agentleFS
cgoinglovecgoinglove/thursday/skills/computer/SKILL.md

computer

Skill56 starsChanged 46 days agoMITView on GitHub

Operates this Mac itself: its apps, windows, screen and clipboard. Use it when the job is in an app rather than a browser.

  • Installs packages

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If anything is missing, say which one in one line (a question to Thursday) and stop.

    line 30
  2. Distinctive:

    Do not retry, do not work around it: without the grant every capture is a black rectangle and every click is refused, and you will spend …

    line 32
  3. Very specific:

    Sign-ins are the user's: Never type a password, an id, or a one-time code into any app.

    line 58
  4. Very specific:

    Verify: instead of sleeping — peekaboo verify polls a predicate and answers satisfied / unsatisfied / unknown.

    line 95
  5. Very specific:

    Run peekaboo learn for the full agent guide, peekaboo tools for the tool catalog, and peekaboo <command> --help for the exact flags of one …

    line 101

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

The Mac

You drive the user's Mac from your shell with peekaboo. This file is how: look at a window, act on what you found, and what to do when the machine says no.

peekaboo is a plain CLI, not an agent. You are the one that looks and decides; it only observes and acts.

In this app

Everything the app has to say about the Mac is in this section. The bots' personas and the prompts do not repeat it — change it here.

Permissions come before everything. Screen Recording and Accessibility are granted by the user, in System Settings, to the process that runs you — you cannot grant them and neither can peekaboo. So start with:

peekaboo permissions status --json

If anything is missing, say which one in one line (a question to Thursday) and stop. Do not retry, do not work around it: without the grant every capture is a black rectangle and every click is refused, and you will spend your steps proving it.

Read as text first. The tree is the cheap way to see a window, and it does not need a screenshot at all:

peekaboo see --app Messages --tree --no-screenshot --json

That gives you the window's elements — their roles, titles and values — which is what you usually wanted when you thought "take a screenshot". Capture pixels only when the layout itself is the question (is this cut off, does this look right), and then look at the file yourself.

Never --analyze, never agent. Those two spend a provider key, and keys do not live in your shell. You are the model here: capture, then read what you captured.

Stay in the background. The user is sitting at this machine. peekaboo delivers input in the background when it can resolve an exact target, so pass --app / --window-title and let it. Add --foreground only when an action genuinely needs the real pointer, and say why in your description. Never quit or close the user's apps unless they asked — minimize instead.

Sign-ins are the user's. Never type a password, an id, or a one-time code into any app. When a window wants one, tell the user in one line (a question to Thursday) and stop.

Files. Captures go under scratch/ while you work (--path scratch/<name>.png); a picture that is the result of the job goes to your folder under artifacts/. Never write to the workspace root.

Ask before you change the user's things. Looking is free — reading a window, listing apps, taking a capture. Sending a message, deleting a mail, posting a form: that is the user's action, so confirm it first unless they already told you to do exactly that.

When something is wrong on this machine, the CLI already knows. Every command takes --json, and a failure carries error.code, error.message and often an error.hint — read the hint before you invent a second approach. Action results also carry effect: confirmed means the app acknowledged it, suspected_noop means nothing changed (your target was probably wrong — look again), refused means a safety gate stopped you.

Quick start

peekaboo permissions status --json           # first, always
peekaboo app list --json                     # what is running
peekaboo see --app Safari --tree --no-screenshot --json   # read a window
peekaboo click --app Safari "Sign in"        # act on what you read
peekaboo type "hello" --app Safari

The loop

  1. Observe — see (tree for text, pixels when layout matters). Every observation returns a snapshot id; later commands target that snapshot.
  2. Act — one command: click, type, press, scroll, menu click.
  3. Observe again. The window changed, so your element ids did too. Never act twice off one see.
  4. Verify instead of sleeping — peekaboo verify polls a predicate and answers satisfied / unsatisfied / unknown. Sleeps are how a job burns a minute and still gets it wrong.

What it can do

Run peekaboo learn for the full agent guide, peekaboo tools for the tool catalog, and peekaboo <command> --help for the exact flags of one command. The surface moves between releases — this list is what exists, not how to spell it.

Observe — see (capture + element map, --tree, --no-screenshot, --no-elements), capture (action, live, video), screen list, verify, permissions.

Act — click, type, press (keys and chords), scroll, move, drag, paste, set-value, action (a named accessibility action).

The system — app (launch, focus, quit, hide, list), window (focus, move, resize, minimize, set-bounds, list), menu (click, list), menubar, dock, dialog, clipboard (get, set, save, restore), space.

Grammar shared by every command

  • Targets: --app, --window-title, --window-id, an element id from a see, a text query, or --at x,y. Coordinates are relative to the targeted window unless you add --global.
  • Durations: 500, 500ms, 2s, 1.5s are the same kind of value.
  • Modifiers: comma-separated — cmd,shift.
  • Output: --json on everything. success, data, error, and — for actions — effect.

Install

The app does not install peekaboo. When it is not on PATH, say so and stop — do not build it from source, and do not install it without the user's yes. It needs macOS 15 or later.

brew tap steipete/tap
brew install steipete/tap/peekaboo
peekaboo --version

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.