Error Coordinator

What it does

The Error Coordinator skill focuses on building resilient multi-agent systems by providing expertise in error handling, failure detection, and recovery. It helps design robust workflows that can detect agent loops, mitigate hallucinations (incorrect or fabricated information), and implement self-healing mechanisms to recover from failures. This skill is specifically designed for managing errors within AI agent interactions, not general application debugging or incident response.

When to use it

• Designing error handling strategies for complex agent systems.
• Implementing retry and recovery processes when agents encounter issues.
• Creating self-healing workflows that automatically recover from failures.
• Detecting and preventing agent loops (repeated, unproductive cycles).
• Mitigating hallucinations in agent outputs by grounding responses with context.

Key capabilities

• Loop Detection System: Tracks agent invocation history to identify and break repeated state patterns, implementing maximum iteration limits and escalation triggers.
• Hallucination Mitigation: Grounds responses with source data, implements output validation, and adds confidence scoring to reduce inaccurate information.
• Circuit Breaker Implementation: Monitors failure rates, defines thresholds, and provides fallback behavior when agents fail repeatedly.
• Error Type Handling: Provides strategies for transient failures (retry with backoff), rate limiting (backoff + queue), invalid outputs (validation + retry), loop detection (break & escalate), hallucinations (grounding & retry), agent timeouts (cancel & fallback), and cascading failures (circuit breaker).

Example prompts

• "How can I prevent an agent from getting stuck in a loop?"
• "What's the best way to handle an agent timeout error?"
• "Suggest a strategy for mitigating hallucinations when this agent is generating text."

Tips & gotchas

• Not for all errors: Do not use this skill for organizing agent teams, debugging application code, handling production incidents, or detecting code-level errors. Use the appropriate specialized skills instead (agent-organizer, debugger, incident-responder, error-detective).
• Timeouts are essential: Always implement timeouts for all agent calls to prevent hung processes and resource exhaustion.
• Backoff strategies: Utilize exponential backoff with jitter when implementing retry mechanisms to avoid "thundering herd" problems (overwhelming a system with simultaneous requests).