How does it handle slow model response times?

The skill monitors execution and triggers a timeout protocol if a model takes longer than 30 seconds, allowing you to increase the timeout or proceed without that specific result.

Does it work with local or embedded models?

Yes, while designed for external API resilience, it includes logic to fall back gracefully to embedded Claude models if external providers are completely unreachable.

Does this skill automatically switch models if one fails?

By default, it prioritizes User Escalation. It will stop and report the failure to you, offering options to retry, switch, or skip, rather than making silent substitutions.

What happens in a multi-model task if only one model succeeds?

It applies a Partial Success strategy, notifying you of the specific failures while presenting the available results, ensuring the entire workflow doesn't crash due to one provider.

Can it manage API rate limiting (429 errors)?

Yes, it detects 429 status codes, respects 'Retry-After' headers, and implements a wait-and-retry strategy before falling back to alternative workflows.

Resilient Multi-Model Error Recovery

Name: Resilient Multi-Model Error Recovery
Author: MadAppGang

byMadAppGang

•

생산성 및 워크플로

Implements robust error handling and recovery patterns for multi-agent workflows using external AI models.

The error-recovery skill provides a comprehensive framework for managing failures in complex AI workflows, ensuring production-ready resilience for multi-model operations. It handles common distributed system issues such as model timeouts, API authentication failures, rate limiting, and partial successes across multiple providers. Most importantly, it enforces a 'User-First' escalation policy that prevents silent model substitutions, ensuring developers maintain full control over model selection and cost management when external dependencies fail.

주요 기능

013 GitHub stars

02Intelligent retry logic with exponential backoff for transient network and API errors

03User-first escalation protocol to prevent silent or unauthorized model substitutions

04Detailed error reporting providing actionable recovery options and diagnostic data

05Automatic timeout detection and management for models exceeding 30-second thresholds

06Partial success strategies to consolidate results when only a subset of agents complete

사용 사례

01Building production-grade multi-agent pipelines that require high availability and fault tolerance

02Maintaining workflow continuity during transient provider outages or API rate limit hits

03Managing complex code review workflows that aggregate feedback from multiple diverse AI models

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/magus error-recovery

For use in Claude.ai and ChatGPT