How mounidar/autonomous-software-engineering-team works
This repository is a small autonomous software-engineering agent prototype. Its implemented workflow starts in main.py: an Architect asks OpenAI for a structured architecture, then a Developer reports selected architecture details and inspects and reads repository files. The Developer also exposes a file-writing method, though the main workflow does not call it. The README describes broader planning, implementation, testing, review, and improvement capabilities; the sampled code does not show those stages implemented, so they are not represented as runtime components.
Main entry points: architect.py Architect agent · main.py Task runner · developer.py Developer agent · filesystem.py Filesystem tools
Components 5
Workflow
- Task runner
main.py
Agents
- Architect agent
agents/architect.py - Developer agent
agents/developer.py
Workspace tools
- Filesystem tools
tools/filesystem.py
Other components
- OpenAI API
Connections 7
- Task runner → Architect agent: requests analysis
- Architect agent → OpenAI API: submits prompt
- OpenAI API → Architect agent: returns parsed plan
- Architect agent → Task runner: returns architecture
- Task runner → Developer agent: passes architecture
- Developer agent → Filesystem tools: lists and reads
- Developer agent → Filesystem tools: writes on request