fix(frontend): simplify review UI - remove exclude button and tone down styling

- Remove confusing Exclude/Include button functionality
- Remove blue border/background from auto-approve toggle for cleaner look
- Toggle now shows by default with subtle gray text
- Disable editing when auto-approve is enabled (not when excluded)
- Remove unused rejection message functionality
- Simpler UX: just approve/reject reviews directly

autogpt_platform/backend/backend/api/features/chat/service.py

@@ -4,14 +4,9 @@ from collections.abc import AsyncGenerator
 from typing import Any
 
 import orjson
-from langfuse import Langfuse
-from openai import (
-    APIConnectionError,
-    APIError,
-    APIStatusError,
-    AsyncOpenAI,
-    RateLimitError,
-)
+from langfuse import get_client, propagate_attributes
+from langfuse.openai import openai  # type: ignore
+from openai import APIConnectionError, APIError, APIStatusError, RateLimitError
 from openai.types.chat import ChatCompletionChunk, ChatCompletionToolParam
 
 from backend.data.understanding import (
@@ -21,7 +16,6 @@ from backend.data.understanding import (
 from backend.util.exceptions import NotFoundError
 from backend.util.settings import Settings
 
-from . import db as chat_db
 from .config import ChatConfig
 from .model import (
     ChatMessage,
@@ -50,10 +44,10 @@ logger = logging.getLogger(__name__)
 
 config = ChatConfig()
 settings = Settings()
-client = AsyncOpenAI(api_key=config.api_key, base_url=config.base_url)
+client = openai.AsyncOpenAI(api_key=config.api_key, base_url=config.base_url)
 
-# Langfuse client (lazy initialization)
-_langfuse_client: Langfuse | None = None
+
+langfuse = get_client()
 
 
 class LangfuseNotConfiguredError(Exception):
@@ -69,65 +63,6 @@ def _is_langfuse_configured() -> bool:
     )
 
 
-def _get_langfuse_client() -> Langfuse:
-    """Get or create the Langfuse client for prompt management and tracing."""
-    global _langfuse_client
-    if _langfuse_client is None:
-        if not _is_langfuse_configured():
-            raise LangfuseNotConfiguredError(
-                "Langfuse is not configured. The chat feature requires Langfuse for prompt management. "
-                "Please set the LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY environment variables."
-            )
-        _langfuse_client = Langfuse(
-            public_key=settings.secrets.langfuse_public_key,
-            secret_key=settings.secrets.langfuse_secret_key,
-            host=settings.secrets.langfuse_host or "https://cloud.langfuse.com",
-        )
-    return _langfuse_client
-
-
-def _get_environment() -> str:
-    """Get the current environment name for Langfuse tagging."""
-    return settings.config.app_env.value
-
-
-def _get_langfuse_prompt() -> str:
-    """Fetch the latest production prompt from Langfuse.
-
-    Returns:
-        The compiled prompt text from Langfuse.
-
-    Raises:
-        Exception: If Langfuse is unavailable or prompt fetch fails.
-    """
-    try:
-        langfuse = _get_langfuse_client()
-        # cache_ttl_seconds=0 disables SDK caching to always get the latest prompt
-        prompt = langfuse.get_prompt(config.langfuse_prompt_name, cache_ttl_seconds=0)
-        compiled = prompt.compile()
-        logger.info(
-            f"Fetched prompt '{config.langfuse_prompt_name}' from Langfuse "
-            f"(version: {prompt.version})"
-        )
-        return compiled
-    except Exception as e:
-        logger.error(f"Failed to fetch prompt from Langfuse: {e}")
-        raise
-
-
-async def _is_first_session(user_id: str) -> bool:
-    """Check if this is the user's first chat session.
-
-    Returns True if the user has 1 or fewer sessions (meaning this is their first).
-    """
-    try:
-        session_count = await chat_db.get_user_session_count(user_id)
-        return session_count <= 1
-    except Exception as e:
-        logger.warning(f"Failed to check session count for user {user_id}: {e}")
-        return False  # Default to non-onboarding if we can't check
-
-
 async def _build_system_prompt(user_id: str | None) -> tuple[str, Any]:
     """Build the full system prompt including business understanding if available.
 
@@ -139,8 +74,6 @@ async def _build_system_prompt(user_id: str | None) -> tuple[str, Any]:
         Tuple of (compiled prompt string, Langfuse prompt object for tracing)
     """
 
-    langfuse = _get_langfuse_client()
-
     # cache_ttl_seconds=0 disables SDK caching to always get the latest prompt
     prompt = langfuse.get_prompt(config.langfuse_prompt_name, cache_ttl_seconds=0)
 
@@ -158,7 +91,7 @@ async def _build_system_prompt(user_id: str | None) -> tuple[str, Any]:
         context = "This is the first time you are meeting the user. Greet them and introduce them to the platform"
 
     compiled = prompt.compile(users_information=context)
-    return compiled, prompt
+    return compiled, understanding
 
 
 async def _generate_session_title(message: str) -> str | None:
@@ -217,6 +150,7 @@ async def assign_user_to_session(
 async def stream_chat_completion(
     session_id: str,
     message: str | None = None,
+    tool_call_response: str | None = None,
     is_user_message: bool = True,
     user_id: str | None = None,
     retry_count: int = 0,
@@ -256,11 +190,6 @@ async def stream_chat_completion(
         yield StreamFinish()
         return
 
-    # Langfuse observations will be created after session is loaded (need messages for input)
-    # Initialize to None so finally block can safely check and end them
-    trace = None
-    generation = None
-
     # Only fetch from Redis if session not provided (initial call)
     if session is None:
         session = await get_chat_session(session_id, user_id)
@@ -336,297 +265,259 @@ async def stream_chat_completion(
             asyncio.create_task(_update_title())
 
     # Build system prompt with business understanding
-    system_prompt, langfuse_prompt = await _build_system_prompt(user_id)
-
-    # Build input messages including system prompt for complete Langfuse logging
-    trace_input_messages = [{"role": "system", "content": system_prompt}] + [
-        m.model_dump() for m in session.messages
-    ]
+    system_prompt, understanding = await _build_system_prompt(user_id)
 
     # Create Langfuse trace for this LLM call (each call gets its own trace, grouped by session_id)
     # Using v3 SDK: start_observation creates a root span, update_trace sets trace-level attributes
-    try:
-        langfuse = _get_langfuse_client()
-        env = _get_environment()
-        trace = langfuse.start_observation(
-            name="chat_completion",
-            input={"messages": trace_input_messages},
-            metadata={
-                "environment": env,
-                "model": config.model,
-                "message_count": len(session.messages),
-                "prompt_name": langfuse_prompt.name if langfuse_prompt else None,
-                "prompt_version": langfuse_prompt.version if langfuse_prompt else None,
-            },
-        )
-        # Set trace-level attributes (session_id, user_id, tags)
-        trace.update_trace(
+    input = message
+    if not message and tool_call_response:
+        input = tool_call_response
+
+    langfuse = get_client()
+    with langfuse.start_as_current_observation(
+        as_type="span",
+        name="user-copilot-request",
+        input=input,
+    ) as span:
+        with propagate_attributes(
             session_id=session_id,
             user_id=user_id,
-            tags=[env, "copilot"],
-        )
-    except Exception as e:
-        logger.warning(f"Failed to create Langfuse trace: {e}")
+            tags=["copilot"],
+            metadata={
+                "users_information": format_understanding_for_prompt(understanding)[
+                    :200
+                ]  # langfuse only accepts upto to 200 chars
+            },
+        ):
 
-    # Initialize variables that will be used in finally block (must be defined before try)
-    assistant_response = ChatMessage(
-        role="assistant",
-        content="",
-    )
-    accumulated_tool_calls: list[dict[str, Any]] = []
-
-    # Wrap main logic in try/finally to ensure Langfuse observations are always ended
-    try:
-        has_yielded_end = False
-        has_yielded_error = False
-        has_done_tool_call = False
-        has_received_text = False
-        text_streaming_ended = False
-        tool_response_messages: list[ChatMessage] = []
-        should_retry = False
-
-        # Generate unique IDs for AI SDK protocol
-        import uuid as uuid_module
-
-        message_id = str(uuid_module.uuid4())
-        text_block_id = str(uuid_module.uuid4())
-
-        # Yield message start
-        yield StreamStart(messageId=message_id)
-
-        # Create Langfuse generation for each LLM call, linked to the prompt
-        # Using v3 SDK: start_observation with as_type="generation"
-        generation = (
-            trace.start_observation(
-                as_type="generation",
-                name="llm_call",
-                model=config.model,
-                input={"messages": trace_input_messages},
-                prompt=langfuse_prompt,
+            # Initialize variables that will be used in finally block (must be defined before try)
+            assistant_response = ChatMessage(
+                role="assistant",
+                content="",
             )
-            if trace
-            else None
-        )
+            accumulated_tool_calls: list[dict[str, Any]] = []
 
-        try:
-            async for chunk in _stream_chat_chunks(
-                session=session,
-                tools=tools,
-                system_prompt=system_prompt,
-                text_block_id=text_block_id,
-            ):
+            # Wrap main logic in try/finally to ensure Langfuse observations are always ended
+            has_yielded_end = False
+            has_yielded_error = False
+            has_done_tool_call = False
+            has_received_text = False
+            text_streaming_ended = False
+            tool_response_messages: list[ChatMessage] = []
+            should_retry = False
 
-                if isinstance(chunk, StreamTextStart):
-                    # Emit text-start before first text delta
-                    if not has_received_text:
+            # Generate unique IDs for AI SDK protocol
+            import uuid as uuid_module
+
+            message_id = str(uuid_module.uuid4())
+            text_block_id = str(uuid_module.uuid4())
+
+            # Yield message start
+            yield StreamStart(messageId=message_id)
+
+            try:
+                async for chunk in _stream_chat_chunks(
+                    session=session,
+                    tools=tools,
+                    system_prompt=system_prompt,
+                    text_block_id=text_block_id,
+                ):
+
+                    if isinstance(chunk, StreamTextStart):
+                        # Emit text-start before first text delta
+                        if not has_received_text:
+                            yield chunk
+                    elif isinstance(chunk, StreamTextDelta):
+                        delta = chunk.delta or ""
+                        assert assistant_response.content is not None
+                        assistant_response.content += delta
+                        has_received_text = True
                         yield chunk
-                elif isinstance(chunk, StreamTextDelta):
-                    delta = chunk.delta or ""
-                    assert assistant_response.content is not None
-                    assistant_response.content += delta
-                    has_received_text = True
-                    yield chunk
-                elif isinstance(chunk, StreamTextEnd):
-                    # Emit text-end after text completes
-                    if has_received_text and not text_streaming_ended:
-                        text_streaming_ended = True
-                        yield chunk
-                elif isinstance(chunk, StreamToolInputStart):
-                    # Emit text-end before first tool call, but only if we've received text
-                    if has_received_text and not text_streaming_ended:
-                        yield StreamTextEnd(id=text_block_id)
-                        text_streaming_ended = True
-                    yield chunk
-                elif isinstance(chunk, StreamToolInputAvailable):
-                    # Accumulate tool calls in OpenAI format
-                    accumulated_tool_calls.append(
-                        {
-                            "id": chunk.toolCallId,
-                            "type": "function",
-                            "function": {
-                                "name": chunk.toolName,
-                                "arguments": orjson.dumps(chunk.input).decode("utf-8"),
-                            },
-                        }
-                    )
-                elif isinstance(chunk, StreamToolOutputAvailable):
-                    result_content = (
-                        chunk.output
-                        if isinstance(chunk.output, str)
-                        else orjson.dumps(chunk.output).decode("utf-8")
-                    )
-                    tool_response_messages.append(
-                        ChatMessage(
-                            role="tool",
-                            content=result_content,
-                            tool_call_id=chunk.toolCallId,
-                        )
-                    )
-                    has_done_tool_call = True
-                    # Track if any tool execution failed
-                    if not chunk.success:
-                        logger.warning(
-                            f"Tool {chunk.toolName} (ID: {chunk.toolCallId}) execution failed"
-                        )
-                    yield chunk
-                elif isinstance(chunk, StreamFinish):
-                    if not has_done_tool_call:
-                        # Emit text-end before finish if we received text but haven't closed it
+                    elif isinstance(chunk, StreamTextEnd):
+                        # Emit text-end after text completes
+                        if has_received_text and not text_streaming_ended:
+                            text_streaming_ended = True
+                            if assistant_response.content:
+                                logger.warn(
+                                    f"StreamTextEnd: Attempting to set output {assistant_response.content}"
+                                )
+                                span.update_trace(output=assistant_response.content)
+                                span.update(output=assistant_response.content)
+                            yield chunk
+                    elif isinstance(chunk, StreamToolInputStart):
+                        # Emit text-end before first tool call, but only if we've received text
                         if has_received_text and not text_streaming_ended:
                             yield StreamTextEnd(id=text_block_id)
                             text_streaming_ended = True
-                        has_yielded_end = True
                         yield chunk
-                elif isinstance(chunk, StreamError):
-                    has_yielded_error = True
-                elif isinstance(chunk, StreamUsage):
-                    session.usage.append(
-                        Usage(
-                            prompt_tokens=chunk.promptTokens,
-                            completion_tokens=chunk.completionTokens,
-                            total_tokens=chunk.totalTokens,
+                    elif isinstance(chunk, StreamToolInputAvailable):
+                        # Accumulate tool calls in OpenAI format
+                        accumulated_tool_calls.append(
+                            {
+                                "id": chunk.toolCallId,
+                                "type": "function",
+                                "function": {
+                                    "name": chunk.toolName,
+                                    "arguments": orjson.dumps(chunk.input).decode(
+                                        "utf-8"
+                                    ),
+                                },
+                            }
                         )
+                    elif isinstance(chunk, StreamToolOutputAvailable):
+                        result_content = (
+                            chunk.output
+                            if isinstance(chunk.output, str)
+                            else orjson.dumps(chunk.output).decode("utf-8")
+                        )
+                        tool_response_messages.append(
+                            ChatMessage(
+                                role="tool",
+                                content=result_content,
+                                tool_call_id=chunk.toolCallId,
+                            )
+                        )
+                        has_done_tool_call = True
+                        # Track if any tool execution failed
+                        if not chunk.success:
+                            logger.warning(
+                                f"Tool {chunk.toolName} (ID: {chunk.toolCallId}) execution failed"
+                            )
+                        yield chunk
+                    elif isinstance(chunk, StreamFinish):
+                        if not has_done_tool_call:
+                            # Emit text-end before finish if we received text but haven't closed it
+                            if has_received_text and not text_streaming_ended:
+                                yield StreamTextEnd(id=text_block_id)
+                                text_streaming_ended = True
+                            has_yielded_end = True
+                            yield chunk
+                    elif isinstance(chunk, StreamError):
+                        has_yielded_error = True
+                    elif isinstance(chunk, StreamUsage):
+                        session.usage.append(
+                            Usage(
+                                prompt_tokens=chunk.promptTokens,
+                                completion_tokens=chunk.completionTokens,
+                                total_tokens=chunk.totalTokens,
+                            )
+                        )
+                    else:
+                        logger.error(
+                            f"Unknown chunk type: {type(chunk)}", exc_info=True
+                        )
+                if assistant_response.content:
+                    langfuse.update_current_trace(output=assistant_response.content)
+                    langfuse.update_current_span(output=assistant_response.content)
+                elif tool_response_messages:
+                    langfuse.update_current_trace(output=str(tool_response_messages))
+                    langfuse.update_current_span(output=str(tool_response_messages))
+
+            except Exception as e:
+                logger.error(f"Error during stream: {e!s}", exc_info=True)
+
+                # Check if this is a retryable error (JSON parsing, incomplete tool calls, etc.)
+                is_retryable = isinstance(
+                    e, (orjson.JSONDecodeError, KeyError, TypeError)
+                )
+
+                if is_retryable and retry_count < config.max_retries:
+                    logger.info(
+                        f"Retryable error encountered. Attempt {retry_count + 1}/{config.max_retries}"
                     )
+                    should_retry = True
                 else:
-                    logger.error(f"Unknown chunk type: {type(chunk)}", exc_info=True)
-        except Exception as e:
-            logger.error(f"Error during stream: {e!s}", exc_info=True)
+                    # Non-retryable error or max retries exceeded
+                    # Save any partial progress before reporting error
+                    messages_to_save: list[ChatMessage] = []
 
