chore(core): remove warning msg for gpt-4

[yuyutaotao]

No description provided.

[chatgpt-codex-connector]

💡 Codex Review

midscene/packages/core/src/ai-model/llm-planning.ts

Lines 53 to 59 in 9f1d3e6

	let imagePayload = screenshotBase64;
	let imageWidth = size.width;
	let imageHeight = size.height;
	const rightLimit = imageWidth;
	const bottomLimit = imageHeight;
	if (vlMode === 'qwen-vl') {
	const paddedResult = await paddingToMatchBlockByBase64(imagePayload);

Planning requests omit element overlays for non-VL models

The planning pipeline now forwards the original screenshot for non-VL models after only handling the qwen-vl padding case. The prompts used for non-VL planning still tell the model that element IDs are “marked with a rectangle in the screenshot” (prompt/llm-planning.ts lines ~272–274). With the overlay removed, GPT‑4o receives an unannotated screenshot and cannot reliably return element IDs for actions, leading to incorrect or null locate parameters. Reintroduce the overlay or revise prompts/consumers so the model’s assumptions match the data it sees.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting

• @codex fix this CI failure
• @codex address that feedback

[chatgpt-codex-connector]

Non-VL element lookup no longer overlays element IDs

The AiLocateElement flow now skips markupImageForLLM whenever vlMode is undefined, so GPT‑4o receives the raw screenshot instead of the annotated version. The locator prompt for non-VL models still asserts that non-text elements “have been highlighted” and expects IDs marked in the image (prompt/llm-locator.ts around lines 83‑86). Without those overlays the model cannot map the textual element list to the screenshot, so element searches for default models will degrade or fail outright. Consider restoring the markup step or updating the prompt and downstream logic to align with the unannotated image.

Useful? React with 👍 / 👎.