Screen captures in Wayland

[collective-thought-media]

Is this a task that Wayland should be able to handle?
Using Wayland GUI, Mistral Nemo model (local)

"I need you to create a series of screen captures of the dev site here's the relevant info
URLs:

1920x1080, mobile: 1280x720 + 375x812
full page
JPG, name should be: pagetitle_Viewport-size.jpg
Output folder: G:\BCD\NWCF\Web\QA\scrnshots"
If so can you provide some guidance on the best prompt to achieve this. I tried it and it seems to just be confused as to what I want... it created a plan to execute the screenshot capture, but then wouldn't run it. Then it kept just creating a basic website post launch plan
And when I would try to steer it back towards the screenshot task it's just kind of looped between making a plan for doing that and then reverting back to a full website post launch plan.

[VectorPeak]

This is a reasonable task for Wayland only if the runtime has the ability to control a browser and write files to disk. The LLM by itself cannot capture the page or save files; it needs an attached tool/skill/command runner such as Playwright, Puppeteer, browser automation, or shell access. From the screenshots, it looks like the model understood the broad intent but stayed in “planning/text generation” mode instead of invoking an actual execution path.

The clearest symptom is in the first screenshot: it correctly restates the screenshot requirements, then asks you to confirm before proceeding, but immediately drifts into “Core Requirements” and the NWCF launch checklist. In the second screenshot it even proposes Puppeteer and says “I will run this script now,” but it only prints a partial script and then asks for information you already provided. The third screenshot shows the same context bleed: after being redirected back to screenshots, it starts summarizing roles, skills, and launch context instead of doing the file/browser operation.

That usually points to one of these:

• The local model does not have a real execution tool available for browser/file tasks.
• Tool use is available, but the model is not reliably selecting it.
• The prompt includes too much project-launch context, so the model keeps falling back to that higher-level task.
• The model is treating “create screenshots” as a planning request rather than an imperative action.

For this task, I would avoid asking for “guidance” or describing the whole launch project. Give it a narrow executor-style prompt and explicitly tell it to stop if it cannot run commands. Something like:

Task: capture screenshots only.

Use the available browser automation or command execution tool. Do not create a launch plan. Do not summarize project context. Do not ask for confirmation unless a required value is missing.

Input:
- URLs:
 1. <url 1>
 2. <url 2>
 3. <url 3>
- Viewports:
 - 1920x1080
 - 1280x720
 - 375x812
- Capture mode: full page
- Format: JPG
- Filename format: <page-title>_<width>x<height>.jpg
- Output folder: G:\BCD\NWCF\Web\QA\scrnshots

Required behavior:
1. Open each URL.
2. Capture a full-page screenshot at each viewport.
3. Save each screenshot as JPG in the output folder.
4. Report only the files created.

If you do not have browser automation and filesystem access, say exactly:
"I cannot perform this task because I do not have browser automation and filesystem access."

If Wayland supports running shell commands, an even more reliable version is to ask it to generate and run a Playwright script, not to “figure out” the whole workflow conversationally:

Use Playwright to capture the screenshots. Create or run a script that:
- creates G:\BCD\NWCF\Web\QA\scrnshots if it does not exist
- visits each URL below
- captures full-page JPG screenshots at 1920x1080, 1280x720, and 375x812
- names each file using the page title and viewport size
- prints the final saved file paths

Do not create a website launch plan. Do not ask for confirmation. Execute the screenshot task now if tools are available.

URLs:
<url 1>
<url 2>
<url 3>

A practical thing to check first is whether the environment running Wayland can actually see G:\BCD\NWCF\Web\QA\scrnshots. If the model/backend is running in WSL, Docker, a sandbox, or another user context, G:\... may not exist from its point of view. In that case the model should either save to an accessible path or tell you it cannot access that destination.

---

If my answer solved your problem, you can click answered the question. I'm really here to help, and along the way I'm also collecting Galaxy Brain badges haha 😆