agentsop-streaming-output

What problem does it solve?

This Skill prevents long-running LLM/agent runs from feeling stuck or noisy by defining a backend streaming protocol that delivers exactly the right “projection” (final tokens, step updates, and custom progress) to the client over the correct transport.

Core Features & Use Cases

• Audience-focused streaming projections: stream only user-relevant final tokens, expose intermediate step updates, and emit in-tool progress events without flooding the client.
• One-wire multiplexing: combine multiple stream modes (tokens, updates, custom) on a single SSE connection with explicit demux tagging.
• Disconnect policy built-in: choose cancel vs detach+persist when the client vanishes, avoiding zombie runs and token waste while preserving resumability for side-effecting flows.
• Cross-framework mapping: aligns LangGraph stream modes, LangChain astream_events, OpenAI streaming, and Anthropic streaming into one consistent SOP.

Quick Start

Ask it to help you design backend streaming for a long agent run by specifying which framework you use, what you need to stream (tokens/steps/progress), and whether disconnects should cancel or detach.