-            # Check if this is a retryable error (JSON parsing, incomplete tool calls, etc.)
-            is_retryable = isinstance(e, (orjson.JSONDecodeError, KeyError, TypeError))
+                    # Add assistant message if it has content or tool calls
+                    if accumulated_tool_calls:
+                        assistant_response.tool_calls = accumulated_tool_calls
+                    if assistant_response.content or assistant_response.tool_calls:
+                        messages_to_save.append(assistant_response)
 
-            if is_retryable and retry_count < config.max_retries:
+                    # Add tool response messages after assistant message
+                    messages_to_save.extend(tool_response_messages)
+
+                    session.messages.extend(messages_to_save)
+                    await upsert_chat_session(session)
+
+                    if not has_yielded_error:
+                        error_message = str(e)
+                        if not is_retryable:
+                            error_message = f"Non-retryable error: {error_message}"
+                        elif retry_count >= config.max_retries:
+                            error_message = f"Max retries ({config.max_retries}) exceeded: {error_message}"
+
+                        error_response = StreamError(errorText=error_message)
+                        yield error_response
+                    if not has_yielded_end:
+                        yield StreamFinish()
+                    return
+
+            # Handle retry outside of exception handler to avoid nesting
+            if should_retry and retry_count < config.max_retries:
                 logger.info(
-                    f"Retryable error encountered. Attempt {retry_count + 1}/{config.max_retries}"
+                    f"Retrying stream_chat_completion for session {session_id}, attempt {retry_count + 1}"
                 )
-                should_retry = True
-            else:
-                # Non-retryable error or max retries exceeded
-                # Save any partial progress before reporting error
-                messages_to_save: list[ChatMessage] = []
+                async for chunk in stream_chat_completion(
+                    session_id=session.session_id,
+                    user_id=user_id,
+                    retry_count=retry_count + 1,
+                    session=session,
+                    context=context,
+                ):
+                    yield chunk
+                return  # Exit after retry to avoid double-saving in finally block
 
-                # Add assistant message if it has content or tool calls
-                if accumulated_tool_calls:
-                    assistant_response.tool_calls = accumulated_tool_calls
-                if assistant_response.content or assistant_response.tool_calls:
-                    messages_to_save.append(assistant_response)
-
-                # Add tool response messages after assistant message
-                messages_to_save.extend(tool_response_messages)
-
-                session.messages.extend(messages_to_save)
-                await upsert_chat_session(session)
-
-                if not has_yielded_error:
-                    error_message = str(e)
-                    if not is_retryable:
-                        error_message = f"Non-retryable error: {error_message}"
-                    elif retry_count >= config.max_retries:
-                        error_message = f"Max retries ({config.max_retries}) exceeded: {error_message}"
-
-                    error_response = StreamError(errorText=error_message)
-                    yield error_response
-                if not has_yielded_end:
-                    yield StreamFinish()
-                return
-
-        # Handle retry outside of exception handler to avoid nesting
-        if should_retry and retry_count < config.max_retries:
+            # Normal completion path - save session and handle tool call continuation
             logger.info(
-                f"Retrying stream_chat_completion for session {session_id}, attempt {retry_count + 1}"
-            )
-            async for chunk in stream_chat_completion(
-                session_id=session.session_id,
-                user_id=user_id,
-                retry_count=retry_count + 1,
-                session=session,
-                context=context,
-            ):
-                yield chunk
-            return  # Exit after retry to avoid double-saving in finally block
-
-        # Normal completion path - save session and handle tool call continuation
-        logger.info(
-            f"Normal completion path: session={session.session_id}, "
-            f"current message_count={len(session.messages)}"
-        )
-
-        # Build the messages list in the correct order
-        messages_to_save: list[ChatMessage] = []
-
-        # Add assistant message with tool_calls if any
-        if accumulated_tool_calls:
-            assistant_response.tool_calls = accumulated_tool_calls
-            logger.info(
-                f"Added {len(accumulated_tool_calls)} tool calls to assistant message"
-            )
-        if assistant_response.content or assistant_response.tool_calls:
-            messages_to_save.append(assistant_response)
-            logger.info(
-                f"Saving assistant message with content_len={len(assistant_response.content or '')}, tool_calls={len(assistant_response.tool_calls or [])}"
+                f"Normal completion path: session={session.session_id}, "
+                f"current message_count={len(session.messages)}"
             )
 
-        # Add tool response messages after assistant message
-        messages_to_save.extend(tool_response_messages)
-        logger.info(
-            f"Saving {len(tool_response_messages)} tool response messages, "
-            f"total_to_save={len(messages_to_save)}"
-        )
+            # Build the messages list in the correct order
+            messages_to_save: list[ChatMessage] = []
 
-        session.messages.extend(messages_to_save)
-        logger.info(
-            f"Extended session messages, new message_count={len(session.messages)}"
-        )
-        await upsert_chat_session(session)
-
-        # If we did a tool call, stream the chat completion again to get the next response
-        if has_done_tool_call:
-            logger.info(
-                "Tool call executed, streaming chat completion again to get assistant response"
-            )
-            async for chunk in stream_chat_completion(
-                session_id=session.session_id,
-                user_id=user_id,
-                session=session,  # Pass session object to avoid Redis refetch
-                context=context,
-            ):
-                yield chunk
-
-    finally:
-        # Always end Langfuse observations to prevent resource leaks
-        # Guard against None and catch errors to avoid masking original exceptions
-        if generation is not None:
-            try:
-                latest_usage = session.usage[-1] if session.usage else None
-                generation.update(
-                    model=config.model,
-                    output={
-                        "content": assistant_response.content,
-                        "tool_calls": accumulated_tool_calls or None,
-                    },
-                    usage_details=(
-                        {
-                            "input": latest_usage.prompt_tokens,
-                            "output": latest_usage.completion_tokens,
-                            "total": latest_usage.total_tokens,
-                        }
-                        if latest_usage
-                        else None
-                    ),
+            # Add assistant message with tool_calls if any
+            if accumulated_tool_calls:
+                assistant_response.tool_calls = accumulated_tool_calls
+                logger.info(
+                    f"Added {len(accumulated_tool_calls)} tool calls to assistant message"
+                )
+            if assistant_response.content or assistant_response.tool_calls:
+                messages_to_save.append(assistant_response)
+                logger.info(
+                    f"Saving assistant message with content_len={len(assistant_response.content or '')}, tool_calls={len(assistant_response.tool_calls or [])}"
                 )
-                generation.end()
-            except Exception as e:
-                logger.warning(f"Failed to end Langfuse generation: {e}")
 
-        if trace is not None:
-            try:
-                if accumulated_tool_calls:
-                    trace.update_trace(output={"tool_calls": accumulated_tool_calls})
-                else:
-                    trace.update_trace(output={"response": assistant_response.content})
-                trace.end()
-            except Exception as e:
-                logger.warning(f"Failed to end Langfuse trace: {e}")
+            # Add tool response messages after assistant message
+            messages_to_save.extend(tool_response_messages)
+            logger.info(
+                f"Saving {len(tool_response_messages)} tool response messages, "
+                f"total_to_save={len(messages_to_save)}"
+            )
+
+            session.messages.extend(messages_to_save)
+            logger.info(
+                f"Extended session messages, new message_count={len(session.messages)}"
+            )
+            await upsert_chat_session(session)
+
+            # If we did a tool call, stream the chat completion again to get the next response
+            if has_done_tool_call:
+                logger.info(
+                    "Tool call executed, streaming chat completion again to get assistant response"
+                )
+                async for chunk in stream_chat_completion(
+                    session_id=session.session_id,
+                    user_id=user_id,
+                    session=session,  # Pass session object to avoid Redis refetch
+                    context=context,
+                    tool_call_response=str(tool_response_messages),
+                ):
+                    yield chunk
 
 
 # Retry configuration for OpenAI API calls
@@ -900,5 +791,4 @@ async def _yield_tool_call(
         session=session,
     )
 
-    logger.info(f"Yielding Tool execution response: {tool_execution_response}")
     yield tool_execution_response

autogpt_platform/backend/backend/api/features/chat/tools/__init__.py

@@ -30,7 +30,7 @@ TOOL_REGISTRY: dict[str, BaseTool] = {
     "find_library_agent": FindLibraryAgentTool(),
     "run_agent": RunAgentTool(),
     "run_block": RunBlockTool(),
-    "agent_output": AgentOutputTool(),
+    "view_agent_output": AgentOutputTool(),
     "search_docs": SearchDocsTool(),
     "get_doc_page": GetDocPageTool(),
 }

autogpt_platform/backend/backend/api/features/chat/tools/add_understanding.py

@@ -3,6 +3,8 @@
 import logging
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 from backend.data.understanding import (
     BusinessUnderstandingInput,
@@ -59,6 +61,7 @@ and automations for the user's specific needs."""
         """Requires authentication to store user-specific data."""
         return True
 
+    @observe(as_type="tool", name="add_understanding")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/agent_generator/core.py

@@ -218,6 +218,7 @@ async def save_agent_to_library(
     library_agents = await library_db.create_library_agent(
         graph=created_graph,
         user_id=user_id,
+        sensitive_action_safe_mode=True,
         create_library_agents_for_sub_graphs=False,
     )
 

autogpt_platform/backend/backend/api/features/chat/tools/agent_output.py

@@ -5,6 +5,7 @@ import re
 from datetime import datetime, timedelta, timezone
 from typing import Any
 
+from langfuse import observe
 from pydantic import BaseModel, field_validator
 
 from backend.api.features.chat.model import ChatSession
@@ -103,7 +104,7 @@ class AgentOutputTool(BaseTool):
 
     @property
     def name(self) -> str:
-        return "agent_output"
+        return "view_agent_output"
 
     @property
     def description(self) -> str:
@@ -328,6 +329,7 @@ class AgentOutputTool(BaseTool):
             total_executions=len(available_executions) if available_executions else 1,
         )
 
+    @observe(as_type="tool", name="view_agent_output")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/create_agent.py

@@ -3,6 +3,8 @@
 import logging
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 
 from .agent_generator import (
@@ -78,6 +80,7 @@ class CreateAgentTool(BaseTool):
             "required": ["description"],
         }
 
+    @observe(as_type="tool", name="create_agent")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/edit_agent.py

@@ -3,6 +3,8 @@
 import logging
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 
 from .agent_generator import (
@@ -85,6 +87,7 @@ class EditAgentTool(BaseTool):
             "required": ["agent_id", "changes"],
         }
 
+    @observe(as_type="tool", name="edit_agent")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/find_agent.py

@@ -2,6 +2,8 @@
 
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 
 from .agent_search import search_agents
@@ -35,6 +37,7 @@ class FindAgentTool(BaseTool):
             "required": ["query"],
         }
 
+    @observe(as_type="tool", name="find_agent")
     async def _execute(
         self, user_id: str | None, session: ChatSession, **kwargs
     ) -> ToolResponseBase:

autogpt_platform/backend/backend/api/features/chat/tools/find_block.py

@@ -1,6 +1,7 @@
 import logging
 from typing import Any
 
+from langfuse import observe
 from prisma.enums import ContentType
 
 from backend.api.features.chat.model import ChatSession
@@ -55,6 +56,7 @@ class FindBlockTool(BaseTool):
     def requires_auth(self) -> bool:
         return True
 
+    @observe(as_type="tool", name="find_block")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/find_library_agent.py

@@ -2,6 +2,8 @@
 
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 
 from .agent_search import search_agents
@@ -41,6 +43,7 @@ class FindLibraryAgentTool(BaseTool):
     def requires_auth(self) -> bool:
         return True
 
