LLM API Integration Doctor

The problem

LLM API calls are notoriously brittle. You are facing silent 401s, 429 rate limit storms, truncated JSON responses, or streaming connections that work in local dev but hang indefinitely once deployed to Vercel or AWS.

What it does

• Triages auth failures by verifying environment variable parity between local and production runtimes.
• Identifies token overflow and truncation causes by analyzing max_tokens settings and model stop reasons.
• Debugs streaming bottlenecks, specifically catching buffered chunks and serverless execution timeouts.
• Optimizes request patterns to handle rate limits using exponential backoff instead of fixed-retry loops.

Frameworks & tools

Works with Node.js, Python, and serverless environments calling Anthropic, OpenAI, Groq, Gemini, or OpenRouter via SDKs or raw HTTP.

Why this beats prompting it yourself

Generic LLM debugging often misses infrastructure-level failures like serverless cold-start timeouts or key-format mismatches. This skill uses a specialized protocol to isolate whether a failure is transport, auth, or dependency-related before looking at the code.

Use cases

• Fixing 'Works on my machine' deployment errors in Vercel or Cloudflare Workers.
• Resolving intermittent JSON parsing errors caused by token limit truncation.
• Diagnosing why a streaming UI appears to hang despite the API returning data.
• Implementing proper concurrency caps to prevent 429 errors during batch processing.

Known limitations

This skill does not debug prompt quality or 'hallucinations.' It requires access to the API call site code and environment configuration to be effective.