What makes jin-swe-fix different from standard Claude coding?

It uses a specialized 4-agent pipeline (Analyst, Fixer, Verifier) that separates thinking from execution and uses independent verification to ensure higher accuracy and less developer bias.

Does it require a GitHub URL to work?

No, while it integrates seamlessly with GitHub to fetch issue details, you can also provide a plain text description of the bug or task directly.

How does the independent verification step work?

A dedicated swe-verifier agent receives only the original issue and the resulting git diff. By excluding the fixer's thought process, it provides an objective assessment of whether the solution truly addresses the problem.

Can it handle complex issues spanning multiple files?

Yes, the skill includes a 'high complexity' path that routes tasks to Claude 3 Opus for deep reasoning and cross-module tracking when three or more files are affected.

SWE Fix (Live-SWE-agent)

Name: SWE Fix (Live-SWE-agent)
Author: cjinzy

bycjinzy

•

開発者ツール

Automates the software engineering lifecycle through a multi-agent pipeline inspired by top-ranking SWE-bench workflows.

jin-swe-fix implements a sophisticated 4-agent orchestration based on the Live-SWE-agent methodology to systematically resolve software bugs and feature requests. It coordinates specialized agents—swe-analyst, swe-agent (Sonnet), swe-agent-high (Opus), and swe-verifier—to handle root cause analysis, complexity-based routing, and independent verification. By separating execution from verification and following a rigorous 6-step workflow (Analysis, Reproduction, Fix, Verification, Edge Case Testing, and Completion), this skill ensures high-quality, bias-free code fixes for projects of any scale.

主な機能

01Multi-agent orchestration separating analysis, implementation, and verification

02Independent verification stage to prevent bias and ensure fix integrity

03Native GitHub integration for fetching and parsing issue details via URL or number

042 GitHub stars

05Complexity-based routing between Claude 3.5 Sonnet and Claude 3 Opus for optimal performance

06Automated 6-stage workflow: Analysis, Reproduction, Fix, Verification, Edge-cases, and Completion

ユースケース

01Performing unbiased independent verification of code changes to maintain high software quality standards

02Automating the end-to-end cycle of bug reproduction, unit testing, and patching for GitHub issues

03Resolving complex, cross-module bugs that require deep architectural reasoning and impact analysis

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add cjinzy/jin-claude jin-swe-fix

For use in Claude.ai and ChatGPT