What problem does it solve? Long-running AI agent work loses all progress when a chat session ends or context limits are hit, forcing expensive reconstruction of decisions, constraints, and next steps. This Skill keeps substantial multi-session work resumable through private local state and checkpoints, without leaking working notes into the repository. ## Core Features & Use Cases - Local Continuity State: Records goals, constraints, complexity, risk, selected skills/tools, verification evidence, and next actions in .agentit/STATE.md and JSON checkpoints. - Checkpoint Commands: Provides agent-facing commands like agentit continuity init and agentit continuity checkpoint to snapshot milestones at meaningful moments. - Resume Protocol: Defines a fresh-session procedure that re-verifies recorded assumptions and explicitly re-selects non-core skills, references, workers, and MCPs instead of trusting stale context. - Use Case: An agent working on a multi-day refactoring hits a context limit mid-task; a new session reads the local state file, verifies the branch and evidence, rebuilds the task decision, and continues from the recorded next actions. ## Quick Start Ask the agent to initialize continuity state for the current substantial task and create a checkpoint before handing off to a new session.
