What problem does it solve?
AI coding agents often stop prematurely during complex, multi-step workflows, leaving tasks unfinished, errors unresolved, and uncommitted code at risk of being lost before it can be pushed to remote repositories.
Core Features & Use Cases
- Stop Hook Interception: Automatically triggers when the agent attempts to end a session, scanning for signs of incomplete work before allowing a stop.
- Comprehensive Incomplete Signal Detection: Checks for pending/in-progress tracked tasks, recent tool errors, uncommitted git worktree changes, unverified LLM prompt edits, and active beads issues to catch all loose ends.
- Built-in Safety Guardrails: Includes configurable loop protection to prevent infinite continuation cycles, worktree safety checks to avoid losing unpushed code, and AI verification gating for edits made via LLM prompts.
- Use Case: Ideal for long debugging sessions, multi-step feature implementation, deployment workflows, and any task tracking setup where you need guaranteed completion of all planned work before the agent stops.
Quick Start
Use the taskmaster skill to configure a stop hook that prevents Claude from ending sessions until all planned tasks, pending errors, and uncommitted work are fully resolved.
