What problem does it solve? AI agent sessions, sandboxes, and processes can crash or lose context mid-task, and relying on conversation history alone makes recovery impossible. This Skill defines a canonical recovery path that rebuilds ticket work from durable project state instead of fragile session memory. ## Core Features & Use Cases - Durable state recovery: Reconstructs work from GitHub Issues, Projects, branches, Draft PRs, stack predecessors, and structured recovery checkpoints rather than native session state. - Split-brain prevention: Uses execution generations, leases, and fencing tokens with compare-and-set ownership so stale agents cannot perform duplicate integrations or external writes. - Validation and side-effect reconciliation: Re-evaluates validation results against the current code snapshot and reconciles external side effects (deploys, migrations, publishes) against remote actual state before retrying. - Use Case: A sandbox running a multi-hour implementation crashes mid-validation. A fresh agent reads the Issue, PR, and latest checkpoint, acquires a new fencing token, reconciles child subagents, and resumes from the last verified state without duplicating side effects. ## Quick Start Ask the agent to recover the interrupted ticket work for a given GitHub Issue by reading the project state, branches, PRs, and latest recovery checkpoint, then resuming from the validated snapshot.
