What problem does it solve? LLM API calls fail unpredictably due to rate limits, quota exhaustion, and malformed JSON responses, breaking automated pipelines. This Skill provides a structured fallback chain that distinguishes per-minute from per-day limits and routes requests across Groq, Cerebras, Ollama, Anthropic, OpenAI, and Google providers. ## Core Features & Use Cases - Rate Limit Classification: Parses 429 error responses to distinguish per-minute limits (wait and retry) from per-day quota exhaustion (skip to next provider). - Multi-Provider Fallback Chains: Defines per-task model chains (relevance scoring, CV tailoring, rejection categorization, interview prep, DOM vision) with a local Ollama model as the final fallback. - Invalid JSON Recovery: Automatically retries with the next model in the chain when a provider returns malformed JSON. - Use Case: When a Groq model returns a 429 daily quota error during relevance scoring, the dispatcher immediately routes the request to Cerebras, then to a local Ollama model if needed, and sends a Telegram alert if all models are exhausted. ## Quick Start Use the rate-limit-fallback skill to debug why my call_with_fallback function raises AllModelsExhaustedError on the relevance_scoring task.
