ArchestackBest of Vibe Coding: List on GitHub ↗ (opens in a new tab)

#128128th of 143 in Coding agents and AI editors

longhorizon-harness

Manager/Executor/Auditor loop that keeps Claude Code or Codex on long tasks

Website ↗ (opens in a new tab)Repository on GitHub ↗ (opens in a new tab)
Stars
1.7k
License
MIT
Last commit
Aug 2026
Last release
Aug 2026
Language
Python

Missed the latest GitHub check; these facts may be older.

Overview

LongHorizon-Harness wraps existing agent CLIs (Claude Code, Codex, OpenCode, DeepSeek Harness) in a plan, act, verify, checkpoint loop. A Manager picks the next bounded step, an Executor runs it with fresh context, and an Auditor checks real files, UI and logs before progress is accepted. It ships a CLI and a browser workbench, and reports gains on WeaveBench, OSWorld 2.0 and Terminal-Bench 2.1.

Who it is for: Engineers running long autonomous Claude Code or Codex tasks on macOS or Linux

Strengths

  • Separate Auditor inspects actual files, logs and tests before a step counts as progress
  • Manager, Executor and Auditor can each use a different backend or model
  • Resumable runs stored under .lh-harness/runs, with follow-ups continuing the same run
  • README reports Terminal-Bench 2.1 69.7 to 77.2 with Qwen 3.7-Plus on Claude Code

Weaknesses

  • macOS is the only tested platform; native Windows is blocked in v0.1.8, WSL2 needed
  • Requires a separate agent CLI (codex, claude, opencode or dsh) installed and authenticated
  • GUI use needs a computer-use plugin and OS permissions granted by hand on macOS
  • DeepSeek backend is CLI-only, returns only final answer, and exposes task text in process args

What it needs

  • no GPU
  • Needs Python 3.10+, uv or pip, Node.js 20+, codex, claude, opencode or dsh CLI, computer-use plugin (optional)
  • Models: Claude, GPT, Qwen, DeepSeek
  • port 8799

Also in Coding agents and AI editors

See all 143
Also in Coding agents and AI editors
RankProjectScore
1ArchonYAML workflow engine that runs AI coding agents through fixed dev steps23.7k starsMIT84 out of 100
2claw-codeRust CLI coding agent harness using Anthropic and OpenAI-compatible APIs195.1k starsMIT83 out of 100
3Open SWEAgent that turns issues into pull requests, reviews PRs and monitors CI11.1k starsMIT83 out of 100
#4piTerminal coding agent with extensions, skills, and an SDK117k starsMIT80 out of 100
#5oh-my-openagentTerminal coding agent with multi-model orchestration and git-backed memory70.3k starscustom license80 out of 100