Each agent is a specialist with its own personality, skills, and work loop. They collaborate autonomously to turn your idea into a production-ready app.
The orchestrator. Routes tasks to the right agent, manages sprint execution, resolves conflicts, and ensures the team delivers on time. Every message flows through the Conductor first.
The planner. Classifies projects, selects quality standards, breaks user requests into sprint tasks, and defines acceptance criteria. Thinks three steps ahead.
Web and deep research. Searches the internet, reads documentation, evaluates APIs, and produces structured research reports for the team.
Translates business requirements into technical specifications. Identifies stakeholders, maps workflows, and writes user stories with measurable acceptance criteria.
Designs system architecture. Selects technology stacks, defines service boundaries, plans data models, and produces architecture decision records.
Integration specialist with access to 775+ connectors. Connects to APIs, databases, SaaS tools, and file sources. Handles authentication, pagination, and rate limiting.
Transforms and analyzes data. Writes SQL queries, builds aggregations, creates visualizations, and produces statistical summaries from collected datasets.
Visual design specialist. Creates component layouts, color schemes, typography systems, and responsive design tokens.
React frontend developer. Builds UI components, pages, routing, state management, and client-side logic.
Express API developer. Creates REST endpoints, database schemas, middleware, authentication, and server-side business logic.
Python specialist. Data pipelines, ML scripts, automation, FastAPI services, and scientific computing workloads.
Can install anything in the sandbox. Docker, databases, tools, runtimes. Manages environment setup, CI/CD pipelines, and deployment configurations.
Automation scheduler. Manages recurring tasks, health checks, data syncs, and timed workflows. Cron jobs added at runtime without restarts.
Monitors connector health. Tests API connections, validates credentials, handles OAuth token refresh, and reports integration status.
BDD testing specialist. Writes test scenarios in Given/When/Then format, executes test suites, and reports coverage.
Code reviewer. Scores deliverables against quality standards, identifies gaps, and pushes work back for revision.
Security specialist. Scans for vulnerabilities, checks secrets exposure, validates auth flows, and enforces security policies.
Performance tuner. Measures load times, optimizes bundle sizes, identifies bottlenecks, and improves Core Web Vitals.
The brain at runtime. Handles live data extraction with useExtract(), form submissions, and dynamic content generation inside deployed apps.
BDD workflow executor. Runs multi-step processes defined in Given/When/Then format. Manages pipeline gates and approval flows.
Outbound communication. Sends emails, SMS, Slack messages, and webhook calls as part of pipeline execution and notifications.
SEO, analytics, and growth strategy. Optimizes content, tracks metrics, configures tracking pixels, and suggests acquisition channels.
Process management. Handles recurring operational tasks, generates reports, monitors SLAs, and coordinates cross-team workflows.
Creates new agents at runtime. When the team needs a specialist that does not exist, the Spawner provisions it with workspace, SOUL, and cron job.
Shared memory manager. Maintains project context, stores lessons learned, manages knowledge base documents, and ensures agents have the context they need.
Error recovery specialist. When something fails, the Resolver diagnoses the root cause, applies fixes, and documents the resolution for future reference.
A SOUL is the system prompt that defines an agent's personality, expertise, decision-making style, assigned skills, and work loop. It is what makes each agent unique.
# SOUL.md — Example: Builder Agent
## Identity
You are Builder, the frontend development specialist.
You write clean React code with TypeScript.
## Personality
Pragmatic. Ship fast, iterate. Prefer libraries over
custom code. Tailwind for styling. Minimal dependencies.
## Skills
- sandbox-canvas (create/update canvas apps)
- sandbox-code-files (read/write source files)
- sandbox-deliverables (publish deliverables)
## Work Loop
1. Read the task and acceptance criteria
2. Check existing code in the project
3. Build components, pages, routing
4. Test in preview, fix issues
5. Mark task complete with deliverable linkDefines communication style, risk tolerance, and decision-making approach unique to each agent.
Lists the sandbox skills (REST API wrappers) the agent is authorized to use.
Step-by-step procedure the agent follows for every task, ensuring consistent quality.
All 25+ agents come pre-configured. Just send a chat message and they start working.