// your software
Agents
Agents take several roles in a security factory. Name the role whenever more than one appears in the same workflow.
Available security workers
| Security worker | What it does for agents |
|---|---|
| Red Team | Tests deployed agents for data, policy, tool, and approval failures |
| Agent Guardrails | Checks external context and monitors or blocks risky actions where agents run |
Agent roles
| Agent role | Meaning |
|---|---|
| Protected AI agent | A deployed agent tested or guarded by the factory |
| Customer coding agent | A destination that investigates findings and prepares remediation |
| Superagent security agent | An actor that performs built-in security work |
| Customer-provided security agent | A customer component that extends the factory |
Add an agent
- Open Agents and select Add new.
- Enter a name and choose the agent type.
- Optionally select a connected repository to scope later work.
- Select Create agent.
- Open the Agent to configure Red Team settings or Runtime Guardrails.
Coding Agents also receive a one-time Runtime Guardrails pairing command during setup.
Use agents as destinations
Findings can start work in Cursor, Codex, Claude Code, Devin, or another customer agent. The agent reads current context through Superagent MCP, acts within its assigned scope, and returns a structured agent report. People approve consequential changes.