v0.9.27
What's New
Added
- Task-start planning: LLM generates a structured step plan before the main agent loop begins, improving task execution quality
- Dialogue state tracking: Multi-turn conversation awareness for smarter completion detection
- ScheduledGoalsTool: New tool for managing recurring/scheduled goals
- display_tool_call helper: Consistent tool call formatting across the codebase
- Terminal re-engagement nudge: Detects stalled terminal sessions and prompts re-engagement
Changed
- Validation state extracted into dedicated module
- Execution state tracking moved to its own module
- Bootstrap phase refactored with task planning integration
- Stopping phase uses dialogue state for smarter completion detection
- Guard thresholds tuned for better stall detection
Fixed
- Dead code warnings for
cfg(not(test))gated task planning - Unused imports in integration tests
Full Changelog: v0.9.26...v0.9.27