What problem does it solve? Long-running, multi-step tasks overwhelm AI agents in large projects because context windows fill up and sessions get interrupted, losing all progress. This Skill persists task state in markdown files so work survives session resets and context limits. ## Core Features & Use Cases - Structured Task Files: Creates CURRENT_DESIGN.md, PLAN.md, TODOs.md, STEP-<N>.md, and KNOWLEDGE.md under .agents/tasks/<task name>/ to track design, plans, and progress. - Session Resumption: Pick up exactly where you left off in a brand-new session by reading the step files and notes. - Plan Verification & Pause Handling: Re-evaluate plans mid-task, add TODOs to existing tasks, and formally pause when user intervention (credentials, sudo commands) is required. - Use Case: You need to refactor user creation across a large codebase. Create a task, let the agent plan it into sequential TODOs, then execute one step per session over several days without losing context. ## Quick Start Read the tasks skill and begin task: my-task-name to start executing the planned steps for that task.