- Enhanced `recall` action with optional `query`, `limit`, and `mode` parameters
- When query is provided, uses embedding-based semantic search with graph cascade
- Added system prompt instruction to use memory when asked "do you remember?"
- Updated test script to verify semantic recall functionality
- All 24 tests pass with both Claude and OpenAI providers
Debug socket enhancements:
- set_model:<model> - Switch model (may change provider)
- set_provider:<name> - Switch provider with default model
- trigger_extraction - Force end-of-session memory extraction
- available_models - List all available models
- destroy_session:<id> - Clean up and remove sessions
- create_session:<path> - Create session with working directory
Agent changes:
- Make extract_session_memories() public and return count
- Enable programmatic testing of memory extraction
Documentation:
- Update CLAUDE.md with new debug socket commands
- Add "Testing with Multiple Providers" example
- Remove deprecated --standalone reference
Testing:
- Add scripts/test_memory.py for comprehensive memory testing
- Tests all memory features with both Claude and OpenAI
- 20/20 tests passing
1jehuang