llm-cost-rate-governance

What problem does it solve? Production pipelines that call the Anthropic API can crash or block all analyses when credits are exhausted, rate limits hit, or errors go unhandled. This Skill enforces a governance checklist so every Claude call site handles 429s, retries transient failures, and degrades gracefully instead of surfacing raw provider errors. ## Core Features & Use Cases - Rate-Limit and Credit-Exhaustion Checks: Confirms every Anthropic SDK call catches 429 and insufficient-credit responses explicitly. - Bounded Retry and Degradation Rules: Requires backoff-based retries and a LIMITED-confidence fallback result on exhaustion. - Cost Traceability: Ensures new call sites document model and approximate token volume for later cost review. - Use Case: When adding a new pipeline stage in lib/stage2 that calls anthropic.messages.create(), apply this Skill to verify error handling, retry bounds, and context truncation before shipping. ## Quick Start Review the new Claude API call I just added in lib/thesis-engine and check it against the LLM cost and rate governance checklist.