Data-driven autonomous mind
The mind's prompt is now assembled from live system state instead of a static hardcoded priority list.
What changed
- Real state queries —
_build_state_snapshot()queries goal_manager, scheduler, memory_manager, identity_manager, and knowledge DB to build the mind's context. Every field is live data, not generic categories. - Bootstrap behavior — When state is empty (first run, no goals/tasks/history), the mind takes initiative: explores tools, searches for opportunities, creates goals. No more sleeping and waiting for the user.
- Fresh timer on resume — User interaction now resets the wakeup timer to config default (
wakeup_seconds) instead of firing immediately on a stale timer from before the conversation. - Cost tracker fix — Fixed
CostTracker.task_cost→task_totalattribute error that crashed every think cycle after tool execution.
Files changed (4)
core/autonomous_mind.py— New_build_state_snapshot(), async_build_prompt(), bootstrap prompt, timer reset, cost fixdocs/26-AUTONOMOUS-MIND.md— Updated context assembly, prompt design, pause/resume, bootstrap docsCHANGELOG.md— New entryREADME.md— Updated autonomous mind descriptions