Agent Task Rescue

Agent Task Rescue turns fragile, multi-step AI-agent work into a bounded, observable job. It defines measurable success before execution, records durable checkpoints, limits attempts, recovery cycles, time, and cost, blocks unsafe duplicate committing actions, redacts secrets from evidence, and verifies results independently before reporting completion. Includes a dependency-free Python ledger, JSON state with atomic writes, a manual Markdown fallback, failure taxonomy, state-transition reference, and automatic task receipts.