Security Agents
Security Agents scan your code for security bugs, risky patterns, and vulnerabilities. They are Cursor-managed agents on Automations, and they require Cloud Agents.
Official: Security Agents.
Two types: Security Reviewer (pull / merge requests) and Vulnerability Scanner (cron). Local skill:
/review-security. Team usage pool — not your personal quota.
Prerequisites
- Cloud Agents available on the account (paid plan; admin has connected source control)
- Access to Automations
- Cursor 3.7+ if you want
/review-security//reviewin the editor or on cursor.com/agents - Team or Enterprise if another product (PR Routing) will consume Security Review Context
Learning objectives
After this page you can:
- Tell Security Reviewer apart from Vulnerability Scanner — and from Bugbot
- Configure a trigger, at least one tool / MCP, and custom instructions
- Run
/review-securityon your branch before you push - Know that usage is billed to the team pool, not to you
Two agent types
| Type | Trigger | Job |
|---|---|---|
| Security Reviewer | Git-based Automations triggers: pull request and merge request events | Catch vulnerabilities during code review |
| Vulnerability Scanner | Cron | Scan the codebase at rest — pre-existing issues, long-standing bugs, things PR review missed |
Both run on the Automations platform. Both need Cloud Agents. You can use Cursor's cloud with no extra VM setup.
This is not Bugbot. Bugbot reviews a PR diff for bugs / security / quality and comments. Security Agents are a separate Cursor-managed automation family. PR Routing can read both: Bugbot Review Context and Security Review Context.
Setup
Open Security Agents in Automations.
- Pick Security Reviewer or Vulnerability Scanner
- Set triggers (PR / MR events vs cron)
- Enable or disable built-in security checks
- Add custom instructions: what to prioritize, project-specific expectations, how the agent should behave
- Attach tools and MCPs — each agent needs at least one tool or MCP to run
- Save. Runs appear in Automations history
Use tools / MCPs to send findings to Slack or an issue tracker, to tell the agent when to call each MCP, and to give extra context before it reports.
Run in your agent
Use /review-security or /review before you push.
| Knob | Official default |
|---|---|
| What diff | Your branch vs the base: committed and uncommitted changes |
| Which base | The default base branch. If that is not main, tell the agent which branch to compare, or let it infer |
Ask it to review only uncommitted changes when you want a narrower pass.
/review and /review-security are available in Cursor 3.7+ and at cursor.com/agents. CLI support is coming soon — do not document a working agent /review-security yet.
Billing
Security Agents are billed at the team usage level:
- Charged to the team's usage pool
- They run under a shared team service account
- They do not consume any individual user's usage
Automations create Cloud Agents; see Cloud agent pricing for the underlying model rates.
Analytics and runs
Three metrics across runs:
| Metric | Meaning |
|---|---|
| Vulnerabilities found | Security findings reported |
| Issues fixed | Findings later resolved |
| Resolution rate | Percentage of reported findings that were fixed |
Cursor uses LLMs on incremental diffs to decide whether a flagged issue was actually fixed.
Every run is tracked in Automations: when it ran, which tools it used, final status, duration. Open a run to inspect the underlying Cloud Agent.
When to use it
- PR gate: Security Reviewer on pull / merge request events
- Recurring baseline: Vulnerability Scanner on a cron, independent of PR traffic
- Local pre-push:
/review-securityon the current branch - Feed PR Routing: enable Security Review Context so low-risk auto-approve waits on these findings
Stay on Bugbot alone when you want a general bug / quality pass, not a dedicated security agent.
Common pitfalls
| Pitfall | Do this |
|---|---|
| Treat this as Bugbot with a new name | Different product. Bugbot comments on a diff; this is a managed Security Agent |
| Forget a tool / MCP | Official: each agent needs at least one to run |
| Expect it on the CLI today | Official: CLI support is coming soon |
| Assume it bills your personal quota | Team usage pool + shared service account |
Ask /review-security for “only the last commit” without saying so | Default is the whole branch, committed and uncommitted |
| Skip Cloud Agents / source-control setup | These agents require Cloud Agents |
Next steps
- Cloud Agents — the runtime
- PR Routing — can wait on Security Review Context
- Cookbook · Bugbot — general PR review
- Official: Security Agents, Automations