Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)
Python
import base64
from typing import Any, Dict
from preview_screenshot import capture_preview_screenshot
from agent.state import AgentFileState
from agent.tools.types import ToolExecutionResult, ToolMultimodalPart
PREVIEW_VIEWPORTS = ("desktop", "mobile")
async def run_screenshot_preview(
_args: Dict[str, Any],
*,
file_state: AgentFileState,
) -> ToolExecutionResult:
"""Render the current HTML and return screenshots.
These previews are for *seeing*, not keeping: the model views them as
attached image bytes (multimodal parts) to verify its work and never
embeds them in its output, so they are NOT persisted as assets. A data
URL is inlined into the summary purely so the UI can show the same preview.
"""
if not file_state.content:
return ToolExecutionResult(
ok=False,
result={"error": "No file exists yet. Call create_file first."},
summary={"error": "No file to screenshot"},
)
screenshots: list[Dict[str, Any]] = []
multimodal_parts: list[ToolMultimodalPart] = []
try:
for viewport in PREVIEW_VIEWPORTS:
image_bytes = await capture_preview_screenshot(
file_state.content,
device=viewport,
full_page=True,
)
display_name = f"preview_{viewport}.png"
image_part_index = len(multimodal_parts)
encoded_image = base64.b64encode(image_bytes).decode("ascii")
data_url = f"data:image/png;base64,{encoded_image}"
screenshots.append(
{
"viewport": viewport,
"full_page": True,
"image_part_index": image_part_index,
"image_display_name": display_name,
"image_bytes": len(image_bytes),
# Inlined for the UI thumbnail only — never stored as an asset.
"image_url": data_url,
"status": "ok",
}
)
multimodal_parts.append(
ToolMultimodalPart(
display_name=display_name,
mime_type="image/png",
data=image_bytes,
)
)
except Exception as exc:
print(f"Preview screenshot failed: {exc}")
return ToolExecutionResult(
ok=False,
result={"error": f"Screenshot failed: {exc}"},
summary={"error": "Screenshot failed"},
)
result: Dict[str, Any] = {
"content": (
"Full-page desktop and mobile screenshots of the current preview "
"are attached."
),
"details": {
"screenshots": [
{
"viewport": screenshot["viewport"],
"full_page": screenshot["full_page"],
"image_part_index": screenshot["image_part_index"],
"image_display_name": screenshot["image_display_name"],
"image_bytes": screenshot["image_bytes"],
}
for screenshot in screenshots
],
},
}
summary: Dict[str, Any] = {
"screenshots": screenshots,
"status": "ok",
}
return ToolExecutionResult(
ok=True,
result=result,
summary=summary,
multimodal_parts=multimodal_parts,
)