+    @observe(as_type="tool", name="find_library_agent")
     async def _execute(
         self, user_id: str | None, session: ChatSession, **kwargs
     ) -> ToolResponseBase:

autogpt_platform/backend/backend/api/features/chat/tools/get_doc_page.py

@@ -4,6 +4,8 @@ import logging
 from pathlib import Path
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 from backend.api.features.chat.tools.base import BaseTool
 from backend.api.features.chat.tools.models import (
@@ -71,6 +73,7 @@ class GetDocPageTool(BaseTool):
         url_path = path.rsplit(".", 1)[0] if "." in path else path
         return f"{DOCS_BASE_URL}/{url_path}"
 
+    @observe(as_type="tool", name="get_doc_page")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/run_agent.py

@@ -3,6 +3,7 @@
 import logging
 from typing import Any
 
+from langfuse import observe
 from pydantic import BaseModel, Field, field_validator
 
 from backend.api.features.chat.config import ChatConfig
@@ -32,7 +33,7 @@ from .models import (
     UserReadiness,
 )
 from .utils import (
-    check_user_has_required_credentials,
+    build_missing_credentials_from_graph,
     extract_credentials_from_schema,
     fetch_graph_from_store_slug,
     get_or_create_library_agent,
@@ -154,6 +155,7 @@ class RunAgentTool(BaseTool):
         """All operations require authentication."""
         return True
 
+    @observe(as_type="tool", name="run_agent")
     async def _execute(
         self,
         user_id: str | None,
@@ -235,15 +237,13 @@ class RunAgentTool(BaseTool):
                 # Return credentials needed response with input data info
                 # The UI handles credential setup automatically, so the message
                 # focuses on asking about input data
-                credentials = extract_credentials_from_schema(
-                    graph.credentials_input_schema
+                requirements_creds_dict = build_missing_credentials_from_graph(
+                    graph, None
                 )
-                missing_creds_check = await check_user_has_required_credentials(
-                    user_id, credentials
+                missing_credentials_dict = build_missing_credentials_from_graph(
+                    graph, graph_credentials
                 )
-                missing_credentials_dict = {
-                    c.id: c.model_dump() for c in missing_creds_check
-                }
+                requirements_creds_list = list(requirements_creds_dict.values())
 
                 return SetupRequirementsResponse(
                     message=self._build_inputs_message(graph, MSG_WHAT_VALUES_TO_USE),
@@ -257,7 +257,7 @@ class RunAgentTool(BaseTool):
                             ready_to_run=False,
                         ),
                         requirements={
-                            "credentials": [c.model_dump() for c in credentials],
+                            "credentials": requirements_creds_list,
                             "inputs": self._get_inputs_list(graph.input_schema),
                             "execution_modes": self._get_execution_modes(graph),
                         },

autogpt_platform/backend/backend/api/features/chat/tools/run_block.py

@@ -4,6 +4,8 @@ import logging
 from collections import defaultdict
 from typing import Any
 
+from langfuse import observe
+
 from backend.api.features.chat.model import ChatSession
 from backend.data.block import get_block
 from backend.data.execution import ExecutionContext
@@ -20,6 +22,7 @@ from .models import (
     ToolResponseBase,
     UserReadiness,
 )
+from .utils import build_missing_credentials_from_field_info
 
 logger = logging.getLogger(__name__)
 
@@ -127,6 +130,7 @@ class RunBlockTool(BaseTool):
 
         return matched_credentials, missing_credentials
 
+    @observe(as_type="tool", name="run_block")
     async def _execute(
         self,
         user_id: str | None,
@@ -186,7 +190,11 @@ class RunBlockTool(BaseTool):
 
         if missing_credentials:
             # Return setup requirements response with missing credentials
-            missing_creds_dict = {c.id: c.model_dump() for c in missing_credentials}
+            credentials_fields_info = block.input_schema.get_credentials_fields_info()
+            missing_creds_dict = build_missing_credentials_from_field_info(
+                credentials_fields_info, set(matched_credentials.keys())
+            )
+            missing_creds_list = list(missing_creds_dict.values())
 
             return SetupRequirementsResponse(
                 message=(
@@ -203,7 +211,7 @@ class RunBlockTool(BaseTool):
                         ready_to_run=False,
                     ),
                     requirements={
-                        "credentials": [c.model_dump() for c in missing_credentials],
+                        "credentials": missing_creds_list,
                         "inputs": self._get_inputs_list(block),
                         "execution_modes": ["immediate"],
                     },

autogpt_platform/backend/backend/api/features/chat/tools/search_docs.py

@@ -3,6 +3,7 @@
 import logging
 from typing import Any
 
+from langfuse import observe
 from prisma.enums import ContentType
 
 from backend.api.features.chat.model import ChatSession
@@ -87,6 +88,7 @@ class SearchDocsTool(BaseTool):
         url_path = path.rsplit(".", 1)[0] if "." in path else path
         return f"{DOCS_BASE_URL}/{url_path}"
 
+    @observe(as_type="tool", name="search_docs")
     async def _execute(
         self,
         user_id: str | None,

autogpt_platform/backend/backend/api/features/chat/tools/utils.py

@@ -8,7 +8,7 @@ from backend.api.features.library import model as library_model
 from backend.api.features.store import db as store_db
 from backend.data import graph as graph_db
 from backend.data.graph import GraphModel
-from backend.data.model import CredentialsMetaInput
+from backend.data.model import CredentialsFieldInfo, CredentialsMetaInput
 from backend.integrations.creds_manager import IntegrationCredentialsManager
 from backend.util.exceptions import NotFoundError
 
@@ -89,6 +89,59 @@ def extract_credentials_from_schema(
     return credentials
 
 
+def _serialize_missing_credential(
+    field_key: str, field_info: CredentialsFieldInfo
+) -> dict[str, Any]:
+    """
+    Convert credential field info into a serializable dict that preserves all supported
+    credential types (e.g., api_key + oauth2) so the UI can offer multiple options.
+    """
+    supported_types = sorted(field_info.supported_types)
+    provider = next(iter(field_info.provider), "unknown")
+    scopes = sorted(field_info.required_scopes or [])
+
+    return {
+        "id": field_key,
+        "title": field_key.replace("_", " ").title(),
+        "provider": provider,
+        "provider_name": provider.replace("_", " ").title(),
+        "type": supported_types[0] if supported_types else "api_key",
+        "types": supported_types,
+        "scopes": scopes,
+    }
+
+
+def build_missing_credentials_from_graph(
+    graph: GraphModel, matched_credentials: dict[str, CredentialsMetaInput] | None
+) -> dict[str, Any]:
+    """
+    Build a missing_credentials mapping from a graph's aggregated credentials inputs,
+    preserving all supported credential types for each field.
+    """
+    matched_keys = set(matched_credentials.keys()) if matched_credentials else set()
+    aggregated_fields = graph.aggregate_credentials_inputs()
+
+    return {
+        field_key: _serialize_missing_credential(field_key, field_info)
+        for field_key, (field_info, _node_fields) in aggregated_fields.items()
+        if field_key not in matched_keys
+    }
+
+
+def build_missing_credentials_from_field_info(
+    credential_fields: dict[str, CredentialsFieldInfo],
+    matched_keys: set[str],
+) -> dict[str, Any]:
+    """
+    Build missing_credentials mapping from a simple credentials field info dictionary.
+    """
+    return {
+        field_key: _serialize_missing_credential(field_key, field_info)
+        for field_key, field_info in credential_fields.items()
+        if field_key not in matched_keys
+    }
+
+
 def extract_credentials_as_dict(
     credentials_input_schema: dict[str, Any] | None,
 ) -> dict[str, CredentialsMetaInput]:

autogpt_platform/backend/backend/api/features/executions/review/model.py

@@ -107,6 +107,13 @@ class ReviewItem(BaseModel):
     reviewed_data: SafeJsonData | None = Field(
         None, description="Optional edited data (ignored if approved=False)"
     )
+    auto_approve_future: bool = Field(
+        default=False,
+        description=(
+            "If true and this review is approved, future executions of this same "
+            "block (node) will be automatically approved. This only affects approved reviews."
+        ),
+    )
 
     @field_validator("reviewed_data")
     @classmethod
@@ -174,6 +181,9 @@ class ReviewRequest(BaseModel):
     This request must include ALL pending reviews for a graph execution.
     Each review will be either approved (with optional data modifications)
     or rejected (data ignored). The execution will resume only after ALL reviews are processed.
+
+    Each review item can individually specify whether to auto-approve future executions
+    of the same block via the `auto_approve_future` field on ReviewItem.
     """
 
     reviews: List[ReviewItem] = Field(

autogpt_platform/backend/backend/api/features/executions/review/review_routes_test.py

@@ -8,6 +8,12 @@ from prisma.enums import ReviewStatus
 from pytest_snapshot.plugin import Snapshot
 
 from backend.api.rest_api import handle_internal_http_error
+from backend.data.execution import (
+    ExecutionContext,
+    ExecutionStatus,
+    NodeExecutionResult,
+)
+from backend.data.graph import GraphSettings
 
 from .model import PendingHumanReviewModel
 from .routes import router
@@ -15,20 +21,24 @@ from .routes import router
 # Using a fixed timestamp for reproducible tests
 FIXED_NOW = datetime.datetime(2023, 1, 1, 0, 0, 0, tzinfo=datetime.timezone.utc)
 
-app = fastapi.FastAPI()
-app.include_router(router, prefix="/api/review")
-app.add_exception_handler(ValueError, handle_internal_http_error(400))
 
-client = fastapi.testclient.TestClient(app)
+@pytest.fixture
+def app():
+    """Create FastAPI app for testing"""
+    test_app = fastapi.FastAPI()
+    test_app.include_router(router, prefix="/api/review")
+    test_app.add_exception_handler(ValueError, handle_internal_http_error(400))
+    return test_app
 
 
-@pytest.fixture(autouse=True)
-def setup_app_auth(mock_jwt_user):
-    """Setup auth overrides for all tests in this module"""
+@pytest.fixture
+def client(app, mock_jwt_user):
+    """Create test client with auth overrides"""
     from autogpt_libs.auth.jwt_utils import get_jwt_payload
 
     app.dependency_overrides[get_jwt_payload] = mock_jwt_user["get_jwt_payload"]
-    yield
+    with fastapi.testclient.TestClient(app) as test_client:
+        yield test_client
     app.dependency_overrides.clear()
 
 
@@ -55,6 +65,7 @@ def sample_pending_review(test_user_id: str) -> PendingHumanReviewModel:
 
 
 def test_get_pending_reviews_empty(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     snapshot: Snapshot,
     test_user_id: str,
@@ -73,6 +84,7 @@ def test_get_pending_reviews_empty(
 
 
 def test_get_pending_reviews_with_data(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     snapshot: Snapshot,
@@ -95,6 +107,7 @@ def test_get_pending_reviews_with_data(
 
 
 def test_get_pending_reviews_for_execution_success(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     snapshot: Snapshot,
@@ -123,6 +136,7 @@ def test_get_pending_reviews_for_execution_success(
 
 
 def test_get_pending_reviews_for_execution_not_available(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
 ) -> None:
     """Test access denied when user doesn't own the execution"""
@@ -138,6 +152,7 @@ def test_get_pending_reviews_for_execution_not_available(
 
 
 def test_process_review_action_approve_success(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     test_user_id: str,
@@ -145,6 +160,12 @@ def test_process_review_action_approve_success(
     """Test successful review approval"""
     # Mock the route functions
 
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = sample_pending_review
+
     mock_get_reviews_for_execution = mocker.patch(
         "backend.api.features.executions.review.routes.get_pending_reviews_for_execution"
     )
@@ -173,6 +194,14 @@ def test_process_review_action_approve_success(
     )
     mock_process_all_reviews.return_value = {"test_node_123": approved_review}
 
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
     mock_has_pending = mocker.patch(
         "backend.api.features.executions.review.routes.has_pending_reviews_for_graph_exec"
     )
@@ -202,6 +231,7 @@ def test_process_review_action_approve_success(
 
 
 def test_process_review_action_reject_success(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     test_user_id: str,
@@ -209,6 +239,20 @@ def test_process_review_action_reject_success(
     """Test successful review rejection"""
     # Mock the route functions
 
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = sample_pending_review
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
     mock_get_reviews_for_execution = mocker.patch(
         "backend.api.features.executions.review.routes.get_pending_reviews_for_execution"
     )
@@ -262,6 +306,7 @@ def test_process_review_action_reject_success(
 
 
 def test_process_review_action_mixed_success(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     test_user_id: str,
@@ -288,6 +333,12 @@ def test_process_review_action_mixed_success(
 
     # Mock the route functions
 
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = sample_pending_review
+
     mock_get_reviews_for_execution = mocker.patch(
         "backend.api.features.executions.review.routes.get_pending_reviews_for_execution"
     )
@@ -337,6 +388,14 @@ def test_process_review_action_mixed_success(
         "test_node_456": rejected_review,
     }
 
