners-policy approver, a pull request autofixer, and Linear or Jira triagers. After a local playground run, agent-sdk deploy builds the agent, assigns a URL, and starts it on Cursor's infrastructure. The same agent can also run as a single Node process on your own machines. Agents wake on pull request, comment, and CI events from GitHub, GitLab, and Bitbucket, on Slack mentions and direct messages, on webhooks, and on a cron schedule. Follow-ups on a pull request or Slack thread resume the same session. Built-in evals send a fixed message and check which tools ran, with what input, and the shape of the reply. Those checks run locally or in CI, and a GitHub replay command can turn a real pull request into signed webhook payloads so the agent can be tested without waiting for the next push.