Design multi-agent teams by deriving roles from domain constraints (process maps, policies, rules) rather than from LLM capabilities. Part of the fips-agents ecosystem.
A developer runs /design-team in Claude Code and enters an iterative dialog that walks through:
- Domain constraint ingestion — BPMN process maps, policy documents, compliance rules, or a plain-language walkthrough of the process
- Process mapping — stages, decision points, handoffs, data flows, exception paths
- Role derivation — candidate roles derived from process gaps, not from LLM capabilities
- Trust and scrutiny profiling — reversibility, impact, data sensitivity mapped to runtime config
- Output generation — role specs, agent.yaml stubs, compatibility matrix, team manifest
Each role must pass the "removal test": if removing it doesn't degrade anything, the role doesn't carry its weight.
fips-agents create agent-team my-team
cd my-team
pip install -e .Then run /design-team in Claude Code to start the interactive design session.
The core logic is also available as a standalone Python package:
pip install agent-team-designCLI commands:
agent-team-design parse-bpmn process.bpmn # Parse BPMN to JSON
agent-team-design generate team-design/my-team # Generate stubs, matrix, manifest- docs/concept.md — Core concept: organizational design for agent teams
- planning/design-notes.md — Design decisions and rationale
- fips-agents/agent-template — Agent templates (agent loop + agentic workflow)
- fips-agents/mcp-server-template — MCP server template
- fips-agents/fips-agents-cli — CLI tool for scaffolding fips-agents projects
Apache License 2.0. See LICENSE.