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
     mock_has_pending = mocker.patch(
         "backend.api.features.executions.review.routes.has_pending_reviews_for_graph_exec"
     )
@@ -369,6 +428,7 @@ def test_process_review_action_mixed_success(
 
 
 def test_process_review_action_empty_request(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     test_user_id: str,
 ) -> None:
@@ -386,10 +446,45 @@ def test_process_review_action_empty_request(
 
 
 def test_process_review_action_review_not_found(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
+    sample_pending_review: PendingHumanReviewModel,
     test_user_id: str,
 ) -> None:
     """Test error when review is not found"""
+    # Create a review with the nonexistent_node ID so the route can find the graph_exec_id
+    nonexistent_review = PendingHumanReviewModel(
+        node_exec_id="nonexistent_node",
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        payload={"data": "test"},
+        instructions="Review",
+        editable=True,
+        status=ReviewStatus.WAITING,
+        review_message=None,
+        was_edited=None,
+        processed=False,
+        created_at=FIXED_NOW,
+        updated_at=None,
+        reviewed_at=None,
+    )
+
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = nonexistent_review
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
     # Mock the functions that extract graph execution ID from the request
     mock_get_reviews_for_execution = mocker.patch(
         "backend.api.features.executions.review.routes.get_pending_reviews_for_execution"
@@ -422,11 +517,26 @@ def test_process_review_action_review_not_found(
 
 
 def test_process_review_action_partial_failure(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     test_user_id: str,
 ) -> None:
     """Test handling of partial failures in review processing"""
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = sample_pending_review
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
     # Mock the route functions
     mock_get_reviews_for_execution = mocker.patch(
         "backend.api.features.executions.review.routes.get_pending_reviews_for_execution"
@@ -456,16 +566,50 @@ def test_process_review_action_partial_failure(
 
 
 def test_process_review_action_invalid_node_exec_id(
+    client: fastapi.testclient.TestClient,
     mocker: pytest_mock.MockerFixture,
     sample_pending_review: PendingHumanReviewModel,
     test_user_id: str,
 ) -> None:
     """Test failure when trying to process review with invalid node execution ID"""
+    # Create a review with the invalid-node-format ID so the route can find the graph_exec_id
+    invalid_review = PendingHumanReviewModel(
+        node_exec_id="invalid-node-format",
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        payload={"data": "test"},
+        instructions="Review",
+        editable=True,
+        status=ReviewStatus.WAITING,
+        review_message=None,
+        was_edited=None,
+        processed=False,
+        created_at=FIXED_NOW,
+        updated_at=None,
+        reviewed_at=None,
+    )
+
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = invalid_review
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
     # Mock the route functions
     mock_get_reviews_for_execution = mocker.patch(
         "backend.api.features.executions.review.routes.get_pending_reviews_for_execution"
     )
-    mock_get_reviews_for_execution.return_value = [sample_pending_review]
+    mock_get_reviews_for_execution.return_value = [invalid_review]
 
     # Mock validation failure - this should return 400, not 500
     mock_process_all_reviews = mocker.patch(
@@ -490,3 +634,595 @@ def test_process_review_action_invalid_node_exec_id(
     # Should be a 400 Bad Request, not 500 Internal Server Error
     assert response.status_code == 400
     assert "Invalid node execution ID format" in response.json()["detail"]
+
+
+def test_process_review_action_auto_approve_creates_auto_approval_records(
+    client: fastapi.testclient.TestClient,
+    mocker: pytest_mock.MockerFixture,
+    sample_pending_review: PendingHumanReviewModel,
+    test_user_id: str,
+) -> None:
+    """Test that auto_approve_future_actions flag creates auto-approval records"""
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = sample_pending_review
+
+    # Mock process_all_reviews
+    mock_process_all_reviews = mocker.patch(
+        "backend.api.features.executions.review.routes.process_all_reviews_for_execution"
+    )
+    approved_review = PendingHumanReviewModel(
+        node_exec_id="test_node_123",
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        payload={"data": "test payload"},
+        instructions="Please review",
+        editable=True,
+        status=ReviewStatus.APPROVED,
+        review_message="Approved",
+        was_edited=False,
+        processed=False,
+        created_at=FIXED_NOW,
+        updated_at=FIXED_NOW,
+        reviewed_at=FIXED_NOW,
+    )
+    mock_process_all_reviews.return_value = {"test_node_123": approved_review}
+
+    # Mock get_node_execution to return node_id
+    mock_get_node_execution = mocker.patch(
+        "backend.api.features.executions.review.routes.get_node_execution"
+    )
+    mock_node_exec = mocker.Mock(spec=NodeExecutionResult)
+    mock_node_exec.node_id = "test_node_def_456"
+    mock_get_node_execution.return_value = mock_node_exec
+
+    # Mock create_auto_approval_record
+    mock_create_auto_approval = mocker.patch(
+        "backend.api.features.executions.review.routes.create_auto_approval_record"
+    )
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
+    # Mock has_pending_reviews_for_graph_exec
+    mock_has_pending = mocker.patch(
+        "backend.api.features.executions.review.routes.has_pending_reviews_for_graph_exec"
+    )
+    mock_has_pending.return_value = False
+
+    # Mock get_graph_settings to return custom settings
+    mock_get_settings = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_settings"
+    )
+    mock_get_settings.return_value = GraphSettings(
+        human_in_the_loop_safe_mode=True,
+        sensitive_action_safe_mode=True,
+    )
+
+    # Mock get_user_by_id to prevent database access
+    mock_get_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_user_by_id"
+    )
+    mock_user = mocker.Mock()
+    mock_user.timezone = "UTC"
+    mock_get_user.return_value = mock_user
+
+    # Mock add_graph_execution
+    mock_add_execution = mocker.patch(
+        "backend.api.features.executions.review.routes.add_graph_execution"
+    )
+
+    request_data = {
+        "reviews": [
+            {
+                "node_exec_id": "test_node_123",
+                "approved": True,
+                "message": "Approved",
+                "auto_approve_future": True,
+            }
+        ],
+    }
+
+    response = client.post("/api/review/action", json=request_data)
+
+    assert response.status_code == 200
+
+    # Verify process_all_reviews_for_execution was called (without auto_approve param)
+    mock_process_all_reviews.assert_called_once()
+
+    # Verify create_auto_approval_record was called for the approved review
+    mock_create_auto_approval.assert_called_once_with(
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        node_id="test_node_def_456",
+        payload={"data": "test payload"},
+    )
+
+    # Verify get_graph_settings was called with correct parameters
+    mock_get_settings.assert_called_once_with(
+        user_id=test_user_id, graph_id="test_graph_789"
+    )
+
+    # Verify add_graph_execution was called with proper ExecutionContext
+    mock_add_execution.assert_called_once()
+    call_kwargs = mock_add_execution.call_args.kwargs
+    execution_context = call_kwargs["execution_context"]
+
+    assert isinstance(execution_context, ExecutionContext)
+    assert execution_context.human_in_the_loop_safe_mode is True
+    assert execution_context.sensitive_action_safe_mode is True
+
+
+def test_process_review_action_without_auto_approve_still_loads_settings(
+    client: fastapi.testclient.TestClient,
+    mocker: pytest_mock.MockerFixture,
+    sample_pending_review: PendingHumanReviewModel,
+    test_user_id: str,
+) -> None:
+    """Test that execution context is created with settings even without auto-approve"""
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = sample_pending_review
+
+    # Mock process_all_reviews
+    mock_process_all_reviews = mocker.patch(
+        "backend.api.features.executions.review.routes.process_all_reviews_for_execution"
+    )
+    approved_review = PendingHumanReviewModel(
+        node_exec_id="test_node_123",
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        payload={"data": "test payload"},
+        instructions="Please review",
+        editable=True,
+        status=ReviewStatus.APPROVED,
+        review_message="Approved",
+        was_edited=False,
+        processed=False,
+        created_at=FIXED_NOW,
+        updated_at=FIXED_NOW,
+        reviewed_at=FIXED_NOW,
+    )
+    mock_process_all_reviews.return_value = {"test_node_123": approved_review}
+
+    # Mock create_auto_approval_record - should NOT be called when auto_approve is False
+    mock_create_auto_approval = mocker.patch(
+        "backend.api.features.executions.review.routes.create_auto_approval_record"
+    )
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
+    # Mock has_pending_reviews_for_graph_exec
+    mock_has_pending = mocker.patch(
+        "backend.api.features.executions.review.routes.has_pending_reviews_for_graph_exec"
+    )
+    mock_has_pending.return_value = False
+
+    # Mock get_graph_settings with sensitive_action_safe_mode enabled
+    mock_get_settings = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_settings"
+    )
+    mock_get_settings.return_value = GraphSettings(
+        human_in_the_loop_safe_mode=False,
+        sensitive_action_safe_mode=True,
+    )
+
+    # Mock get_user_by_id to prevent database access
+    mock_get_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_user_by_id"
+    )
+    mock_user = mocker.Mock()
+    mock_user.timezone = "UTC"
+    mock_get_user.return_value = mock_user
+
+    # Mock add_graph_execution
+    mock_add_execution = mocker.patch(
+        "backend.api.features.executions.review.routes.add_graph_execution"
+    )
+
+    # Request WITHOUT auto_approve_future (defaults to False)
+    request_data = {
+        "reviews": [
+            {
+                "node_exec_id": "test_node_123",
+                "approved": True,
+                "message": "Approved",
+                # auto_approve_future defaults to False
+            }
+        ],
+    }
+
+    response = client.post("/api/review/action", json=request_data)
+
+    assert response.status_code == 200
+
+    # Verify process_all_reviews_for_execution was called
+    mock_process_all_reviews.assert_called_once()
+
+    # Verify create_auto_approval_record was NOT called (auto_approve_future=False)
+    mock_create_auto_approval.assert_not_called()
+
+    # Verify settings were loaded
+    mock_get_settings.assert_called_once()
+
+    # Verify ExecutionContext has proper settings
+    mock_add_execution.assert_called_once()
+    call_kwargs = mock_add_execution.call_args.kwargs
+    execution_context = call_kwargs["execution_context"]
+
+    assert isinstance(execution_context, ExecutionContext)
+    assert execution_context.human_in_the_loop_safe_mode is False
+    assert execution_context.sensitive_action_safe_mode is True
+
+
+def test_process_review_action_auto_approve_only_applies_to_approved_reviews(
+    client: fastapi.testclient.TestClient,
+    mocker: pytest_mock.MockerFixture,
+    test_user_id: str,
+) -> None:
+    """Test that auto_approve record is created only for approved reviews"""
+    # Create two reviews - one approved, one rejected
+    approved_review = PendingHumanReviewModel(
+        node_exec_id="node_exec_approved",
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        payload={"data": "approved"},
+        instructions="Review",
+        editable=True,
+        status=ReviewStatus.APPROVED,
+        review_message=None,
+        was_edited=False,
+        processed=False,
+        created_at=FIXED_NOW,
+        updated_at=FIXED_NOW,
+        reviewed_at=FIXED_NOW,
+    )
+    rejected_review = PendingHumanReviewModel(
+        node_exec_id="node_exec_rejected",
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        payload={"data": "rejected"},
+        instructions="Review",
+        editable=True,
+        status=ReviewStatus.REJECTED,
+        review_message="Rejected",
+        was_edited=False,
+        processed=False,
+        created_at=FIXED_NOW,
+        updated_at=FIXED_NOW,
+        reviewed_at=FIXED_NOW,
+    )
+
+    # Mock get_pending_review_by_node_exec_id (called to find the graph_exec_id)
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+    mock_get_reviews_for_user.return_value = approved_review
+
+    # Mock process_all_reviews
+    mock_process_all_reviews = mocker.patch(
+        "backend.api.features.executions.review.routes.process_all_reviews_for_execution"
+    )
+    mock_process_all_reviews.return_value = {
+        "node_exec_approved": approved_review,
+        "node_exec_rejected": rejected_review,
+    }
+
+    # Mock get_node_execution to return node_id (only called for approved review)
+    mock_get_node_execution = mocker.patch(
+        "backend.api.features.executions.review.routes.get_node_execution"
+    )
+    mock_node_exec = mocker.Mock(spec=NodeExecutionResult)
+    mock_node_exec.node_id = "test_node_def_approved"
+    mock_get_node_execution.return_value = mock_node_exec
+
+    # Mock create_auto_approval_record
+    mock_create_auto_approval = mocker.patch(
+        "backend.api.features.executions.review.routes.create_auto_approval_record"
+    )
+
+    # Mock get_graph_execution_meta to return execution in REVIEW status
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
+    # Mock has_pending_reviews_for_graph_exec
+    mock_has_pending = mocker.patch(
+        "backend.api.features.executions.review.routes.has_pending_reviews_for_graph_exec"
+    )
+    mock_has_pending.return_value = False
+
+    # Mock get_graph_settings
+    mock_get_settings = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_settings"
+    )
+    mock_get_settings.return_value = GraphSettings()
+
+    # Mock get_user_by_id to prevent database access
+    mock_get_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_user_by_id"
+    )
+    mock_user = mocker.Mock()
+    mock_user.timezone = "UTC"
+    mock_get_user.return_value = mock_user
+
+    # Mock add_graph_execution
+    mock_add_execution = mocker.patch(
+        "backend.api.features.executions.review.routes.add_graph_execution"
+    )
+
+    request_data = {
+        "reviews": [
+            {
+                "node_exec_id": "node_exec_approved",
+                "approved": True,
+                "auto_approve_future": True,
+            },
+            {
+                "node_exec_id": "node_exec_rejected",
+                "approved": False,
+                "auto_approve_future": True,  # Should be ignored since rejected
+            },
+        ],
+    }
+
+    response = client.post("/api/review/action", json=request_data)
+
+    assert response.status_code == 200
+
+    # Verify process_all_reviews_for_execution was called
+    mock_process_all_reviews.assert_called_once()
+
+    # Verify create_auto_approval_record was called ONLY for the approved review
+    # (not for the rejected one)
+    mock_create_auto_approval.assert_called_once_with(
+        user_id=test_user_id,
+        graph_exec_id="test_graph_exec_456",
+        graph_id="test_graph_789",
+        graph_version=1,
+        node_id="test_node_def_approved",
+        payload={"data": "approved"},
+    )
+
+    # Verify get_node_execution was called only for approved review
+    mock_get_node_execution.assert_called_once_with("node_exec_approved")
+
+    # Verify ExecutionContext was created (auto-approval is now DB-based)
+    call_kwargs = mock_add_execution.call_args.kwargs
+    execution_context = call_kwargs["execution_context"]
+    assert isinstance(execution_context, ExecutionContext)
+
+
+def test_process_review_action_per_review_auto_approve_granularity(
+    client: fastapi.testclient.TestClient,
+    mocker: pytest_mock.MockerFixture,
+    sample_pending_review: PendingHumanReviewModel,
+    test_user_id: str,
+) -> None:
+    """Test that auto-approval can be set per-review (granular control)"""
+    # Mock get_pending_review_by_node_exec_id - return different reviews based on node_exec_id
+    mock_get_reviews_for_user = mocker.patch(
+        "backend.api.features.executions.review.routes.get_pending_review_by_node_exec_id"
+    )
+
+    # Create a mapping of node_exec_id to review
+    review_map = {
+        "node_1_auto": PendingHumanReviewModel(
+            node_exec_id="node_1_auto",
+            user_id=test_user_id,
+            graph_exec_id="test_graph_exec",
+            graph_id="test_graph",
+            graph_version=1,
+            payload={"data": "node1"},
+            instructions="Review 1",
+            editable=True,
+            status=ReviewStatus.WAITING,
+            review_message=None,
+            was_edited=False,
+            processed=False,
+            created_at=FIXED_NOW,
+        ),
+        "node_2_manual": PendingHumanReviewModel(
+            node_exec_id="node_2_manual",
+            user_id=test_user_id,
+            graph_exec_id="test_graph_exec",
+            graph_id="test_graph",
+            graph_version=1,
+            payload={"data": "node2"},
+            instructions="Review 2",
+            editable=True,
+            status=ReviewStatus.WAITING,
+            review_message=None,
+            was_edited=False,
+            processed=False,
+            created_at=FIXED_NOW,
+        ),
+        "node_3_auto": PendingHumanReviewModel(
+            node_exec_id="node_3_auto",
+            user_id=test_user_id,
+            graph_exec_id="test_graph_exec",
+            graph_id="test_graph",
+            graph_version=1,
+            payload={"data": "node3"},
+            instructions="Review 3",
+            editable=True,
+            status=ReviewStatus.WAITING,
+            review_message=None,
+            was_edited=False,
+            processed=False,
+            created_at=FIXED_NOW,
+        ),
+    }
+
+    # Use side_effect to return different reviews based on node_exec_id parameter
+    def mock_get_review_by_id(node_exec_id: str, _user_id: str):
+        return review_map.get(node_exec_id)
+
+    mock_get_reviews_for_user.side_effect = mock_get_review_by_id
+
+    # Mock process_all_reviews - return 3 approved reviews
+    mock_process_all_reviews = mocker.patch(
+        "backend.api.features.executions.review.routes.process_all_reviews_for_execution"
+    )
+    mock_process_all_reviews.return_value = {
+        "node_1_auto": PendingHumanReviewModel(
+            node_exec_id="node_1_auto",
+            user_id=test_user_id,
+            graph_exec_id="test_graph_exec",
+            graph_id="test_graph",
+            graph_version=1,
+            payload={"data": "node1"},
+            instructions="Review 1",
+            editable=True,
+            status=ReviewStatus.APPROVED,
+            review_message=None,
+            was_edited=False,
+            processed=False,
+            created_at=FIXED_NOW,
+            updated_at=FIXED_NOW,
+            reviewed_at=FIXED_NOW,
+        ),
+        "node_2_manual": PendingHumanReviewModel(
+            node_exec_id="node_2_manual",
+            user_id=test_user_id,
+            graph_exec_id="test_graph_exec",
+            graph_id="test_graph",
+            graph_version=1,
+            payload={"data": "node2"},
+            instructions="Review 2",
+            editable=True,
+            status=ReviewStatus.APPROVED,
+            review_message=None,
+            was_edited=False,
+            processed=False,
+            created_at=FIXED_NOW,
+            updated_at=FIXED_NOW,
+            reviewed_at=FIXED_NOW,
+        ),
+        "node_3_auto": PendingHumanReviewModel(
+            node_exec_id="node_3_auto",
+            user_id=test_user_id,
+            graph_exec_id="test_graph_exec",
+            graph_id="test_graph",
+            graph_version=1,
+            payload={"data": "node3"},
+            instructions="Review 3",
+            editable=True,
+            status=ReviewStatus.APPROVED,
+            review_message=None,
+            was_edited=False,
+            processed=False,
+            created_at=FIXED_NOW,
+            updated_at=FIXED_NOW,
+            reviewed_at=FIXED_NOW,
+        ),
+    }
+
+    # Mock get_node_execution
+    mock_get_node_execution = mocker.patch(
+        "backend.api.features.executions.review.routes.get_node_execution"
+    )
+
+    def mock_get_node(node_exec_id: str):
+        mock_node = mocker.Mock(spec=NodeExecutionResult)
+        mock_node.node_id = f"node_def_{node_exec_id}"
+        return mock_node
+
+    mock_get_node_execution.side_effect = mock_get_node
+
+    # Mock create_auto_approval_record
+    mock_create_auto_approval = mocker.patch(
+        "backend.api.features.executions.review.routes.create_auto_approval_record"
+    )
+
+    # Mock get_graph_execution_meta
+    mock_get_graph_exec = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_execution_meta"
+    )
+    mock_graph_exec_meta = mocker.Mock()
+    mock_graph_exec_meta.status = ExecutionStatus.REVIEW
+    mock_get_graph_exec.return_value = mock_graph_exec_meta
+
+    # Mock has_pending_reviews_for_graph_exec
+    mock_has_pending = mocker.patch(
+        "backend.api.features.executions.review.routes.has_pending_reviews_for_graph_exec"
+    )
+    mock_has_pending.return_value = False
+
+    # Mock settings and execution
+    mock_get_settings = mocker.patch(
+        "backend.api.features.executions.review.routes.get_graph_settings"
+    )
+    mock_get_settings.return_value = GraphSettings(
+        human_in_the_loop_safe_mode=False, sensitive_action_safe_mode=False
+    )
+
+    mocker.patch("backend.api.features.executions.review.routes.add_graph_execution")
+    mocker.patch("backend.api.features.executions.review.routes.get_user_by_id")
+
+    # Request with granular auto-approval:
+    # - node_1_auto: auto_approve_future=True
+    # - node_2_manual: auto_approve_future=False (explicit)
+    # - node_3_auto: auto_approve_future=True
+    request_data = {
+        "reviews": [
+            {
+                "node_exec_id": "node_1_auto",
+                "approved": True,
+                "auto_approve_future": True,
+            },
+            {
+                "node_exec_id": "node_2_manual",
+                "approved": True,
+                "auto_approve_future": False,  # Don't auto-approve this one
+            },
+            {
+                "node_exec_id": "node_3_auto",
+                "approved": True,
+                "auto_approve_future": True,
+            },
+        ],
+    }
+
+    response = client.post("/api/review/action", json=request_data)
+
+    assert response.status_code == 200
+
+    # Verify create_auto_approval_record was called ONLY for reviews with auto_approve_future=True
+    assert mock_create_auto_approval.call_count == 2
+
+    # Check that it was called for node_1 and node_3, but NOT node_2
+    call_args_list = [call.kwargs for call in mock_create_auto_approval.call_args_list]
+    node_ids_with_auto_approval = [args["node_id"] for args in call_args_list]
+
+    assert "node_def_node_1_auto" in node_ids_with_auto_approval
+    assert "node_def_node_3_auto" in node_ids_with_auto_approval
+    assert "node_def_node_2_manual" not in node_ids_with_auto_approval

