Agent skill · software engineering · jeremylongshore
clay-reliability-patterns
Build fault-tolerant Clay integrations with circuit breakers, dead letter
Why this skill is useful
Adds executable scripts for implementing fault tolerance in Clay integrations, including circuit breakers and dead letter queues.
What it needs
Requires redis installed locally. About 5k tokens when loaded. Last updated 2026-08-07. 2,604 stars on the source repository.
What this skill does
Clay Reliability Patterns Overview Production reliability patterns for Clay data enrichment pipelines. Clay's async enrichment model, credit-based billing, and dependency on 150+ external data providers require specific resilience strategies: credit budget circuit breakers, webhook delivery tracking, dead letter queues for failed batches, and graceful degradation when Clay is unavailable. Prerequisites Clay integration in production or pre-production Redis or similar for state tracking Understanding of Clay's async enrichment model Monitoring infrastructure (see clay-observability) Instructions Step 1: Credit Budget Circuit Breaker Stop processing when credit burn exceeds budget to prevent runaway costs: Step 2: Dead Letter Queue for Failed Submissions Step 3: Webhook Health Monitor Step 4: Graceful Degradation When Clay Is Down Step 5: Combine All Patterns Error Handling Issue Cause Solution ------- ------- ---------- Runaway credit spend No budget circuit breaker Implement credit budget limiter Lost leads during outage No DLQ Queue failed submissions for retry Silent webhook failures No health monitoring Track success/failure rates Clay outage blocks pipeline No fallback Implement cache + queue fallback Output Produce a reliability receipt with incoming workload scope, idempotency key, queue/DLQ state, circuit and credit status, retry count, health result, operator decision, and recovery evidence. A queued or acknowledged request is not complete until its downstream processing and required audit state are confirmed. Examples Simulate a staging webhook outage, prove that the circuit opens and the lead is stored once in the DLQ without exposing its raw data in logs, then restore the endpoint and replay through the reviewed worker. If a replay fails validation, leave it quarantined for operator review rather than repeatedly retrying it. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills jeremylongshore/clay-reliability-patterns