anthropic-api-resiliencelisted
Install: claude install-skill mickolasjae/mick-applied-ai-toolkit
# Anthropic API Resilience
Production-grade reliability wrapper for any `@anthropic-ai/sdk` (TS) or `anthropic` (Python) client. Adds three layers: **round-robin key rotation**, **exponential backoff**, and a **circuit breaker** on per-key failure streaks.
## When to use
Trigger this skill whenever the user is:
- Hitting `429 RateLimitError` or `529 overloaded_error` from Claude
- Asking for "Anthropic API retries", "multiple Anthropic keys", "round-robin Anthropic", "production-grade Anthropic client"
- Running an agent loop, batch job, or any long-running pipeline that calls `client.messages.create()` more than ~1/sec
- Reviewing code that calls Anthropic with no retry/backoff wrapper
Do **not** trigger for: simple one-shot scripts, prompt-engineering questions, or non-Anthropic providers.
## Quick assessment (ask this first)
> **How many Anthropic API keys do you have available?**
- **1 key**: round-robin buys nothing. Install the wrapper anyway — backoff + circuit breaker still prevent thundering-herd retries that get you rate-limited harder. Skip the multi-key env-var step.
- **2+ keys**: full payoff. Distinct keys from different Anthropic accounts/orgs give independent rate-limit buckets. Same-account keys share the bucket and only help with key-specific failures (revoked, network, etc.).
## Spec (verified pattern)
Sourced from `RL-Studio-Anthropic/ANTHROPIC_KEY_MANAGEMENT.md`:
| Knob | Value |
|---|---|
| Base backoff | 1 second |
| Max backoff | 5 minutes (3