Hand a GUI task on this Mac (click, type, read a window, move through an app) to Codex Computer Use, one bounded subtask at a time, and get back a result the runner has checked against what actually happened. Use when a task needs a desktop app driven or read and no API, CLI, or connector covers it, or when the user says use the computer, drive the app, or computer use.
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Computer Use
Codex Computer Use (the Codex desktop plugin, model gpt-6-astra) can read and operate
Mac apps through their accessibility trees and screenshots. This skill hands it one
bounded subtask per call, headless, through `codex exec`, and audits the result against
the recorded event stream rather than the model's own account. You write the brief, run
the script, read the audited result and the screenshots, and report to the user in plain
words.
Every call is a Codex model call billed to whatever the user's Codex profile is
configured for. Preflight and each subtask take tens of seconds to a few minutes.
## What has and has not been verified
Verified live on 2026-09-20 with this runner, on the development Mac:
- Four Codex profiles, each with `CODEX_HOME` set to it: live preflight passed (one
read-only Sky call listed the running apps) and a read-only `run` against Chrome came
back `done` with a screenshot the reviewer opened and matched to the screen. All four
resolve to one signed helper inside the ChatGPT app. Every call was routed through the
user's local proxy, as the `routing` line in preflight showed beforehand.
…
The line at the top of this page pins the label git:20260921.ccd46de, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
B 16 of 17 checks passed. Fixed automatic checks with no AI involved, so the same checks give the same answer every run.
fail: No prompt-injection phrasing (matched: Do not tell the user)
pass: Frontmatter block present
pass: Frontmatter declares a name
pass: Frontmatter declares a description
pass: Size between 200 bytes and 200 KB (14608 bytes)
pass: No zero-width or bidi control characters
pass: No instruction hidden inside an HTML comment
pass: No link to an exfiltration or paste host
pass: No credential-shaped string
pass: No instruction to send local credentials anywhere
pass: No text hidden with inline styles
pass: No curl or wget piped into a shell
pass: No recursive delete of root, home or parent
pass: No instruction to read or print local credentials
pass: No base64 blob over 200 characters
pass: No link to a raw IP address
pass: No script tag
Starting a new one? Use a template for SKILL.md on mdpantry, which passes every one of these checks and says what to change in it.
Source
GitHub
timharris707/skills · 44 stars · license MIT · pushed 2026-10-19 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_r2esmwy7eejguy2v
GET https://markdownregistry.com/api/v1/resolve?ref=timharris707/skills/computer-use
GET https://markdownregistry.com/api/v1/blob/c6c7b768e969c30db9d276c10b9bf297d387b98637fac67de735d057875b9bab
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
timharris707/skills · editions/codex/overrides/skills/investigate/ingest/SKILL.md · For a media file or URL shared with a goal, transcription requests, or takeaways from a call, demo, playtest, or video…
timharris707/skills · plugins/clickai-codex/skills/author/huh/SKILL.md · When a message did not land or pasted agent output needs translation, restate it plainly, expand shorthand, separate…
timharris707/skills · plugins/clickai-codex/skills/author/plainspoken/SKILL.md · For chat replies, status updates, review comments, and commit messages, write plainly without AI tells. Also use to…
timharris707/skills · plugins/clickai-codex/skills/author/writing-for-agents/SKILL.md · For skills, standing agent instructions, recorded rules, or documents agents skip, write and prune reliable triggers…
timharris707/skills · plugins/clickai-codex/skills/author/writing-for-humans/SKILL.md · For public-facing drafts, pages that read like agent documentation, or a final AI-tell scrub, write clear human copy in…
timharris707/skills · plugins/clickai-codex/skills/decide/advisory-board/SKILL.md · For an explicitly requested advisory board, roundtable, panel, idea tournament, multi-model review or debate, red-team…
timharris707/skills · plugins/clickai-codex/skills/decide/decision-map/SKILL.md · When asked to chart a decision map, work a map ticket, or plan work whose open decisions gate each other, map the…
timharris707/skills · plugins/clickai-codex/skills/decide/grilling/SKILL.md · When asked to grill, stress-test, or pressure-test an idea, or before a decision map or spec rests on an untested…
timharris707/skills · plugins/clickai-codex/skills/investigate/codebase-review/SKILL.md · Before an upcoming feature, when merged work crosses the recorded review threshold, or when implementation exposes…
timharris707/skills · plugins/clickai-codex/skills/investigate/ingest/SKILL.md · For a media file or URL shared with a goal, transcription requests, or takeaways from a call, demo, playtest, or video…
timharris707/skills · plugins/clickai-codex/skills/investigate/prototype/SKILL.md · When discussion or static mockups cannot settle how something should look, behave, or feel, build a disposable…
qwenlm/qwen-code · packages/core/src/skills/bundled/computer-use/SKILL.md · Control local desktop applications through Computer Use for tasks that require reading or operating app UI. Prefer…
thinkinaixyz/deepchat · plugins/cua/skills/computer-use/SKILL.md · Drive native desktop apps through DeepChat's built-in Computer Use tools. Use when the user asks to operate, inspect…
clacky-ai/openclacky · lib/clacky/default_extensions/computer-use/skills/computer-use/SKILL.md · Drive the desktop — macOS, or the Windows desktop through WSL — when a task needs a native app, dialog, menu bar, file…
xuzhougeng/wisp-science · skills/computer-use/SKILL.md · Use Cua Driver through MCP to inspect and operate the user's native desktop apps on Windows, macOS, or Linux. Trigger…
autonomous-ai/lampos · skills/computer-use/SKILL.md · Operate apps/websites on the paired Mac via Buddy: Calendar, Notes, forms, screenshots, files. Never use the headless…
autonomous-ai/autonomous-os · skills/computer-use/SKILL.md · Operate apps/websites on the paired Mac via Buddy: Calendar, Notes, forms, screenshots, files. Never use the headless…
autonomous-ai/physical-ai-operating-system · skills/computer-use/SKILL.md · Operate apps/websites on the paired Mac via Buddy: Calendar, Notes, forms, screenshots, files. Never use the headless…
