Autonomous Development Loop using OODA (Observe, Orient, Decide, Act)
- Launch Claude Code in this directory
- Read
NEXT_STEPS.mdfor the implementation plan - Start with: "Implement Phase 1: Project Setup"
Build an AI system that can autonomously:
- Read code and run tests
- Learn from errors and previous fixes
- Generate working code patches
- Apply patches and verify success
- Iterate until all tests pass
✅ Phase 6 Complete - All phases implemented and validated (91 tests passing)
- Phase 1: Project Setup
- Phase 2: Orient Phase - Context Memory (18 tests passing)
- Phase 3: Decide Phase - LLM Integration (24 tests passing)
- Phase 4: Act Phase - Patch Application (25 tests passing)
- Phase 5: Full Loop Integration (14 tests passing)
- Phase 6: Testing & Validation (10 tests passing)
| Phase | Module | Tests | Status |
|---|---|---|---|
| 1 | Project Setup | - | ✅ Complete |
| 2 | Orient Phase | 18 | ✅ All passing |
| 3 | Decide Phase | 24 | ✅ All passing |
| 4 | Act Phase | 25 | ✅ All passing |
| 5 | Loop Integration | 14 | ✅ All passing |
| 6 | Validation | 10 | ✅ All passing |
| Total | 91 | ✅ All passing |
# Run AutoDev with defaults
python autodev.py
# Fix a specific file
python autodev.py -f test.py
# Limit iterations
python autodev.py --max-iterations 5
# Verbose output
python autodev.py --verbose
# Require human approval for patches
python autodev.py --approve
# Watch progress
python autodev.py --watchOBSERVE → ORIENT → DECIDE → ACT → (repeat)
↓ ↓ ↓ ↓
Code Context LLM Patches
Tests Memory Verify
Errors
- NEXT_STEPS.md - Full implementation plan
- docs/ - Detailed documentation (coming soon)
- mal-python-impl - Mal interpreter (Steps 0-3)
- HeathAudio/.claude/skills/byterover - Context tree and GLM integration
Created: 2025-02-15 Location: C:\dev\AutoDev\