What problem does it solve?
Integrating stateful, long-running, and non-deterministic LLM agents with conventional web application patterns often leads to blocked event loops, silent failures, duplicate runs, leaked user state, and no traceability; this Skill helps you choose the right integration pattern and discipline to avoid those failures. It provides actionable guidance to map latency and delivery requirements to REST, WebSocket/SSE, Webhook, or Message Queue architectures and prevents common integration bugs caused by using the wrong pattern.
Core Features & Use Cases
- Pattern selection: Decision rules for synchronous REST, async REST + polling, SSE, WebSocket, Webhook, and Message Queue integrations based on initiator and latency requirements.
- Interface design: Concrete API shapes and messaging contracts for run, poll, stream, and cancel flows.
- Reliability & safety: Best practices for offloading blocking calls, timeouts, idempotency, deduplication, state isolation, and health endpoints.
- Security & observability: Signature verification for webhooks, secret management, structured logging, and trace continuity using run_id for correlation.
- Use case examples: A browser client receiving progress via SSE, a webhook that acks within 3 seconds and processes async, and a worker pipeline consuming jobs from a queue with retries.
Quick Start
Design an async webhook or streaming API that acks requests immediately, offloads agent runs to a background worker or thread pool, enforces timeouts and signature verification, and returns a run_id for tracing.
