yahyaDesoky/InfCode
🤖 Analyze and fix code issues automatically with InfCode, an intelligent Code Agent System achieving a 79.4% solution rate using LLMs.
What's novel
🤖 Analyze and fix code issues automatically with InfCode, an intelligent Code Agent System achieving a 79.4% solution rate using LLMs.
Code Analysis
8 files read · 3 roundsA SWE-bench evaluation harness that runs multiple LLM-powered code-fixing agents in Docker containers (Generator-Selector pattern) to solve programming issues, then selects the best patch from candidates.
Strengths
Well-structured Generator-Selector architecture with clean separation between tools, LLM API layer, and agent loops; the tool system (bash, edit, search) is robust with proper ABC patterns, container support, and persistent shell sessions using a clever sentinel-based output reading approach.
Weaknesses
The README is completely misleading (describes a consumer GUI app when this is an ML research/evaluation tool), there are zero tests, Chinese text is mixed into prompts and logs, and the core agent loop is a straightforward LLM tool-calling pattern derived from Anthropic/ByteDance open-source code rather than a novel algorithm.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
19
Contributors
3
Files
63
Active weeks
3
Repository
Language
Python
Stars
3
Forks
1
License
MIT