How bp974/safe-agent works
Safe Agent isolates coding agents in Docker against safe project copies, then uses reviewed Git commits to transfer changes to and from real repositories. The README documents the workflow and integrations; the only sampled implementation is the project-config validator, so command-to-command internals are not inferred.
Main entry points: agent-start Start agent · agent-pull Pull changes · agent-push Push changes · Dockerfile Docker container
Components 14
Project workflow
- Real repositories
- Safe project copies
- Pull changes
bin/agent-pull - Push changes
bin/agent-push
Agent isolation
- Start agent
bin/agent-start - Docker container
container/Dockerfile - Coding agents
container - Persistent agent state
container
Runtime configuration
- Project config
- Config validator
bin/read-safe-agent-config.py - Local settings
.env.example - Extra host mounts
bin/read-safe-agent-config.py
Optional integrations
- Fusion 360 integration
docs/integrations/fusion-360-wsl-mcp.md
Other components
- Developer
Connections 17
- Developer → Pull changes: runs
- Pull changes → Real repositories: reads commits
- Pull changes → Safe project copies: updates commits
- Developer → Start agent: runs
- Start agent → Safe project copies: mounts project
- Start agent → Docker container: starts
- Docker container → Coding agents: runs selected
- Docker container → Persistent agent state: persists state
- Coding agents → Safe project copies: edits and commits
- Developer → Push changes: runs and confirms
- Push changes → Safe project copies: fetches commits
- Push changes → Real repositories: fast-forwards
- Start agent → Local settings: reads settings
- Start agent → Project config: loads project config
- Project config → Config validator: validated by
- Start agent → Extra host mounts: configures mounts
- Coding agents → Fusion 360 integration: optional integration