autogpt_platform/backend/backend/api/features/executions/review/routes.py

@@ -5,13 +5,23 @@ import autogpt_libs.auth as autogpt_auth_lib
 from fastapi import APIRouter, HTTPException, Query, Security, status
 from prisma.enums import ReviewStatus
 
-from backend.data.execution import get_graph_execution_meta
+from backend.data.execution import (
+    ExecutionContext,
+    ExecutionStatus,
+    get_graph_execution_meta,
+    get_node_execution,
+)
+from backend.data.graph import get_graph_settings
 from backend.data.human_review import (
+    create_auto_approval_record,
+    get_pending_review_by_node_exec_id,
     get_pending_reviews_for_execution,
     get_pending_reviews_for_user,
     has_pending_reviews_for_graph_exec,
     process_all_reviews_for_execution,
 )
+from backend.data.model import USER_TIMEZONE_NOT_SET
+from backend.data.user import get_user_by_id
 from backend.executor.utils import add_graph_execution
 
 from .model import PendingHumanReviewModel, ReviewRequest, ReviewResponse
@@ -127,17 +137,80 @@ async def process_review_action(
             detail="At least one review must be provided",
         )
 
-    # Build review decisions map
+    # Get graph execution ID by looking up all requested reviews
+    # Use direct lookup to avoid pagination issues (can't miss reviews beyond first page)
+    # Also validate that all reviews belong to the same execution
+    matching_review = None
+    graph_exec_ids: set[str] = set()
+
+    for node_exec_id in all_request_node_ids:
+        review = await get_pending_review_by_node_exec_id(node_exec_id, user_id)
+        if not review:
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail=f"No pending review found for node execution {node_exec_id}",
+            )
+        if matching_review is None:
+            matching_review = review
+        graph_exec_ids.add(review.graph_exec_id)
+
+    # Ensure all reviews belong to the same execution
+    if len(graph_exec_ids) > 1:
+        raise HTTPException(
+            status_code=status.HTTP_409_CONFLICT,
+            detail="All reviews in a single request must belong to the same execution.",
+        )
+
+    # Safety check (matching_review should never be None here due to validation above)
+    if matching_review is None:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Internal error: No matching review found despite validation",
+        )
+
+    graph_exec_id = matching_review.graph_exec_id
+
+    # Validate execution status before processing reviews
+    graph_exec_meta = await get_graph_execution_meta(
+        user_id=user_id, execution_id=graph_exec_id
+    )
+
+    if not graph_exec_meta:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"Graph execution #{graph_exec_id} not found",
+        )
+
+    # Only allow processing reviews if execution is paused for review
+    # or incomplete (partial execution with some reviews already processed)
+    if graph_exec_meta.status not in (
+        ExecutionStatus.REVIEW,
+        ExecutionStatus.INCOMPLETE,
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_409_CONFLICT,
+            detail=f"Cannot process reviews while execution status is {graph_exec_meta.status}. "
+            f"Reviews can only be processed when execution is paused (REVIEW status). "
+            f"Current status: {graph_exec_meta.status}",
+        )
+
+    # Build review decisions map and track which reviews requested auto-approval
+    # Auto-approved reviews use original data (no modifications allowed)
     review_decisions = {}
+    auto_approve_requests = {}  # Map node_exec_id -> auto_approve_future flag
+
     for review in request.reviews:
         review_status = (
             ReviewStatus.APPROVED if review.approved else ReviewStatus.REJECTED
         )
+        # If this review requested auto-approval, don't allow data modifications
+        reviewed_data = None if review.auto_approve_future else review.reviewed_data
         review_decisions[review.node_exec_id] = (
             review_status,
-            review.reviewed_data,
+            reviewed_data,
             review.message,
         )
+        auto_approve_requests[review.node_exec_id] = review.auto_approve_future
 
     # Process all reviews
     updated_reviews = await process_all_reviews_for_execution(
@@ -145,6 +218,32 @@ async def process_review_action(
         review_decisions=review_decisions,
     )
 
+    # Create auto-approval records for approved reviews that requested it
+    # Note: Processing sequentially to avoid event loop issues in tests
+    for node_exec_id, review_result in updated_reviews.items():
+        # Only create auto-approval if:
+        # 1. This review was approved
+        # 2. The review requested auto-approval
+        if review_result.status == ReviewStatus.APPROVED and auto_approve_requests.get(
+            node_exec_id, False
+        ):
+            try:
+                node_exec = await get_node_execution(node_exec_id)
+                if node_exec:
+                    await create_auto_approval_record(
+                        user_id=user_id,
+                        graph_exec_id=review_result.graph_exec_id,
+                        graph_id=review_result.graph_id,
+                        graph_version=review_result.graph_version,
+                        node_id=node_exec.node_id,
+                        payload=review_result.payload,
+                    )
+            except Exception as e:
+                logger.error(
+                    f"Failed to create auto-approval record for {node_exec_id}",
+                    exc_info=e,
+                )
+
     # Count results
     approved_count = sum(
         1
@@ -157,22 +256,37 @@ async def process_review_action(
         if review.status == ReviewStatus.REJECTED
     )
 
-    # Resume execution if we processed some reviews
+    # Resume execution only if ALL pending reviews for this execution have been processed
     if updated_reviews:
-        # Get graph execution ID from any processed review
-        first_review = next(iter(updated_reviews.values()))
-        graph_exec_id = first_review.graph_exec_id
-
-        # Check if any pending reviews remain for this execution
         still_has_pending = await has_pending_reviews_for_graph_exec(graph_exec_id)
 
         if not still_has_pending:
-            # Resume execution
+            # Get the graph_id from any processed review
+            first_review = next(iter(updated_reviews.values()))
+
             try:
+                # Fetch user and settings to build complete execution context
+                user = await get_user_by_id(user_id)
+                settings = await get_graph_settings(
+                    user_id=user_id, graph_id=first_review.graph_id
+                )
+
+                # Preserve user's timezone preference when resuming execution
+                user_timezone = (
+                    user.timezone if user.timezone != USER_TIMEZONE_NOT_SET else "UTC"
+                )
+
+                execution_context = ExecutionContext(
+                    human_in_the_loop_safe_mode=settings.human_in_the_loop_safe_mode,
+                    sensitive_action_safe_mode=settings.sensitive_action_safe_mode,
+                    user_timezone=user_timezone,
+                )
+
                 await add_graph_execution(
                     graph_id=first_review.graph_id,
                     user_id=user_id,
                     graph_exec_id=graph_exec_id,
+                    execution_context=execution_context,
                 )
                 logger.info(f"Resumed execution {graph_exec_id}")
             except Exception as e:

autogpt_platform/backend/backend/api/features/library/db.py

@@ -401,27 +401,11 @@ async def add_generated_agent_image(
     )
 
 
-def _initialize_graph_settings(graph: graph_db.GraphModel) -> GraphSettings:
-    """
-    Initialize GraphSettings based on graph content.
-
-    Args:
-        graph: The graph to analyze
-
-    Returns:
-        GraphSettings with appropriate human_in_the_loop_safe_mode value
-    """
-    if graph.has_human_in_the_loop:
-        # Graph has HITL blocks - set safe mode to True by default
-        return GraphSettings(human_in_the_loop_safe_mode=True)
-    else:
-        # Graph has no HITL blocks - keep None
-        return GraphSettings(human_in_the_loop_safe_mode=None)
-
-
 async def create_library_agent(
     graph: graph_db.GraphModel,
     user_id: str,
+    hitl_safe_mode: bool = True,
+    sensitive_action_safe_mode: bool = False,
     create_library_agents_for_sub_graphs: bool = True,
 ) -> list[library_model.LibraryAgent]:
     """
@@ -430,6 +414,8 @@ async def create_library_agent(
     Args:
         agent: The agent/Graph to add to the library.
         user_id: The user to whom the agent will be added.
+        hitl_safe_mode: Whether HITL blocks require manual review (default True).
+        sensitive_action_safe_mode: Whether sensitive action blocks require review.
         create_library_agents_for_sub_graphs: If True, creates LibraryAgent records for sub-graphs as well.
 
