deskclawlisted

# deskclaw — the read-only desktop eye You CAN see the Windows desktop. Before this existed, Wes had to paste screenshots by hand; that is the workflow you are replacing. Reach for this instead of asking him what a window says. **Read-only.** There is no click, type, key or focus verb. Do not promise one. **Cheapest form first:** `declick desk windows`, `declick desk tree <title> --interactive --grep <re>`, `declick desk read <title> "<Type:Name>"` return JSON with `--fields`/`--limit`, work from any subagent, and cost a fraction of a screenshot. Take the screenshot only when the question is visual. `~/.claude/tools/deskclaw/` — spec at `~/.claude/docs/superpowers/specs/2026-08-12-deskclaw-design.md`. ## Use it, or use something else | Target | Tool | |---|---| | A web page, logged-in or not | `agent-browser` / Playwright over CDP. NOT deskclaw. | | Wes's iPhone | `sidetap` (the `phone` skill). NOT deskclaw. | | An Electron app (Magnetic, VS Code, Slack) | Playwright Electron or `agent-browser skills get electron` — it renders web UI, so the DOM is richer than the UIA tree. | | Blender, Unity | Their headless paths (`feeders/blender/render.py`, Unity CLI). They draw their own UI in OpenGL and expose almost nothing to UIA. | | Native Windows: dialogs, installers, Explorer, Task Scheduler, Office, legacy apps | **deskclaw** | | "What is open right now?" / "what does that window say?" | **deskclaw** | ## The four verbs ```bash ~/.claude/tools/deskclaw/desk windows