Progressive disclosure index for all specialized agents in the AI Product OS.
Each agent has a dedicated definition file in /agents/ with full role instructions, responsibilities, and constraints.
| Agent | Role | Definition |
|---|---|---|
| Research | Validates ideas, explores problems and market feasibility | research-agent.md |
| Product | Converts validated ideas into clear product specifications | product-agent.md |
| Design | Defines UX structure before engineering work begins | design-agent.md |
| Backend Architect | Designs technical system architecture before implementation | backend-architect-agent.md |
| Database Architect | Designs data model before backend implementation | database-architect-agent.md |
| Frontend Engineer | Implements user interface based on design specification | frontend-engineer-agent.md |
| Backend Engineer | Implements backend services based on architecture and schema | backend-engineer-agent.md |
| Deslop | Cleans and polishes AI-generated code before code review | deslop-agent.md |
| Code Review | Reviews code for violations before acceptance | code-review-agent.md |
| Peer Review | Performs adversarial architecture review of implementation | peer-review-agent.md |
| QA Testing | Validates system reliability before release | qa-agent.md |
| Metric Plan | Defines measurement plan before product ships | metric-plan-agent.md |
| Deploy | Ensures system is safe to deploy to production | deploy-agent.md |
| Analytics | Defines how product usage will be measured | analytics-agent.md |
| Learning | Improves the system after each iteration via postmortems | learning-agent.md |
| Agent | Role | Definition |
|---|---|---|
| Documentation | Maintains clear documentation (CODEBASE-CONTEXT.md) | docs-agent.md |
| Linear | Syncs repo workflow state into Linear for PM use | linear-agent.md |
Agents are activated by their corresponding command. Never activate an agent outside its designated pipeline stage. See CLAUDE.md for the command-to-agent mapping.
The AI Product OS is designed for native portability across AI-assisted engineering tools. To maintain system integrity, follow the MCP Runtime Matrix when configuring tool-specific integrations.
- Durable Context: Always prefer repo-local context (AGENTS.md, CLAUDE.md, project-state.md) over tool-specific memory.
- Executable Validation: Use Bun for all script execution. Enforce standards via
bun run validatebefore accepting agent work. - Bounded Exploration: Use specialized subagents for high-volume or speculative tasks to preserve the main session context.