     Returns:
@@ -465,7 +451,11 @@ async def create_library_agent(
                             }
                         },
                         settings=SafeJson(
-                            _initialize_graph_settings(graph_entry).model_dump()
+                            GraphSettings.from_graph(
+                                graph_entry,
+                                hitl_safe_mode=hitl_safe_mode,
+                                sensitive_action_safe_mode=sensitive_action_safe_mode,
+                            ).model_dump()
                         ),
                     ),
                     include=library_agent_include(
@@ -627,33 +617,6 @@ async def update_library_agent(
         raise DatabaseError("Failed to update library agent") from e
 
 
-async def update_library_agent_settings(
-    user_id: str,
-    agent_id: str,
-    settings: GraphSettings,
-) -> library_model.LibraryAgent:
-    """
-    Updates the settings for a specific LibraryAgent.
-
-    Args:
-        user_id: The owner of the LibraryAgent.
-        agent_id: The ID of the LibraryAgent to update.
-        settings: New GraphSettings to apply.
-
-    Returns:
-        The updated LibraryAgent.
-
-    Raises:
-        NotFoundError: If the specified LibraryAgent does not exist.
-        DatabaseError: If there's an error in the update operation.
-    """
-    return await update_library_agent(
-        library_agent_id=agent_id,
-        user_id=user_id,
-        settings=settings,
-    )
-
-
 async def delete_library_agent(
     library_agent_id: str, user_id: str, soft_delete: bool = True
 ) -> None:
@@ -838,7 +801,7 @@ async def add_store_agent_to_library(
                 "isCreatedByUser": False,
                 "useGraphIsActiveVersion": False,
                 "settings": SafeJson(
-                    _initialize_graph_settings(graph_model).model_dump()
+                    GraphSettings.from_graph(graph_model).model_dump()
                 ),
             },
             include=library_agent_include(
@@ -1228,8 +1191,15 @@ async def fork_library_agent(
         )
         new_graph = await on_graph_activate(new_graph, user_id=user_id)
 
-        # Create a library agent for the new graph
-        return (await create_library_agent(new_graph, user_id))[0]
+        # Create a library agent for the new graph, preserving safe mode settings
+        return (
+            await create_library_agent(
+                new_graph,
+                user_id,
+                hitl_safe_mode=original_agent.settings.human_in_the_loop_safe_mode,
+                sensitive_action_safe_mode=original_agent.settings.sensitive_action_safe_mode,
+            )
+        )[0]
     except prisma.errors.PrismaError as e:
         logger.error(f"Database error cloning library agent: {e}")
         raise DatabaseError("Failed to fork library agent") from e

autogpt_platform/backend/backend/api/features/library/model.py

@@ -73,6 +73,12 @@ class LibraryAgent(pydantic.BaseModel):
     has_external_trigger: bool = pydantic.Field(
         description="Whether the agent has an external trigger (e.g. webhook) node"
     )
+    has_human_in_the_loop: bool = pydantic.Field(
+        description="Whether the agent has human-in-the-loop blocks"
+    )
+    has_sensitive_action: bool = pydantic.Field(
+        description="Whether the agent has sensitive action blocks"
+    )
     trigger_setup_info: Optional[GraphTriggerInfo] = None
 
     # Indicates whether there's a new output (based on recent runs)
@@ -180,6 +186,8 @@ class LibraryAgent(pydantic.BaseModel):
                 graph.credentials_input_schema if sub_graphs is not None else None
             ),
             has_external_trigger=graph.has_external_trigger,
+            has_human_in_the_loop=graph.has_human_in_the_loop,
+            has_sensitive_action=graph.has_sensitive_action,
             trigger_setup_info=graph.trigger_setup_info,
             new_output=new_output,
             can_access_graph=can_access_graph,

autogpt_platform/backend/backend/api/features/library/routes_test.py

@@ -52,6 +52,8 @@ async def test_get_library_agents_success(
                 output_schema={"type": "object", "properties": {}},
                 credentials_input_schema={"type": "object", "properties": {}},
                 has_external_trigger=False,
+                has_human_in_the_loop=False,
+                has_sensitive_action=False,
                 status=library_model.LibraryAgentStatus.COMPLETED,
                 recommended_schedule_cron=None,
                 new_output=False,
@@ -75,6 +77,8 @@ async def test_get_library_agents_success(
                 output_schema={"type": "object", "properties": {}},
                 credentials_input_schema={"type": "object", "properties": {}},
                 has_external_trigger=False,
+                has_human_in_the_loop=False,
+                has_sensitive_action=False,
                 status=library_model.LibraryAgentStatus.COMPLETED,
                 recommended_schedule_cron=None,
                 new_output=False,
@@ -150,6 +154,8 @@ async def test_get_favorite_library_agents_success(
                 output_schema={"type": "object", "properties": {}},
                 credentials_input_schema={"type": "object", "properties": {}},
                 has_external_trigger=False,
+                has_human_in_the_loop=False,
+                has_sensitive_action=False,
                 status=library_model.LibraryAgentStatus.COMPLETED,
                 recommended_schedule_cron=None,
                 new_output=False,
@@ -218,6 +224,8 @@ def test_add_agent_to_library_success(
         output_schema={"type": "object", "properties": {}},
         credentials_input_schema={"type": "object", "properties": {}},
         has_external_trigger=False,
+        has_human_in_the_loop=False,
+        has_sensitive_action=False,
         status=library_model.LibraryAgentStatus.COMPLETED,
         new_output=False,
         can_access_graph=True,

autogpt_platform/backend/backend/api/features/store/embeddings.py

@@ -6,6 +6,7 @@ Handles generation and storage of OpenAI embeddings for all content types
 """
 
 import asyncio
+import contextvars
 import logging
 import time
 from typing import Any
@@ -21,6 +22,11 @@ from backend.util.json import dumps
 
 logger = logging.getLogger(__name__)
 
+# Context variable to track errors logged in the current task/operation
+# This prevents spamming the same error multiple times when processing batches
+_logged_errors: contextvars.ContextVar[set[str]] = contextvars.ContextVar(
+    "_logged_errors"
+)
 
 # OpenAI embedding model configuration
 EMBEDDING_MODEL = "text-embedding-3-small"
@@ -31,6 +37,42 @@ EMBEDDING_DIM = 1536
 EMBEDDING_MAX_TOKENS = 8191
 
 
+def log_once_per_task(error_key: str, log_fn, message: str, **kwargs) -> bool:
+    """
+    Log an error/warning only once per task/operation to avoid log spam.
+
+    Uses contextvars to track what has been logged in the current async context.
+    Useful when processing batches where the same error might occur for many items.
+
+    Args:
+        error_key: Unique identifier for this error type
+        log_fn: Logger function to call (e.g., logger.error, logger.warning)
+        message: Message to log
+        **kwargs: Additional arguments to pass to log_fn
+
+    Returns:
+        True if the message was logged, False if it was suppressed (already logged)
+
+    Example:
+        log_once_per_task("missing_api_key", logger.error, "API key not set")
+    """
+    # Get current logged errors, or create a new set if this is the first call in this context
+    logged = _logged_errors.get(None)
+    if logged is None:
+        logged = set()
+        _logged_errors.set(logged)
+
+    if error_key in logged:
+        return False
+
+    # Log the message with a note that it will only appear once
+    log_fn(f"{message} (This message will only be shown once per task.)", **kwargs)
+
+    # Mark as logged
+    logged.add(error_key)
+    return True
+
+
 def build_searchable_text(
     name: str,
     description: str,
@@ -73,7 +115,11 @@ async def generate_embedding(text: str) -> list[float] | None:
     try:
         client = get_openai_client()
         if not client:
-            logger.error("openai_internal_api_key not set, cannot generate embedding")
+            log_once_per_task(
+                "openai_api_key_missing",
+                logger.error,
+                "openai_internal_api_key not set, cannot generate embeddings",
+            )
             return None
 
         # Truncate text to token limit using tiktoken
@@ -154,6 +200,7 @@ async def store_content_embedding(
 
         # Upsert the embedding
         # WHERE clause in DO UPDATE prevents PostgreSQL 15 bug with NULLS NOT DISTINCT
+        # Use unqualified ::vector - pgvector is in search_path on all environments
         await execute_raw_with_schema(
             """
             INSERT INTO {schema_prefix}"UnifiedContentEmbedding" (
@@ -177,7 +224,6 @@ async def store_content_embedding(
             searchable_text,
             metadata_json,
             client=client,
-            set_public_search_path=True,
         )
 
         logger.info(f"Stored embedding for {content_type}:{content_id}")
@@ -236,7 +282,6 @@ async def get_content_embedding(
             content_type,
             content_id,
             user_id,
-            set_public_search_path=True,
         )
 
         if result and len(result) > 0:
@@ -291,7 +336,12 @@ async def ensure_embedding(
         # Generate new embedding
         embedding = await generate_embedding(searchable_text)
         if embedding is None:
-            logger.warning(f"Could not generate embedding for version {version_id}")
+            log_once_per_task(
+                "embedding_generation_failed",
+                logger.warning,
+                "Could not generate embeddings (missing API key or service unavailable). "
+                "Embedding generation is disabled for this task.",
+            )
             return False
 
         # Store the embedding with metadata using new function
@@ -610,8 +660,11 @@ async def ensure_content_embedding(
         # Generate new embedding
         embedding = await generate_embedding(searchable_text)
         if embedding is None:
-            logger.warning(
-                f"Could not generate embedding for {content_type}:{content_id}"
+            log_once_per_task(
+                "embedding_generation_failed",
+                logger.warning,
+                "Could not generate embeddings (missing API key or service unavailable). "
+                "Embedding generation is disabled for this task.",
             )
             return False
 
@@ -871,31 +924,45 @@ async def semantic_search(
         # Add content type parameters and build placeholders dynamically
         content_type_start_idx = len(params) + 1
         content_type_placeholders = ", ".join(
-            f'${content_type_start_idx + i}::{{{{schema_prefix}}}}"ContentType"'
+            "$" + str(content_type_start_idx + i) + '::{schema_prefix}"ContentType"'
             for i in range(len(content_types))
         )
         params.extend([ct.value for ct in content_types])
 
-        sql = f"""
+        # Build min_similarity param index before appending
+        min_similarity_idx = len(params) + 1
+        params.append(min_similarity)
+
+        # Use unqualified ::vector and <=> operator - pgvector is in search_path on all environments
+        sql = (
+            """
             SELECT
                 "contentId" as content_id,
                 "contentType" as content_type,
                 "searchableText" as searchable_text,
                 metadata,
-                1 - (embedding <=> '{embedding_str}'::vector) as similarity
-            FROM {{{{schema_prefix}}}}"UnifiedContentEmbedding"
-            WHERE "contentType" IN ({content_type_placeholders})
-            {user_filter}
-            AND 1 - (embedding <=> '{embedding_str}'::vector) >= ${len(params) + 1}
+                1 - (embedding <=> '"""
+            + embedding_str
+            + """'::vector) as similarity
+            FROM {schema_prefix}"UnifiedContentEmbedding"
+            WHERE "contentType" IN ("""
+            + content_type_placeholders
+            + """)
+            """
+            + user_filter
+            + """
+            AND 1 - (embedding <=> '"""
+            + embedding_str
+            + """'::vector) >= $"""
+            + str(min_similarity_idx)
+            + """
             ORDER BY similarity DESC
             LIMIT $1
         """
-        params.append(min_similarity)
+        )
 
         try:
-            results = await query_raw_with_schema(
-                sql, *params, set_public_search_path=True
-            )
+            results = await query_raw_with_schema(sql, *params)
             return [
                 {
                     "content_id": row["content_id"],
@@ -922,31 +989,41 @@ async def semantic_search(
     # Add content type parameters and build placeholders dynamically
     content_type_start_idx = len(params_lexical) + 1
     content_type_placeholders_lexical = ", ".join(
-        f'${content_type_start_idx + i}::{{{{schema_prefix}}}}"ContentType"'
+        "$" + str(content_type_start_idx + i) + '::{schema_prefix}"ContentType"'
         for i in range(len(content_types))
     )
     params_lexical.extend([ct.value for ct in content_types])
 
-    sql_lexical = f"""
+    # Build query param index before appending
+    query_param_idx = len(params_lexical) + 1
+    params_lexical.append(f"%{query}%")
+
+    # Use regular string (not f-string) for template to preserve {schema_prefix} placeholders
+    sql_lexical = (
+        """
         SELECT
             "contentId" as content_id,
             "contentType" as content_type,
             "searchableText" as searchable_text,
             metadata,
             0.0 as similarity
-        FROM {{{{schema_prefix}}}}"UnifiedContentEmbedding"
-        WHERE "contentType" IN ({content_type_placeholders_lexical})
-        {user_filter}
-        AND "searchableText" ILIKE ${len(params_lexical) + 1}
+        FROM {schema_prefix}"UnifiedContentEmbedding"
+        WHERE "contentType" IN ("""
+        + content_type_placeholders_lexical
+        + """)
+        """
+        + user_filter
+        + """
+        AND "searchableText" ILIKE $"""
+        + str(query_param_idx)
+        + """
         ORDER BY "updatedAt" DESC
         LIMIT $1
     """
-    params_lexical.append(f"%{query}%")
+    )
 
     try:
-        results = await query_raw_with_schema(
-            sql_lexical, *params_lexical, set_public_search_path=True
-        )
+        results = await query_raw_with_schema(sql_lexical, *params_lexical)
         return [
             {
                 "content_id": row["content_id"],

autogpt_platform/backend/backend/api/features/store/embeddings_test.py

@@ -155,18 +155,14 @@ async def test_store_embedding_success(mocker):
     )
 
     assert result is True
-    # execute_raw is called twice: once for SET search_path, once for INSERT
-    assert mock_client.execute_raw.call_count == 2
+    # execute_raw is called once for INSERT (no separate SET search_path needed)
+    assert mock_client.execute_raw.call_count == 1
 
