Index of every page on this site: llms.txt
All skills
Run it yourself

Drives a real browser or desktop app to test, check or automate anything that needs a real interface - opening pages, clicking through flows, filling forms, taking screenshots, confirming a feature works. Also checks the agent's own work in the real interface after a change, when the user asks it to or the testing skill needs it. Uses browser-use (the user's own Chrome) first, then agent-browser (a separate, clean Chrome), then agent-desktop for desktop apps only after asking. Use when the user asks to automate a web or desktop task, test or check something in the browser, verify or check your own work, click through a flow, reproduce a UI bug, take a screenshot of a page, or fill a form. Not for writing test files that run again (use testing), proving a change against an answer key before a push (use prove), or fetching a public page or API that a plain request can read.

  • 4 files
  • 13.4 KB
  • Updated 3 weeks ago
  • GitHub
Use this Skill: https://skilld.dev/gh/avik-creator/agent-skills/automation

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesagent-desktop.md

≈613 tokens on demand. Your agent reads this file only when SKILL.md points to it.

agent-desktop

Read when using desktop control (step 3), after the user has said yes: native apps, or when neither browser tool works.

Set up

npm install -g agent-desktop          # the user runs this, not you
agent-desktop version
agent-desktop permissions --guide     # macOS: grants Accessibility permission
  • macOS needs Accessibility permission.
  • Windows needs no grant for apps running at the terminal's privilege level.
  • Linux needs the user's graphical session with at-spi2-core.

Load its guide first

agent-desktop skills get desktop      # the loop, references, verification and retry rules
agent-desktop skills get platform     # what differs on this operating system
agent-desktop <command> --help        # flags for one command

Follow these over this file. Why: they ship with the installed version.

The loop

Take a snapshot, act on a reference, then take a snapshot again to check:

agent-desktop list-apps
agent-desktop snapshot --app Finder -i      # interactive elements, each with a reference
agent-desktop click @s8f3k2p9:e8            # a reference from that snapshot
agent-desktop snapshot --app Finder -i      # did it do what you expected?
  • A reference belongs to its snapshot (@<snapshot>:eN). A bare @e8 is rejected.
  • STALE_REF or AMBIGUOUS_TARGET means the screen changed: take a new snapshot and use its references.
  • With several windows open, snapshot lists the window ids. Pass one with --window-id.
  • For a busy app, map it first (snapshot --skeleton --app Slack -i --compact), then open one region (snapshot --root <ref> -i --compact).
  • Every command prints JSON with ok, and either data or error. Retry only when error.disposition.retry is safe.

The mouse and keyboard

By default it acts through the accessibility system, without moving the pointer or taking focus, so the user can keep working. The --headed flag allows real mouse and keyboard input. Ask before using it. Why: it takes over the user's mouse and keyboard while it runs.

Privacy

Snapshots, screenshots, clipboard contents and debug files can hold whatever is on the user's screen: messages, documents, passwords.

  • Read the clipboard only when the task needs it.
  • Keep that output out of public logs, issues and uploads.
  • Don't use its optional JEV helper without asking. Why: it sends screen descriptions to an outside model provider.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at a0324a3. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 days ago.

Activeupdated 3 weeks ago

README badge

README badge for avik-creator/agent-skills/automation