desktop-inspect

name: desktop-inspect description: Inspect the running Limn desktop app via its debug HTTP server. Use this skill whenever you need to check what the app is rendering, verify DOM state, look for JS errors, capture screenshots, check node counts, or investigate any runtime behavior of the desktop app. Also use when debugging issues that require seeing what the WKWebView is actually displaying. Trigger on phrases like "check the app", "what does it look like", "is it rendering", "any JS errors", "screenshot the app", "how many nodes", or any ad-hoc investigation of the running desktop app. allowed-tools: Bash(just desktop-inspect-*), Read, Write

Inspect Running Desktop App

The Limn desktop app (Debug builds only) runs an HTTP server on localhost:9876 for programmatic inspection. All interaction goes through just recipes -- never call curl directly.

Prerequisites

The app must be running in debug mode (just desktop-dev). If commands return empty output or connection refused, the app is not running -- ask Ted to start it.

Output directory

All inspection artifacts (screenshots, logs, scripts, notes) go in .llm/inspect/. This directory is gitignored.

Commands

All commands below are just recipes. Run them from the project root. The optional filename argument is positional (not file=).

List open windows

just desktop-inspect-windows

Returns JSON array of open windows with index, filename, and URL. Run this first to see what's available.

Evaluate JavaScript

just desktop-inspect-eval '<js-expression>'
just desktop-inspect-eval '<js-expression>' <filename>

Returns {"result": <value>} or {"error": "..."}. Pass the filename as the second argument to target a specific window.

Capture screenshot

just desktop-inspect-screenshot
just desktop-inspect-screenshot <filename>

Saves PNG to .llm/inspect/screenshot-<timestamp>.png. After capturing, read the saved PNG file to visually verify rendering.

Get editor state

just desktop-inspect-state
just desktop-inspect-state <filename>

Returns {"nodeCount": N, "selectedId": "...", "filename": "...", "zoom": N}.

Get document JSON

just desktop-inspect-json
just desktop-inspect-json <filename>

Returns the full mind map document as pretty-printed JSON (the same structure as a .limn file). Useful for verifying file content, node structure, and data integrity without needing to save and read from disk.

Saving results

When investigating an issue, save findings to .llm/inspect/:

• Screenshots: automatically saved there by just desktop-inspect-screenshot
• Investigation logs: save as .llm/inspect/<topic>.md with findings, commands run, and conclusions
• Helper scripts: save one-off inspection scripts to .llm/inspect/ (not the project source tree)

Investigation patterns

Check for JS errors

just desktop-inspect-eval 'window.__lmnErrors || "none"'

Verify file loaded correctly

just desktop-inspect-state test-a.limn
just desktop-inspect-eval 'document.querySelectorAll("[data-node-id]").length' test-a.limn

Visual verification

just desktop-inspect-screenshot test-a.limn

Then read the saved PNG to see what the app is rendering.

Multi-window comparison

just desktop-inspect-windows
just desktop-inspect-state test-a.limn
just desktop-inspect-state test-b.limn

Verify document data

just desktop-inspect-json test-a.limn

Inspect the full JSON to check node text, positions, parent-child relationships, or any document-level metadata.

Inspect DOM structure

just desktop-inspect-eval 'document.querySelector("[data-node-id]")?.outerHTML'
just desktop-inspect-eval 'document.querySelector(".selected")?.textContent'

Tips

• All commands except desktop-inspect-windows accept an optional filename as their last positional argument to target a window.
• If no filename is given, the first window is used.
• The debug server only exists in #if DEBUG builds -- it is not in Release.
• For complex JS, use single quotes around the whole expression in just desktop-inspect-eval.
• Run multiple independent queries in parallel for faster investigation.