-    # First call: SET search_path
-    first_call_args = mock_client.execute_raw.call_args_list[0][0]
-    assert "SET search_path" in first_call_args[0]
-
-    # Second call: INSERT query with the actual data
-    second_call_args = mock_client.execute_raw.call_args_list[1][0]
-    assert "test-version-id" in second_call_args
-    assert "[0.1,0.2,0.3]" in second_call_args
-    assert None in second_call_args  # userId should be None for store agents
+    # Verify the INSERT query with the actual data
+    call_args = mock_client.execute_raw.call_args_list[0][0]
+    assert "test-version-id" in call_args
+    assert "[0.1,0.2,0.3]" in call_args
+    assert None in call_args  # userId should be None for store agents
 
 
 @pytest.mark.asyncio(loop_scope="session")

autogpt_platform/backend/backend/api/features/store/hybrid_search.py

@@ -12,7 +12,7 @@ from dataclasses import dataclass
 from typing import Any, Literal
 
 from prisma.enums import ContentType
-from rank_bm25 import BM25Okapi
+from rank_bm25 import BM25Okapi  # type: ignore[import-untyped]
 
 from backend.api.features.store.embeddings import (
     EMBEDDING_DIM,
@@ -363,9 +363,7 @@ async def unified_hybrid_search(
         LIMIT {limit_param} OFFSET {offset_param}
     """
 
-    results = await query_raw_with_schema(
-        sql_query, *params, set_public_search_path=True
-    )
+    results = await query_raw_with_schema(sql_query, *params)
 
     total = results[0]["total_count"] if results else 0
     # Apply BM25 reranking
@@ -688,9 +686,7 @@ async def hybrid_search(
         LIMIT {limit_param} OFFSET {offset_param}
     """
 
-    results = await query_raw_with_schema(
-        sql_query, *params, set_public_search_path=True
-    )
+    results = await query_raw_with_schema(sql_query, *params)
 
     total = results[0]["total_count"] if results else 0
 

autogpt_platform/backend/backend/api/features/v1.py

@@ -761,10 +761,8 @@ async def create_new_graph(
     graph.reassign_ids(user_id=user_id, reassign_graph_id=True)
     graph.validate_graph(for_run=False)
 
-    # The return value of the create graph & library function is intentionally not used here,
-    # as the graph already valid and no sub-graphs are returned back.
     await graph_db.create_graph(graph, user_id=user_id)
-    await library_db.create_library_agent(graph, user_id=user_id)
+    await library_db.create_library_agent(graph, user_id)
     activated_graph = await on_graph_activate(graph, user_id=user_id)
 
     if create_graph.source == "builder":
@@ -888,21 +886,19 @@ async def set_graph_active_version(
 async def _update_library_agent_version_and_settings(
     user_id: str, agent_graph: graph_db.GraphModel
 ) -> library_model.LibraryAgent:
-    # Keep the library agent up to date with the new active version
     library = await library_db.update_agent_version_in_library(
         user_id, agent_graph.id, agent_graph.version
     )
-    # If the graph has HITL node, initialize the setting if it's not already set.
-    if (
-        agent_graph.has_human_in_the_loop
-        and library.settings.human_in_the_loop_safe_mode is None
-    ):
-        await library_db.update_library_agent_settings(
+    updated_settings = GraphSettings.from_graph(
+        graph=agent_graph,
+        hitl_safe_mode=library.settings.human_in_the_loop_safe_mode,
+        sensitive_action_safe_mode=library.settings.sensitive_action_safe_mode,
+    )
+    if updated_settings != library.settings:
+        library = await library_db.update_library_agent(
+            library_agent_id=library.id,
             user_id=user_id,
-            agent_id=library.id,
-            settings=library.settings.model_copy(
-                update={"human_in_the_loop_safe_mode": True}
-            ),
+            settings=updated_settings,
         )
     return library
 
@@ -919,21 +915,18 @@ async def update_graph_settings(
     user_id: Annotated[str, Security(get_user_id)],
 ) -> GraphSettings:
     """Update graph settings for the user's library agent."""
-    # Get the library agent for this graph
     library_agent = await library_db.get_library_agent_by_graph_id(
         graph_id=graph_id, user_id=user_id
     )
     if not library_agent:
         raise HTTPException(404, f"Graph #{graph_id} not found in user's library")
 
-    # Update the library agent settings
-    updated_agent = await library_db.update_library_agent_settings(
+    updated_agent = await library_db.update_library_agent(
+        library_agent_id=library_agent.id,
         user_id=user_id,
-        agent_id=library_agent.id,
         settings=settings,
     )
 
-    # Return the updated settings
     return GraphSettings.model_validate(updated_agent.settings)
 
 

autogpt_platform/backend/backend/blocks/basic.py

@@ -116,6 +116,7 @@ class PrintToConsoleBlock(Block):
             input_schema=PrintToConsoleBlock.Input,
             output_schema=PrintToConsoleBlock.Output,
             test_input={"text": "Hello, World!"},
+            is_sensitive_action=True,
             test_output=[
                 ("output", "Hello, World!"),
                 ("status", "printed"),

autogpt_platform/backend/backend/blocks/data_manipulation.py

@@ -680,3 +680,58 @@ class ListIsEmptyBlock(Block):
 
     async def run(self, input_data: Input, **kwargs) -> BlockOutput:
         yield "is_empty", len(input_data.list) == 0
+
+
+class ConcatenateListsBlock(Block):
+    class Input(BlockSchemaInput):
+        lists: List[List[Any]] = SchemaField(
+            description="A list of lists to concatenate together. All lists will be combined in order into a single list.",
+            placeholder="e.g., [[1, 2], [3, 4], [5, 6]]",
+        )
+
+    class Output(BlockSchemaOutput):
+        concatenated_list: List[Any] = SchemaField(
+            description="The concatenated list containing all elements from all input lists in order."
+        )
+        error: str = SchemaField(
+            description="Error message if concatenation failed due to invalid input types."
+        )
+
+    def __init__(self):
+        super().__init__(
+            id="3cf9298b-5817-4141-9d80-7c2cc5199c8e",
+            description="Concatenates multiple lists into a single list. All elements from all input lists are combined in order.",
+            categories={BlockCategory.BASIC},
+            input_schema=ConcatenateListsBlock.Input,
+            output_schema=ConcatenateListsBlock.Output,
+            test_input=[
+                {"lists": [[1, 2, 3], [4, 5, 6]]},
+                {"lists": [["a", "b"], ["c"], ["d", "e", "f"]]},
+                {"lists": [[1, 2], []]},
+                {"lists": []},
+            ],
+            test_output=[
+                ("concatenated_list", [1, 2, 3, 4, 5, 6]),
+                ("concatenated_list", ["a", "b", "c", "d", "e", "f"]),
+                ("concatenated_list", [1, 2]),
+                ("concatenated_list", []),
+            ],
+        )
+
+    async def run(self, input_data: Input, **kwargs) -> BlockOutput:
+        concatenated = []
+        for idx, lst in enumerate(input_data.lists):
+            if lst is None:
+                # Skip None values to avoid errors
+                continue
+            if not isinstance(lst, list):
+                # Type validation: each item must be a list
+                # Strings are iterable and would cause extend() to iterate character-by-character
+                # Non-iterable types would raise TypeError
+                yield "error", (
+                    f"Invalid input at index {idx}: expected a list, got {type(lst).__name__}. "
+                    f"All items in 'lists' must be lists (e.g., [[1, 2], [3, 4]])."
+                )
+                return
+            concatenated.extend(lst)
+        yield "concatenated_list", concatenated

autogpt_platform/backend/backend/blocks/helpers/review.py

@@ -9,7 +9,7 @@ from typing import Any, Optional
 from prisma.enums import ReviewStatus
 from pydantic import BaseModel
 
-from backend.data.execution import ExecutionContext, ExecutionStatus
+from backend.data.execution import ExecutionStatus
 from backend.data.human_review import ReviewResult
 from backend.executor.manager import async_update_node_execution_status
 from backend.util.clients import get_database_manager_async_client
@@ -28,6 +28,11 @@ class ReviewDecision(BaseModel):
 class HITLReviewHelper:
     """Helper class for Human-In-The-Loop review operations."""
 
+    @staticmethod
+    async def check_approval(**kwargs) -> Optional[ReviewResult]:
+        """Check if there's an existing approval for this node execution."""
+        return await get_database_manager_async_client().check_approval(**kwargs)
+
     @staticmethod
     async def get_or_create_human_review(**kwargs) -> Optional[ReviewResult]:
         """Create or retrieve a human review from the database."""
@@ -55,11 +60,11 @@ class HITLReviewHelper:
     async def _handle_review_request(
         input_data: Any,
         user_id: str,
+        node_id: str,
         node_exec_id: str,
         graph_exec_id: str,
         graph_id: str,
         graph_version: int,
-        execution_context: ExecutionContext,
         block_name: str = "Block",
         editable: bool = False,
     ) -> Optional[ReviewResult]:
@@ -69,11 +74,11 @@ class HITLReviewHelper:
         Args:
             input_data: The input data to be reviewed
             user_id: ID of the user requesting the review
+            node_id: ID of the node in the graph definition
             node_exec_id: ID of the node execution
             graph_exec_id: ID of the graph execution
             graph_id: ID of the graph
             graph_version: Version of the graph
-            execution_context: Current execution context
             block_name: Name of the block requesting review
             editable: Whether the reviewer can edit the data
 
@@ -83,15 +88,41 @@ class HITLReviewHelper:
         Raises:
             Exception: If review creation or status update fails
         """
-        # Skip review if safe mode is disabled - return auto-approved result
-        if not execution_context.safe_mode:
+        # Note: Safe mode checks (human_in_the_loop_safe_mode, sensitive_action_safe_mode)
+        # are handled by the caller:
+        # - HITL blocks check human_in_the_loop_safe_mode in their run() method
+        # - Sensitive action blocks check sensitive_action_safe_mode in is_block_exec_need_review()
+        # This function only handles checking for existing approvals.
+
+        # Check if this node has already been approved (normal or auto-approval)
+        if approval_result := await HITLReviewHelper.check_approval(
+            node_exec_id=node_exec_id,
+            graph_exec_id=graph_exec_id,
+            node_id=node_id,
+            user_id=user_id,
+            input_data=input_data,
+        ):
             logger.info(
-                f"Block {block_name} skipping review for node {node_exec_id} - safe mode disabled"
+                f"Block {block_name} skipping review for node {node_exec_id} - "
+                f"found existing approval"
+            )
+            # Return a new ReviewResult with the current node_exec_id but approved status
+            # For auto-approvals, always use current input_data
+            # For normal approvals, use approval_result.data unless it's None
+            is_auto_approval = approval_result.node_exec_id != node_exec_id
+            approved_data = (
+                input_data
+                if is_auto_approval
+                else (
+                    approval_result.data
+                    if approval_result.data is not None
+                    else input_data
+                )
             )
             return ReviewResult(
-                data=input_data,
+                data=approved_data,
                 status=ReviewStatus.APPROVED,
-                message="Auto-approved (safe mode disabled)",
+                message=approval_result.message,
                 processed=True,
                 node_exec_id=node_exec_id,
             )
@@ -129,11 +160,11 @@ class HITLReviewHelper:
     async def handle_review_decision(
         input_data: Any,
         user_id: str,
+        node_id: str,
         node_exec_id: str,
         graph_exec_id: str,
         graph_id: str,
         graph_version: int,
-        execution_context: ExecutionContext,
         block_name: str = "Block",
         editable: bool = False,
     ) -> Optional[ReviewDecision]:
@@ -143,11 +174,11 @@ class HITLReviewHelper:
         Args:
             input_data: The input data to be reviewed
             user_id: ID of the user requesting the review
+            node_id: ID of the node in the graph definition
             node_exec_id: ID of the node execution
             graph_exec_id: ID of the graph execution
             graph_id: ID of the graph
             graph_version: Version of the graph
-            execution_context: Current execution context
             block_name: Name of the block requesting review
             editable: Whether the reviewer can edit the data
 
@@ -158,11 +189,11 @@ class HITLReviewHelper:
         review_result = await HITLReviewHelper._handle_review_request(
             input_data=input_data,
             user_id=user_id,
+            node_id=node_id,
             node_exec_id=node_exec_id,
             graph_exec_id=graph_exec_id,
             graph_id=graph_id,
             graph_version=graph_version,
-            execution_context=execution_context,
             block_name=block_name,
             editable=editable,
         )

autogpt_platform/backend/backend/blocks/human_in_the_loop.py

@@ -97,6 +97,7 @@ class HumanInTheLoopBlock(Block):
         input_data: Input,
         *,
         user_id: str,
+        node_id: str,
         node_exec_id: str,
         graph_exec_id: str,
         graph_id: str,
@@ -104,7 +105,7 @@ class HumanInTheLoopBlock(Block):
         execution_context: ExecutionContext,
         **_kwargs,
     ) -> BlockOutput:
-        if not execution_context.safe_mode:
+        if not execution_context.human_in_the_loop_safe_mode:
             logger.info(
                 f"HITL block skipping review for node {node_exec_id} - safe mode disabled"
             )
@@ -115,11 +116,11 @@ class HumanInTheLoopBlock(Block):
         decision = await self.handle_review_decision(
             input_data=input_data.data,
             user_id=user_id,
+            node_id=node_id,
             node_exec_id=node_exec_id,
             graph_exec_id=graph_exec_id,
             graph_id=graph_id,
             graph_version=graph_version,
-            execution_context=execution_context,
             block_name=self.name,
             editable=input_data.editable,
         )

autogpt_platform/backend/backend/blocks/llm.py

@@ -79,6 +79,10 @@ class ModelMetadata(NamedTuple):
     provider: str
     context_window: int
     max_output_tokens: int | None
+    display_name: str
+    provider_name: str
+    creator_name: str
+    price_tier: Literal[1, 2, 3]
 
 
 class LlmModelMeta(EnumMeta):
@@ -171,6 +175,26 @@ class LlmModel(str, Enum, metaclass=LlmModelMeta):
     V0_1_5_LG = "v0-1.5-lg"
     V0_1_0_MD = "v0-1.0-md"
 
