Agent Skills
› NeverSight/learn-skills.dev
› error-coordinator
error-coordinator
GitHub专注于构建高韧性多智能体系统，专长于检测循环、幻觉及故障，实现自愈工作流、重试策略、熔断机制和状态恢复，确保AI流程的稳定性与容错能力。
Trigger Scenarios
设计多智能体系统的错误处理机制
实现重试与恢复策略
构建具备自愈能力的AI工作流
检测智能体循环或无限递归
减轻智能体输出中的幻觉现象
实施智能体熔断器以应对级联故障
Install
npx skills add NeverSight/learn-skills.dev --skill error-coordinator -g -y
SKILL.md
Frontmatter
{
"name": "error-coordinator",
"description": "Expert in making multi-agent systems resilient. Specializes in detecting loops, hallucinations, and failures, and implementing self-healing workflows. Use when designing error handling for agent systems, implementing retry strategies, or building resilient AI workflows."
}
Error Coordinator
Purpose
Provides expertise in building resilient multi-agent systems with robust error handling, failure detection, and recovery mechanisms. Covers loop detection, hallucination mitigation, and self-healing agent workflows.
When to Use
- Designing error handling for agent systems
- Implementing retry and recovery strategies
- Building self-healing AI workflows
- Detecting agent loops and infinite recursion
- Mitigating hallucinations in agent outputs
- Implementing circuit breakers for agents
- Coordinating failure recovery across agents
Quick Start
Invoke this skill when:
- Designing error handling for agent systems
- Implementing retry and recovery strategies
- Building self-healing AI workflows
- Detecting agent loops and infinite recursion
- Coordinating failure recovery across agents
Do NOT invoke when:
- Organizing agent teams (use agent-organizer)
- Debugging application errors (use debugger)
- Handling production incidents (use incident-responder)
- Detecting code error patterns (use error-detective)
Decision Framework
Error Type Handling:
├── Transient failure → Retry with backoff
├── Rate limiting → Backoff + queue
├── Invalid output → Validation + retry with feedback
├── Loop detected → Break + escalate
├── Hallucination → Ground with context, retry
├── Agent timeout → Cancel + fallback
└── Cascading failure → Circuit breaker
Recovery Strategy:
├── Idempotent operation → Simple retry
├── Stateful operation → Checkpoint + resume
├── Critical path → Fallback agent
└── Best effort → Log + continue
Core Workflows
1. Loop Detection System
- Track agent invocation history
- Detect repeated state patterns
- Set maximum iteration limits
- Implement escape hatch triggers
- Log loop occurrences for analysis
- Escalate to supervisor or human
2. Hallucination Mitigation
- Ground responses with source data
- Implement output validation
- Cross-check with retrieval
- Add confidence scoring
- Flag low-confidence outputs
- Provide feedback for retry
3. Circuit Breaker Implementation
- Track failure rates per agent
- Define failure threshold
- Open circuit on threshold breach
- Provide fallback behavior
- Implement half-open state for testing
- Close circuit on recovery
- Monitor and alert on breaker state
Best Practices
- Implement timeouts for all agent calls
- Use exponential backoff with jitter
- Log all failures with full context
- Design for graceful degradation
- Test failure scenarios explicitly
- Monitor error rates and patterns
Anti-Patterns
| Anti-Pattern | Problem | Correct Approach |
|---|---|---|
| Infinite retries | Resource exhaustion | Max retry limits |
| Silent failures | Hidden problems | Log and alert |
| No timeouts | Hung processes | Always set timeouts |
| Same retry interval | Thundering herd | Exponential backoff |
| No fallbacks | Complete failure | Graceful degradation |
Version History
- e0220ca Current 2026-07-05 21:13
