swe-bench 开源项目推荐（396 个）

smallcloudai /refact

AI Agent that handles engineering tasks end-to-end: integrates with developers’ tools, plans, executes, and iterates until it achieves a successful result.

huggingface /Repo2RLEnv

Turn any repository into verifiable RL environments for coding agents - Harbor tasks you can train on, evaluate and share on the Hugging Face Hub

JARVIS-Xs /SE-Agent

SE-Agent 是一个面向 LLM 代码代理的自我演进框架。它使轨迹级演化能够通过修订、重组和精炼等推理路径交换信息，扩展搜索空间并逃离局部最优。在 SWE-bench Verified 上，它实现了 SOTA 性能

facebookresearch /swe-sweep

How many bugs can LMs find & fix in large codebases?

microsoft /RepoLaunch

[NeurIPS 2026 Main] Automate the build, execution and test of software repositories across programming languages and operating systems.

BichengWang /multi-agents

Multi-agent teams for real work: run teams of coding agents (Claude Code, Codex, any model) in isolated git worktrees, and measure whether the team beats a solo agent on cost and results.

UiPath /coder_eval

Playwright for coding agents. Test that your skills, MCP servers, and CLIs actually work when an agent uses them — sandboxed YAML suites, A/B experiments, CI gates.

hwfengcs /DM-Code-Agent

Local-first, auditable Python code agent. Ships its own 30-task hidden-test benchmark plus SWE-bench Verified scored by the official harness -- every number reproducible, negative results included.

usetig /sage

An LLM council that reviews your coding agent's every move

TIGER-AI-Lab /WebsiteBench

Can AI agents rebuild websites through browser exploration?

vieval-dev /vieval

🧪💥 Evaluation framework based on Vitest, the testing framework you familiar with, for agents, models, and more.

strands-labs /benchmark-harnesses

Strands-based agents and harnesses for agentic benchmarks.

wbopan /retro-harness

RHO: Evolving Agents in the Dark — Retrospective Harness Optimization via Self-Preference. Improving LLM agents from unlabeled past trajectories (arXiv:2606.05922).

FareedKhan-dev /agentic-loop-engineering-course

An 18 notebook course that isolates and measures each component of agentic loop engineering on real, industry standard software datasets.

xianyu-sheng /Xenon

Verifiable AI coding agent harness / 可验证的 AI 编码代理：evidence hash-chain · hollow-answer gate · manual verification checklist · 6 engines + MCP · multi-model · Windows/CJK · listed in awesome-opensource-ai & Collabnix awesome-mcp-lists

Kotlin /kotlin-swe-bench

The official Kotlin benchmark by JetBrains

logic-star-ai /insights

We track and analyze the activity and performance of autonomous code agents in the wild

SihyeonJeon /why-was-fable-banned

Fable-style spec + evidence gate for Claude Code + Codex. Makes Opus/Codex work under Fable-like discipline: blocks every edit until a deterministic spec passes, and there is no "done" without live acceptance evidence. Spec-first, verification-gated, forbidden-paths enforced.

semi-hollow /NanoHarness

Compact AI agent runtime control plane with governed tools, HITL approval, resumable execution, trace-driven evaluation, and SWE-bench-shaped evidence.

aravelo7 /AgentPatchCheck

Coding Agent Runtime & Evaluation Harness for controlled repository-level software repair