Skip to content

Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)

Python

Fix select-and-edit losing the element reference on imported projects

Abi Raja3aebd3c37d98
diff --git a/backend/prompts/prompt_types.py b/backend/prompts/prompt_types.py
index 08020652..68324199 100644
--- a/backend/prompts/prompt_types.py
+++ b/backend/prompts/prompt_types.py
@@ -1,14 +1,20 @@
from typing import List, Literal, TypedDict
-class UserTurnInput(TypedDict):
- """Normalized current user turn payload from the request."""
-
+class _UserTurnInputRequired(TypedDict):
text: str
images: List[str]
videos: List[str]
+class UserTurnInput(_UserTurnInputRequired, total=False):
+ """Normalized current user turn payload from the request."""
+
+ # Full instruction for the model when it differs from the display text
+ # (e.g. includes the selected-element reference, built by the frontend).
+ full_text: str
+
+
class PromptHistoryMessage(TypedDict):
"""Explicit role-based message structure for edit history."""
diff --git a/backend/prompts/request_parsing.py b/backend/prompts/request_parsing.py
index 6ac7f81a..b54bfdfd 100644
--- a/backend/prompts/request_parsing.py
+++ b/backend/prompts/request_parsing.py
@@ -16,12 +16,18 @@ def parse_prompt_content(raw_prompt: object) -> UserTurnInput:
prompt_dict = cast(dict[str, object], raw_prompt)
text = prompt_dict.get("text")
- return {
+ parsed: UserTurnInput = {
"text": text if isinstance(text, str) else "",
"images": _to_string_list(prompt_dict.get("images")),
"videos": _to_string_list(prompt_dict.get("videos")),
}
+ full_text = prompt_dict.get("fullText")
+ if isinstance(full_text, str) and full_text.strip():
+ parsed["full_text"] = full_text
+
+ return parsed
+
def parse_prompt_history(raw_history: object) -> List[PromptHistoryMessage]:
if not isinstance(raw_history, list):
diff --git a/backend/prompts/update/from_file_snapshot.py b/backend/prompts/update/from_file_snapshot.py
index cc79be40..937b6df6 100644
--- a/backend/prompts/update/from_file_snapshot.py
+++ b/backend/prompts/update/from_file_snapshot.py
@@ -17,7 +17,13 @@ def build_update_prompt_from_file_snapshot(
design_system: str | None = None,
) -> Prompt:
path = file_state.get("path", "index.html")
- request_text = prompt.get("text", "").strip() or "Apply the requested update."
+ # full_text carries the complete model-facing instruction (e.g. with the
+ # selected-element reference); text is the user-typed display string.
+ request_text = (
+ prompt.get("full_text", "").strip()
+ or prompt.get("text", "").strip()
+ or "Apply the requested update."
+ )
selected_stack = build_selected_stack_policy(stack)
image_policy = build_user_image_policy(image_generation_enabled)
design_system_block = build_design_system_prompt_block(design_system)
diff --git a/backend/tests/test_update_from_file_snapshot.py b/backend/tests/test_update_from_file_snapshot.py
new file mode 100644
index 00000000..163111f1
--- /dev/null
+++ b/backend/tests/test_update_from_file_snapshot.py
@@ -0,0 +1,60 @@
+from typing import Any, cast
+
+from prompts.message_builder import Prompt
+from prompts.request_parsing import parse_prompt_content
+from prompts.update.from_file_snapshot import build_update_prompt_from_file_snapshot
+
+
+FILE_STATE = {"path": "index.html", "content": "<html><body><button>Buy</button></body></html>"}
+
+
+def _user_text(prompt_messages: Prompt) -> str:
+ content = prompt_messages[1].get("content")
+ if isinstance(content, str):
+ return content
+ parts = cast(list[dict[str, Any]], content)
+ return cast(str, parts[0]["text"])
+
+
+def test_snapshot_prompt_prefers_full_text_over_display_text() -> None:
+ prompt = parse_prompt_content(
+ {
+ "text": "Make this button red",
+ "fullText": 'Make this button red\n\nApply the change to this element: <button class="btn">Buy</button>',
+ "images": [],
+ "videos": [],
+ }
+ )
+
+ messages = build_update_prompt_from_file_snapshot(
+ stack="html_tailwind",
+ prompt=prompt,
+ file_state=FILE_STATE,
+ image_generation_enabled=False,
+ )
+
+ text = _user_text(messages)
+ assert '<button class="btn">Buy</button>' in text
+
+
+def test_snapshot_prompt_falls_back_to_display_text() -> None:
+ prompt = parse_prompt_content(
+ {"text": "Make the header blue", "images": [], "videos": []}
+ )
+
+ messages = build_update_prompt_from_file_snapshot(
+ stack="html_tailwind",
+ prompt=prompt,
+ file_state=FILE_STATE,
+ image_generation_enabled=False,
+ )
+
+ assert "Make the header blue" in _user_text(messages)
+
+
+def test_parse_prompt_content_ignores_blank_full_text() -> None:
+ prompt = parse_prompt_content(
+ {"text": "Hello", "fullText": " ", "images": [], "videos": []}
+ )
+
+ assert "full_text" not in prompt
diff --git a/frontend/src/App.tsx b/frontend/src/App.tsx
index 3fe11fd2..20040e26 100644
--- a/frontend/src/App.tsx
+++ b/frontend/src/App.tsx
@@ -678,6 +678,7 @@ function App() {
inputMode,
prompt: {
text: updateInstruction,
+ fullText: modifiedUpdateInstruction,
images: updateImages,
videos: [],
selectedElementHtml,
diff --git a/frontend/src/types.ts b/frontend/src/types.ts
index 1c715ba0..8a70ef76 100644
--- a/frontend/src/types.ts
+++ b/frontend/src/types.ts
@@ -57,7 +57,10 @@ export interface PromptAsset {
}
export interface PromptContent {
- text: string;
+ text: string; // What the user typed (displayed in the UI)
+ // Full instruction for the model when it differs from `text`
+ // (e.g. includes the selected-element reference)
+ fullText?: string;
images: string[]; // Array of data URLs
videos?: string[]; // Array of data URLs
selectedElementHtml?: string; // Raw HTML of selected element (for display only)