#350350th of 455 in Skills, plugins and prompts

macos-harness

CLI that lets coding agents read and control macOS apps in the background

Stars
1k
License
MIT
Last commit
Oct 2026
Last release
Oct 2026
Language
Python
Stars, last 12 weeks+478
19 Aug 2026: 535 stars26 Aug 2026: 787 stars2 Sept 2026: 819 stars9 Sept 2026: 837 stars16 Sept 2026: 869 stars23 Sept 2026: 886 stars30 Sept 2026: 892 stars7 Oct 2026: 901 stars14 Oct 2026: 915 stars21 Oct 2026: 964 stars28 Oct 2026: 971 stars4 Nov 2026: 1,006 stars11 Nov 2026: 1,013 stars

535 on 19 Aug 20261,013 on 11 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

macos-harness gives a coding agent eleven shell commands (get_app_state, click, type_text, press_key, set_value, scroll, drag, paste and others) to operate any macOS app. It reads a window's accessibility tree as numbered text plus a screenshot, then acts through accessibility actions or events sent to the app's process, so the real pointer and front app are not disturbed. It installs a skill for Claude Code, Codex, Pi or OpenCode.

Who it is for: Developers who want coding agents to drive native Mac apps

Strengths

  • Same eleven commands work for any app; no per-app tools
  • Acts in the background without moving the pointer or changing the front app
  • Repeat reads print only what changed, keeping agent context small
  • Optional policy.toml: ask before first use per app, blocked_urls, strict mode

Weaknesses

  • macOS only and labelled experimental
  • Needs Accessibility and Screen Recording permissions for every call
  • Anonymous telemetry is on by default; opt out with a command
  • No stop button; headless agents must be stopped by killing the process

What it needs

  • Needs macOS, uv, Python 3.12, Accessibility permission, Screen Recording permission
  • Models: Claude Code, Codex, Pi, OpenCode

What changed

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in Skills, plugins and prompts

See all 455
Also in Skills, plugins and prompts
RankProjectScore
1claude-memPersistent cross-session memory plugin for Claude Code and other coding agents116.4k stars+16,587 (+17%) in 30 daysApache-2.097 out of 100
2impeccableDesign skill with 24 slash commands for AI coding agents' frontend output Live demo ↗ (opens in a new tab)108.8k stars+28,464 (+35%) in 30 daysApache-2.097 out of 100
3gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles141.3k stars+5,544 (+4.1%) in 30 daysMIT96 out of 100
#4rufloAgent harness adding swarms, memory and hooks to Claude Code and Codex Live demo ↗ (opens in a new tab)81.5k stars+7,374 (+9.9%) in 30 daysMIT96 out of 100
#5cavemanShortens coding-agent replies and compresses tool output before the model reads it Live demo ↗ (opens in a new tab)119.3k stars+8,422 (+7.6%) in 30 daysApache-2.095 out of 100