How are 'overloaded_error' events handled in a stream?

Since these errors can occur even after a stream has successfully started (returning HTTP 200), the skill includes logic to detect error-type events within the stream and trigger appropriate retry mechanisms with backoff.

How does this skill handle API timeouts?

It provides specific configurations for HTTP clients, recommending read timeouts of at least 60 seconds to accommodate the potentially long duration of streaming responses from complex queries.

Why shouldn't I parse JSON mid-stream during tool use?

Streaming delivers partial JSON chunks that are syntactically incomplete. Attempting to parse them immediately will cause errors; you must buffer the chunks and parse the complete string only after receiving the 'content_block_stop' event.

What is the benefit of streaming extended thinking?

Streaming thinking allows users to see the model's internal reasoning process as it occurs, providing transparency and maintaining user engagement during high-latency reasoning tasks.

Real-Time Claude Streaming

Name: Real-Time Claude Streaming
Author: markus41

bymarkus41

•

API 开发

Implements Server-Sent Events (SSE) to enable real-time, progressive response streaming for Claude API integrations.

This skill provides a comprehensive framework for handling Claude's streaming capabilities, allowing developers to build highly responsive interfaces that display text, tool calls, and extended thinking as they are generated. It includes robust patterns for managing the complex SSE event flow, safely buffering partial JSON from tool outputs to avoid parsing errors, and implementing advanced error handling for connection stability. By providing standardized implementations for both Python and TypeScript, it ensures efficient resource management and a superior user experience for long-running AI generations and complex reasoning tasks.

主要功能

01Robust error handling with exponential backoff for retriable API errors

02Server-Sent Events (SSE) integration for real-time progressive output

03Optimized connection management including keep-alive and pooling configurations

04Safe tool use streaming with partial JSON buffering and late parsing

059 GitHub stars

06Extended thinking visualization for monitoring deep reasoning models

使用场景

01Implementing progressive tool execution where inputs are visualized as they form

02Building interactive chat interfaces with immediate, low-latency feedback

03Monitoring long-running complex reasoning or coding tasks in real-time

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add markus41/claude streaming

For use in Claude.ai and ChatGPT