AI Agent that handles engineering tasks end-to-end: integrates with developers’ tools, plans, executes, and iterates until it achieves a successful result.
SE-Agent is a self-evolution framework for LLM Code agents. It enables trajectory-level evolution to exchange information across reasoning paths via Revision, Recombination, and Refinement, expanding the search space and escaping local optima. On SWE-bench Verified, it achieves SOTA performance
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.
Test that your Claude Code skills, MCP servers, and CLIs actually work when an agent uses them — sandboxed YAML suites, activation checks, A/B experiments, CI gates.
An LLM council that reviews your coding agent's every move
Xenon — Agent Harness for AI coding agents. 7 种可替换推理范式 + Evidence Runtime 验证闭环（工具结果是 Evidence，LLM 输出只是 Claim）+ 执行隔离边界 + 可复现评测链路。交互与 SWE-bench 评测共享同一组约束，故 40.0% 实例级通过率（同模型 A/B +6.7pp）对日常使用同样成立。Python，可当 CLI 用也可当库嵌入。
RHO: Evolving Agents in the Dark — Retrospective Harness Optimization via Self-Preference. Improving LLM agents from unlabeled past trajectories (arXiv:2606.05922).
A Kotlin software-engineering benchmark for evaluating coding agents.
Strands-based agents and harnesses for agentic benchmarks.
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.
An 18 notebook course that isolates and measures each component of agentic loop engineering on real, industry standard software datasets.
🧪💥 Evaluation framework based on Vitest, the testing framework you familiar with, for agents, models, and more.
Simple Long Horizon Agent - A simple yet effective AI agent for learning, experimentation, and long horizon work.
SWE-bench for your codebase — mine your merged PRs into local, contamination-free coding-agent benchmarks. Adapters: claude-code, aider (Opus 4.7 / GPT-5.5 / Sonnet 4.6 / Gemini 3.1 Pro).
Microsoft Build AI 2026 National 1st Place. Local codebase memory for AI coding agents: 63.8% Coverage@5, 75.2% Coverage@10, 60-65x lower context.
Squeeze verbose LLM agent tool output down to only the relevant lines
Per-step LLM routing benchmark with 970 static labels, live SWE-bench evaluation, an open data pipeline, and a public leaderboard.
Runtime safety net for LLM agents. Detects token spirals, kills doomed tasks early, tells you exactly why. Rust core, Python SDK. pip install state-harness
Coding agent harness, achieved SWE Bench Lite resolve rate of 52.4%
Add a description, image, and links to the swe-bench topic page so that developers can more easily learn about it.
To associate your repository with the swe-bench topic, visit your repo's landing page and select "manage topics."