How skaslam1407/multi-agent-system works
This is a manager-led multi-agent system that accepts a task, optionally researches it, generates Python code, and reviews it; review feedback can trigger another coding pass until approval or the revision limit. The CLI presents the resulting research, code, and review. Supporting layers provide shared workflow state, task validation, permission demonstration, and a retrying LLM client configured for an OpenAI-compatible endpoint. The graph follows README-documented behavior and shown imports/calls; unsampled manager and CLI details are not inferred beyond those references.
Main entry points: llm.py LLM client · manager.py Workflow manager · main.py CLI and results · research_agent.py Research agent
Components 14
CLI entry
- CLI and results
main.py - Task validation
validation.py - Tool permissions
permissions.py
Workflow orchestration
- Workflow manager
agent/manager.py - Shared workflow state
agent/state.py
Task agents
- Research agent
workers/research_agent.py - Coding agent
workers/coding_agent.py - Review agent
workers/reviewer_agent.py
LLM runtime
- LLM client
llm.py - Runtime configuration
config.py - Thinking indicator
animation.py - Agent logging
logger.py
Other components
- Task submitter
- OpenAI-compatible LLM
Connections 15
- Task submitter → CLI and results: submits task
README.md - CLI and results → Workflow manager: runs workflow
main.py - CLI and results → Tool permissions: demonstrates permissions
main.py - Workflow manager → Task validation: validates task
agent/manager.py - Workflow manager → Shared workflow state: creates state
agent/manager.py - Workflow manager → Research agent: calls research
agent/manager.py - Workflow manager → Coding agent: calls coding
agent/manager.py - Workflow manager → Review agent: calls review
agent/manager.py - Research agent → LLM client: requests completion
workers/research_agent.py - Coding agent → LLM client: requests completion
workers/coding_agent.py - Review agent → LLM client: requests completion
workers/reviewer_agent.py - LLM client → Runtime configuration: reads settings
llm.py - LLM client → Agent logging: logs requests
llm.py - LLM client → Thinking indicator: shows progress
llm.py - LLM client → OpenAI-compatible LLM: sends completions
llm.py