Configuring Agent Types
Each agent type can be configured independently with trust level, cost cap, and capability set.
Agent types
| Agent | Role | Best used for |
|---|---|---|
| 🎭 Composer | Intent shaping | Turning raw intents into acceptance criteria and design stubs |
| 💻 Coder | Implementation | Writing code, implementing features, fixing bugs |
| 🔍 Reviewer | Code review | PRs, quality, patterns, correctness |
| 🔐 Security | Security assessment | OWASP scanning, vulnerability detection |
| 🧪 Tester | Test generation | Unit, integration, and E2E tests |
| 🔬 Researcher | Investigation | Spikes, feasibility, technology comparisons |
| 📋 Compliance | Compliance checking | GDPR, SOC 2, ISO 27001 policy compliance |
| ♿ Accessibility | WCAG compliance | WCAG audit, aria fixes, keyboard navigation |
| ⚡ Performance | Optimisation | Profiling, bundle size, runtime performance |
Trust levels
- 1 — Minimal: Agent provides suggestions only; human implements
- 2 — Low: Plan Gate required; human reviews every step
- 3 — Medium: Plan Gate required; execution autonomous; human reviews output
- 4 — High: Plan Gate optional; agent runs end-to-end
- 5 — Full: Fully autonomous; human receives completed output only
Start all agents at trust level 2. Increase only after reviewing at least 10 runs for that type and finding output quality consistently acceptable.
Capabilities
Each agent has individually toggleable capabilities (e.g. "Write code", "Delete files", "Run tests"). Disable any capability inappropriate for your context — for example, "Delete files" can be disabled to prevent agents from removing source files.