AI Agent Guide
Quick reference for the built-in agents: what they do, how to trigger them in Cursor/Codex (`@agent`) or Claude in the terminal (`/agent`), and when to pick each one.
How to call agents
Cursor/Codex chat: type `@agent-id` in a message (e.g., `@analyst run a market scan`).
Claude (terminal): prefix the message with `/agent-id` (e.g., `/analyst summarize this doc`).
You can also say “As {agent-id}, ...” in either interface. Agents follow their personas but you stay in control—ask, get an answer, then continue or switch.
Agent roster
Pick the specialist that fits the task—call syntax works in both Cursor/Codex (`@agent`) and Claude terminal (`/agent`).
- ux-expert — UI/UX partner for flows, wireframes, specs, and polishing frontend details. Call with
@ux-expert(Cursor/Codex) or/ux-expert(Claude). Great for turning rough ideas into layouts and checking microcopy/accessibility. - sm — Scrum Master that drafts stories/epics, keeps checklists tight, and clarifies acceptance criteria. Call with
@smor/sm. Use when you need structured backlog items without writing code. - qa — Test Architect who maps requirements to tests, flags risk, and proposes quality gates. Call with
@qaor/qa. Good for test plans, coverage gaps, and risk/impact summaries. - po — Product Owner for backlog shaping, acceptance criteria, and prioritization tradeoffs. Call with
@poor/po. Use when you need user-value framing and crisp acceptance tests. - pm — Product Manager that crafts PRDs, roadmap narratives, and stakeholder comms. Call with
@pmor/pm. Helpful for framing problems, success metrics, and scope choices. - dev — Full-stack developer for implementation guidance, debugging, and refactors. Call with
@devor/dev. Use for code walkthroughs, fixes, and best-practice suggestions. - architect — System design and API strategist for integrations, scalability, and trade studies. Call with
@architector/architect. Great for diagrams-in-words, interface definitions, and migration plans. - analyst — Business/market research and competitive analysis with concise briefs. Call with
@analystor/analyst. Use for market sizing, positioning, and quick evidence-backed summaries.
Tips for good results
Be specific about inputs and outputs (e.g., “3 bullet risk summary” or “draft a PRD outline with 5 sections”).
Stay in one agent until you finish a subtask; switch when the work changes (design → testing → docs).
If you’re unsure which agent fits, start with the orchestrator to route you.