Live Tutorial

Live Tutorial

Demonstrate the requested workflow, then return a verified MP4 made from its real screenshots.

Write captions and annotation text in the language of the user's tutorial request, keeping app control names as displayed.

The request already authorizes completing and rendering the tutorial. Do not ask whether to continue or offer screenshots/text instead of the requested video merely because execution paused. Resume with the same recording, captures, and manifest; ask only for genuinely missing information or authorization.

Use PNG paths returned in each observation's artifacts array. The recording output_dir is the destination for the video, not the location of those screenshots; do not search it for frames. The renderer is ready to run with the manifest below: no source inspection, image-processing code, or dependency setup is needed.

Use the live tool results for recording directories, screenshot paths, and current target tokens. After a pause, read a specific saved artifact only if the checkpoint omits a required value; do not scan directories or reread evidence already available. Each new window observation invalidates earlier tokens, so make one observation with any needed query and screenshot, then act immediately from that same result. Use the shell tool only for the final renderer command.

1. Select the target application and one short route from its current view to the requested result. Demonstrate only the requested workflow, not every related option. Do not reset the application or visit unrelated settings merely to create a starting frame. If already at the result, show only the navigation needed to teach the requested step; do not claim a pre-existing state was caused by an action you did not perform.
2. Start computer_use.cua.start_recording with record_video: false; retain output_dir.
3. Use the shortest supported route: a known keyboard shortcut, an accessibility-backed action, or computer_use.cua.invoke_menu for native menus; never shell automation. Capture the starting state before acting. During recording, supported input actions automatically return a fresh screenshot and element queries include screenshots. Reuse the returned observation for verification and the next step instead of taking another identical screenshot. Observe again only if the result lacks needed evidence. Captions must describe what you actually did, including any shortcut; retain the meaningful intermediate states.
4. Always stop recording. Select 2–6 distinct captures from the successful route, normally one per action plus the result; omit setup, failed attempts, retries, and polling. Keep valid captures across retries rather than re-recording the workflow. The final result must agree with the demonstrated actions. Use a short instruction for each action and a confirmation caption for the result.
5. Write the manifest below with the file tool, then run the bundled renderer through the shell tool. Do not replace this with a raw screen recording or a custom automation/rendering script. Do not inspect or rewrite the renderer unless it reports an error.
6. Check the renderer's resolvedTargets against the intended controls and the source observations. Correct any mismatch before returning fileMarker verbatim. The returned previewPaths identify the rendered step frames. If unfinished, explain the obstacle.

Render

"$NODE" "$LEON_CODEBASE_PATH/skills/agent/live-tutorial/scripts/render-tutorial.mjs" --manifest "/exact/path/to/tutorial.json"

Leon supplies NODE and LEON_CODEBASE_PATH. Do not set a working directory or construct executable paths. No system Node, Python, or FFmpeg installation is needed.

On PowerShell, use & $env:NODE "$env:LEON_CODEBASE_PATH/skills/agent/live-tutorial/scripts/render-tutorial.mjs" --manifest "/exact/path/to/tutorial.json".

Manifest:

{
  "outputDir": "<output_dir from start_recording>",
  "steps": [
    {"screenshotPath": "<pre-action PNG path>", "targetToken": "<element_token from this exact capture>", "instruction": "Open the relevant setting."},
    {"screenshotPath": "<result PNG path>", "instruction": "Read the displayed value."}
  ]
}

Use exact current-session paths and captions of at most 240 characters. Prefer targetToken: the renderer reads capture-bound geometry from the PNG's companion JSON and draws the control outline, arrow, and step number automatically.

Use point: {x, y, coordinateWidth, coordinateHeight} when the intended control has no accessible target geometry, even if other controls in the capture do. Copy the exact x/y and coordinate dimensions from the successful action performed on that capture; do not estimate a new annotation point. Never copy a point from another capture or mix window and desktop coordinates. The final result frame must have no point annotation; omit its marker or use a verified targetToken when highlighting a result is genuinely useful.

A tutorial request authorizes demonstrating navigation, not choosing an unspecified setting value or completing a destructive, financial, publishing, or security-sensitive action. Stop before committing such changes unless the owner explicitly authorized them.