error-recovery-retry

What problem does it solve?

This Skill helps production AI agents recover from transient failures and avoid cascading outages by adding structured retry, circuit breaking, and escalation paths when things go wrong.

Core Features & Use Cases

• Failure taxonomy & decisioning: Classifies errors (transient, rate limited, validation, model error, permanent) to decide what to retry versus when to escalate.
• Exponential backoff + jitter retries: Implements bounded retries with delay growth and random jitter to reduce load spikes.
• Circuit breaker + graceful degradation: Prevents repeated calls to failing dependencies and switches capabilities or tool availability under sustained failure.
• State recovery + dead-letter queue: Checkpoints agent state for replay and routes unrecoverable tasks to a DLQ for human review or later replay.

Quick Start

Ask an agent to apply error-recovery-retry patterns to your tool calls and model requests by implementing bounded exponential backoff retries, circuit breakers per dependency, and a dead-letter queue for unrecoverable failures.