terminal-bench

Framework for evaluating and improving agents

A harness optimized to smaller LLMs

Long Horizon Terminal Benchmark with Dense Reward Grading

Official Implementation of "CLI-Gym: Scalable CLI Task Generation via Agentic Environment Inversion"

Playwright for coding agents. Benchmark Claude Code, Codex, Gemini, and OpenCode on your own tasks - and test that your skills, MCP servers, and CLIs work when an agent uses them. Sandboxed YAML suites, activation checks, A/B experiments, CI gates.

Some tools help. Some tools assist. Villani Code intervenes.

A research framework for principled agent self-improvement under frozen evaluators and declared mutation boundaries, recording verifiable lineage to make it reproducible and auditable.

Strands-based agents and harnesses for agentic benchmarks.

Simple Long Horizon Agent - A simple yet effective AI agent for learning, experimentation, and long horizon work.

Spoox CLI - Terminal Agent - SPlit lOOp eXand agent

A coding agent that builds it's own tools and mines it's own skills and always ships with receipts.

Fast, Multi-Cloud Sandbox Engine for AI Agents

Evidence-first, crash-resumable self-evolution engine for DeepSeek Harness and Harbor.

Which host should serve your open model? Run your workload across pinned, verified providers and get one table: success, cost, latency, cache hits.

Trajectories for running OpenHands on Terminal Bench

OpenCode TUI plugin for model benchmark scores, token costs, and token efficiency.

Self-owned TerminalBench 2.0 coding-agent harness with a Rust Worker, Harbor/verifier-grounded evaluation, and guarded Heuristic Learning.

Your model, many harnesses, many benchmarks.

Open-source desktop coding agent (Electron + React) with plan/execute modes, MCP, skills, and a terminal-bench 2.0 harness adapter

Central repository for Project Terminus task submissions. These environments and Oracle solutions are designed to evaluate state-of-the-art AI agents (like GPT-5.2 and Claude Opus 4.6) on complex, multi-step engineering challenges within a sandboxed terminal.