Linux App Playbooks

Files, editor, viewer, browser via computer_use.

Skill metadata

Source	Bundled (installed by default)
Path	skills/desktop/linux-app-playbooks
Version	1.0.0
Author	VibeOS
License	MIT
Platforms	linux
Tags	linux, desktop, computer-use, nautilus, gedit, evince, firefox
Related skills	linux-desktop, choose-interaction-surface, computer-use, desktop-operate

Reference: full SKILL.md

信息

The following is the complete skill definition that VibeOS loads when this skill is triggered. This is what the agent sees as instructions when the skill is active.

Linux App Playbooks

GUI procedures for common Linux desktop apps. Prefer terminal and file tools for paths; prefer browser_* for web pages. Long Operate jobs → desktop-operate (≤2 retries + scorecard).

When to Use

• File manager UI when the user wants a visible browse path
• Simple text editor draft when no project workspace file is wanted
• PDF/image viewer export (Evince / Eye of GNOME / Okular)
• Desktop browser fallback when the browser toolset is unavailable

Prerequisites

• Display session (X11 or Wayland)
• computer_use enabled; load linux-desktop for clipboard helpers

How to Run

1. Confirm native desktop app → computer_use.
2. Capture SoM/AX; click by element index; re-capture after moves.

Quick Reference

App class	Examples	Prefer instead
File manager	Nautilus, Dolphin, Thunar	terminal / file tools
Text editor	Gedit, Kate, Mousepad	write_file / patch
Document / image viewer	Evince, EOG, Okular	read_file / vision_analyze
Browser	Firefox, Chromium	browser_* tools

Procedure

File manager — open a folder

1. Capture the file manager window.
2. Use location / path bar when available; type absolute path + Enter.
3. Otherwise walk sidebar → folders with re-capture each step.

Text editor — draft

1. Capture editor; focus the text buffer.
2. Type or paste; save only if asked (capture Save dialog).

Document / image viewer — open and export

1. Open a known path via file manager or terminal (xdg-open), then capture.
2. Export/Save As: re-capture the dialog; prefer the task workspace path.
3. Do not overwrite the original without human approve.
4. Empty AT-SPI tree → one re-capture; after two stuck tries escalate (desktop-operate scorecard).

Desktop browser — open URL (fallback)

1. Prefer browser_navigate when the browser toolset is loaded.
2. Else capture address bar, type URL, Enter.

Pitfalls

• Wayland vs X11 changes which apps expose a usable a11y tree.
• Snap/Flatpak sandboxes may hide UI from AT-SPI — fall back to terminal.
• Do not click through polkit / sudo graphical prompts without the user.
• Web CAPTCHA in Firefox/Chrome: use browser_* + challenge-gate, not computer_use pixel clicks on the challenge.
• Blind clicks burn the Operate retry budget — use element indices.

Verification

• File manager shows the requested directory.
• Editor contains the drafted text.
• Viewer export exists when requested.
• Browser loaded the URL (or browser_* was used).