Timmy was exhibiting severe incoherence (no memory between messages, tool call
leakage, chain-of-thought narration, random tool invocations) due to creating
a brand new agent per HTTP request and giving a 3B model (llama3.2) a 73-line
system prompt with complex tool-calling instructions it couldn't follow.
Key changes:
- Add session.py singleton with stable session_id for conversation continuity
- Add _model_supports_tools() to strip tools from small models (< 7B)
- Add two-tier prompts: lite (12 lines) for small models, full for capable ones
- Add response sanitizer to strip leaked JSON tool calls and CoT narration
- Set show_tool_calls=False to prevent raw tool JSON in output
- Wire ConversationManager for user name extraction
- Deprecate orphaned memory_layers.py (unused 4-layer system)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add infrastructure for running swarm agents as isolated Docker
containers with HTTP-based coordination, startup recovery, and
enhanced dashboard UI for agent management.
- Dockerfile and docker-compose.yml for multi-service orchestration
- DockerAgentRunner for programmatic container lifecycle management
- Internal HTTP API for container agents to poll tasks and submit bids
- Startup recovery system to reconcile orphaned tasks and stale agents
- Enhanced UI partials for agent panels, chat, and task assignment
- Timmy docker entry point with heartbeat and task polling
- New Makefile targets for Docker workflows
- Tests for swarm recovery
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add dashboard/store.py: MessageLog dataclass singleton tracking
user/agent/error messages for the lifetime of the server process
- agents.py: write each chat turn to MessageLog; add GET and DELETE
/agents/timmy/history routes returning the history.html partial
- partials/history.html: render stored messages by role (YOU / TIMMY /
SYSTEM); falls back to the Mission Control init message when empty
- index.html: chat-log loads history via hx-get on page start; new
CLEAR button in panel header sends hx-delete to reset the log
- style.css: add .mc-btn-clear (muted, red-on-hover for the header)
- tests: autouse reset_message_log fixture in conftest; 5 new history
tests covering empty state, recording, offline errors, clear, and
post-clear state → 32 tests total, all passing
https://claude.ai/code/session_01KZMfwBpLuiv6x9GbzTqbys
