Projects and Agents

Workbench organizes work by workspace, project, and protected agent.

Workspace

A workspace contains projects, members, settings, and billing state.

Workspace members can have roles:

• owner
• admin
• member

Project

A project represents one agent system or product surface under test.

Projects track:

• project name and slug
• risk score
• runs
• findings
• scenarios
• attack-pack coverage
• API keys
• CI setup

Protected Agent

An agent is the target being protected.

Agent fields:

• name
• type: http, cli, browser, or mock
• environment: production, staging, or local
• target
• connected status
• latest risk
• coverage

Creating A Project

Use the onboarding flow in workbench:

1. Create a project.
2. Choose agent type.
3. Set environment and target.
4. Copy the project API key.
5. Run the included CLI locally.
6. Upload sanitized findings.
7. Confirm the first uploaded run appears.

The onboarding flow creates:

• project
• first protected agent
• project API key
• CI setup

Agent Attribution During Uploads

When a project has multiple agents, pass the agent name during upload:

roleplay upload all --agent support-agent-staging

Or set:

ROLEPLAY_AGENT_NAME=support-agent-staging

This keeps findings and risk attached to the correct protected agent.