+    @classmethod
+    def __get_pydantic_json_schema__(cls, schema, handler):
+        json_schema = handler(schema)
+        llm_model_metadata = {}
+        for model in cls:
+            model_name = model.value
+            metadata = model.metadata
+            llm_model_metadata[model_name] = {
+                "creator": metadata.creator_name,
+                "creator_name": metadata.creator_name,
+                "title": metadata.display_name,
+                "provider": metadata.provider,
+                "provider_name": metadata.provider_name,
+                "name": model_name,
+                "price_tier": metadata.price_tier,
+            }
+        json_schema["llm_model"] = True
+        json_schema["llm_model_metadata"] = llm_model_metadata
+        return json_schema
+
     @property
     def metadata(self) -> ModelMetadata:
         return MODEL_METADATA[self]
@@ -190,119 +214,291 @@ class LlmModel(str, Enum, metaclass=LlmModelMeta):
 
 MODEL_METADATA = {
     # https://platform.openai.com/docs/models
-    LlmModel.O3: ModelMetadata("openai", 200000, 100000),
-    LlmModel.O3_MINI: ModelMetadata("openai", 200000, 100000),  # o3-mini-2025-01-31
-    LlmModel.O1: ModelMetadata("openai", 200000, 100000),  # o1-2024-12-17
-    LlmModel.O1_MINI: ModelMetadata("openai", 128000, 65536),  # o1-mini-2024-09-12
+    LlmModel.O3: ModelMetadata("openai", 200000, 100000, "O3", "OpenAI", "OpenAI", 2),
+    LlmModel.O3_MINI: ModelMetadata(
+        "openai", 200000, 100000, "O3 Mini", "OpenAI", "OpenAI", 1
+    ),  # o3-mini-2025-01-31
+    LlmModel.O1: ModelMetadata(
+        "openai", 200000, 100000, "O1", "OpenAI", "OpenAI", 3
+    ),  # o1-2024-12-17
+    LlmModel.O1_MINI: ModelMetadata(
+        "openai", 128000, 65536, "O1 Mini", "OpenAI", "OpenAI", 2
+    ),  # o1-mini-2024-09-12
     # GPT-5 models
-    LlmModel.GPT5_2: ModelMetadata("openai", 400000, 128000),
-    LlmModel.GPT5_1: ModelMetadata("openai", 400000, 128000),
-    LlmModel.GPT5: ModelMetadata("openai", 400000, 128000),
-    LlmModel.GPT5_MINI: ModelMetadata("openai", 400000, 128000),
-    LlmModel.GPT5_NANO: ModelMetadata("openai", 400000, 128000),
-    LlmModel.GPT5_CHAT: ModelMetadata("openai", 400000, 16384),
-    LlmModel.GPT41: ModelMetadata("openai", 1047576, 32768),
-    LlmModel.GPT41_MINI: ModelMetadata("openai", 1047576, 32768),
+    LlmModel.GPT5_2: ModelMetadata(
+        "openai", 400000, 128000, "GPT-5.2", "OpenAI", "OpenAI", 3
+    ),
+    LlmModel.GPT5_1: ModelMetadata(
+        "openai", 400000, 128000, "GPT-5.1", "OpenAI", "OpenAI", 2
+    ),
+    LlmModel.GPT5: ModelMetadata(
+        "openai", 400000, 128000, "GPT-5", "OpenAI", "OpenAI", 1
+    ),
+    LlmModel.GPT5_MINI: ModelMetadata(
+        "openai", 400000, 128000, "GPT-5 Mini", "OpenAI", "OpenAI", 1
+    ),
+    LlmModel.GPT5_NANO: ModelMetadata(
+        "openai", 400000, 128000, "GPT-5 Nano", "OpenAI", "OpenAI", 1
+    ),
+    LlmModel.GPT5_CHAT: ModelMetadata(
+        "openai", 400000, 16384, "GPT-5 Chat Latest", "OpenAI", "OpenAI", 2
+    ),
+    LlmModel.GPT41: ModelMetadata(
+        "openai", 1047576, 32768, "GPT-4.1", "OpenAI", "OpenAI", 1
+    ),
+    LlmModel.GPT41_MINI: ModelMetadata(
+        "openai", 1047576, 32768, "GPT-4.1 Mini", "OpenAI", "OpenAI", 1
+    ),
     LlmModel.GPT4O_MINI: ModelMetadata(
-        "openai", 128000, 16384
+        "openai", 128000, 16384, "GPT-4o Mini", "OpenAI", "OpenAI", 1
     ),  # gpt-4o-mini-2024-07-18
-    LlmModel.GPT4O: ModelMetadata("openai", 128000, 16384),  # gpt-4o-2024-08-06
+    LlmModel.GPT4O: ModelMetadata(
+        "openai", 128000, 16384, "GPT-4o", "OpenAI", "OpenAI", 2
+    ),  # gpt-4o-2024-08-06
     LlmModel.GPT4_TURBO: ModelMetadata(
-        "openai", 128000, 4096
+        "openai", 128000, 4096, "GPT-4 Turbo", "OpenAI", "OpenAI", 3
     ),  # gpt-4-turbo-2024-04-09
-    LlmModel.GPT3_5_TURBO: ModelMetadata("openai", 16385, 4096),  # gpt-3.5-turbo-0125
+    LlmModel.GPT3_5_TURBO: ModelMetadata(
+        "openai", 16385, 4096, "GPT-3.5 Turbo", "OpenAI", "OpenAI", 1
+    ),  # gpt-3.5-turbo-0125
     # https://docs.anthropic.com/en/docs/about-claude/models
     LlmModel.CLAUDE_4_1_OPUS: ModelMetadata(
-        "anthropic", 200000, 32000
+        "anthropic", 200000, 32000, "Claude Opus 4.1", "Anthropic", "Anthropic", 3
     ),  # claude-opus-4-1-20250805
     LlmModel.CLAUDE_4_OPUS: ModelMetadata(
-        "anthropic", 200000, 32000
+        "anthropic", 200000, 32000, "Claude Opus 4", "Anthropic", "Anthropic", 3
     ),  # claude-4-opus-20250514
     LlmModel.CLAUDE_4_SONNET: ModelMetadata(
-        "anthropic", 200000, 64000
+        "anthropic", 200000, 64000, "Claude Sonnet 4", "Anthropic", "Anthropic", 2
     ),  # claude-4-sonnet-20250514
     LlmModel.CLAUDE_4_5_OPUS: ModelMetadata(
-        "anthropic", 200000, 64000
+        "anthropic", 200000, 64000, "Claude Opus 4.5", "Anthropic", "Anthropic", 3
     ),  # claude-opus-4-5-20251101
     LlmModel.CLAUDE_4_5_SONNET: ModelMetadata(
-        "anthropic", 200000, 64000
+        "anthropic", 200000, 64000, "Claude Sonnet 4.5", "Anthropic", "Anthropic", 3
     ),  # claude-sonnet-4-5-20250929
     LlmModel.CLAUDE_4_5_HAIKU: ModelMetadata(
-        "anthropic", 200000, 64000
+        "anthropic", 200000, 64000, "Claude Haiku 4.5", "Anthropic", "Anthropic", 2
     ),  # claude-haiku-4-5-20251001
     LlmModel.CLAUDE_3_7_SONNET: ModelMetadata(
-        "anthropic", 200000, 64000
+        "anthropic", 200000, 64000, "Claude 3.7 Sonnet", "Anthropic", "Anthropic", 2
     ),  # claude-3-7-sonnet-20250219
     LlmModel.CLAUDE_3_HAIKU: ModelMetadata(
-        "anthropic", 200000, 4096
+        "anthropic", 200000, 4096, "Claude 3 Haiku", "Anthropic", "Anthropic", 1
     ),  # claude-3-haiku-20240307
     # https://docs.aimlapi.com/api-overview/model-database/text-models
-    LlmModel.AIML_API_QWEN2_5_72B: ModelMetadata("aiml_api", 32000, 8000),
-    LlmModel.AIML_API_LLAMA3_1_70B: ModelMetadata("aiml_api", 128000, 40000),
-    LlmModel.AIML_API_LLAMA3_3_70B: ModelMetadata("aiml_api", 128000, None),
-    LlmModel.AIML_API_META_LLAMA_3_1_70B: ModelMetadata("aiml_api", 131000, 2000),
-    LlmModel.AIML_API_LLAMA_3_2_3B: ModelMetadata("aiml_api", 128000, None),
-    # https://console.groq.com/docs/models
-    LlmModel.LLAMA3_3_70B: ModelMetadata("groq", 128000, 32768),
-    LlmModel.LLAMA3_1_8B: ModelMetadata("groq", 128000, 8192),
-    # https://ollama.com/library
-    LlmModel.OLLAMA_LLAMA3_3: ModelMetadata("ollama", 8192, None),
-    LlmModel.OLLAMA_LLAMA3_2: ModelMetadata("ollama", 8192, None),
-    LlmModel.OLLAMA_LLAMA3_8B: ModelMetadata("ollama", 8192, None),
-    LlmModel.OLLAMA_LLAMA3_405B: ModelMetadata("ollama", 8192, None),
-    LlmModel.OLLAMA_DOLPHIN: ModelMetadata("ollama", 32768, None),
-    # https://openrouter.ai/models
-    LlmModel.GEMINI_2_5_PRO: ModelMetadata("open_router", 1050000, 8192),
-    LlmModel.GEMINI_3_PRO_PREVIEW: ModelMetadata("open_router", 1048576, 65535),
-    LlmModel.GEMINI_2_5_FLASH: ModelMetadata("open_router", 1048576, 65535),
-    LlmModel.GEMINI_2_0_FLASH: ModelMetadata("open_router", 1048576, 8192),
-    LlmModel.GEMINI_2_5_FLASH_LITE_PREVIEW: ModelMetadata(
-        "open_router", 1048576, 65535
+    LlmModel.AIML_API_QWEN2_5_72B: ModelMetadata(
+        "aiml_api", 32000, 8000, "Qwen 2.5 72B Instruct Turbo", "AI/ML", "Qwen", 1
+    ),
+    LlmModel.AIML_API_LLAMA3_1_70B: ModelMetadata(
+        "aiml_api",
+        128000,
+        40000,
+        "Llama 3.1 Nemotron 70B Instruct",
+        "AI/ML",
+        "Nvidia",
+        1,
+    ),
+    LlmModel.AIML_API_LLAMA3_3_70B: ModelMetadata(
+        "aiml_api", 128000, None, "Llama 3.3 70B Instruct Turbo", "AI/ML", "Meta", 1
+    ),
+    LlmModel.AIML_API_META_LLAMA_3_1_70B: ModelMetadata(
+        "aiml_api", 131000, 2000, "Llama 3.1 70B Instruct Turbo", "AI/ML", "Meta", 1
+    ),
+    LlmModel.AIML_API_LLAMA_3_2_3B: ModelMetadata(
+        "aiml_api", 128000, None, "Llama 3.2 3B Instruct Turbo", "AI/ML", "Meta", 1
+    ),
+    # https://console.groq.com/docs/models
+    LlmModel.LLAMA3_3_70B: ModelMetadata(
+        "groq", 128000, 32768, "Llama 3.3 70B Versatile", "Groq", "Meta", 1
+    ),
+    LlmModel.LLAMA3_1_8B: ModelMetadata(
+        "groq", 128000, 8192, "Llama 3.1 8B Instant", "Groq", "Meta", 1
+    ),
+    # https://ollama.com/library
+    LlmModel.OLLAMA_LLAMA3_3: ModelMetadata(
+        "ollama", 8192, None, "Llama 3.3", "Ollama", "Meta", 1
+    ),
+    LlmModel.OLLAMA_LLAMA3_2: ModelMetadata(
+        "ollama", 8192, None, "Llama 3.2", "Ollama", "Meta", 1
+    ),
+    LlmModel.OLLAMA_LLAMA3_8B: ModelMetadata(
+        "ollama", 8192, None, "Llama 3", "Ollama", "Meta", 1
+    ),
+    LlmModel.OLLAMA_LLAMA3_405B: ModelMetadata(
+        "ollama", 8192, None, "Llama 3.1 405B", "Ollama", "Meta", 1
+    ),
+    LlmModel.OLLAMA_DOLPHIN: ModelMetadata(
+        "ollama", 32768, None, "Dolphin Mistral Latest", "Ollama", "Mistral AI", 1
+    ),
+    # https://openrouter.ai/models
+    LlmModel.GEMINI_2_5_PRO: ModelMetadata(
+        "open_router",
+        1050000,
+        8192,
+        "Gemini 2.5 Pro Preview 03.25",
+        "OpenRouter",
+        "Google",
+        2,
+    ),
+    LlmModel.GEMINI_3_PRO_PREVIEW: ModelMetadata(
+        "open_router", 1048576, 65535, "Gemini 3 Pro Preview", "OpenRouter", "Google", 2
+    ),
+    LlmModel.GEMINI_2_5_FLASH: ModelMetadata(
+        "open_router", 1048576, 65535, "Gemini 2.5 Flash", "OpenRouter", "Google", 1
+    ),
+    LlmModel.GEMINI_2_0_FLASH: ModelMetadata(
+        "open_router", 1048576, 8192, "Gemini 2.0 Flash 001", "OpenRouter", "Google", 1
+    ),
+    LlmModel.GEMINI_2_5_FLASH_LITE_PREVIEW: ModelMetadata(
+        "open_router",
+        1048576,
+        65535,
+        "Gemini 2.5 Flash Lite Preview 06.17",
+        "OpenRouter",
+        "Google",
+        1,
+    ),
+    LlmModel.GEMINI_2_0_FLASH_LITE: ModelMetadata(
+        "open_router",
+        1048576,
+        8192,
+        "Gemini 2.0 Flash Lite 001",
+        "OpenRouter",
+        "Google",
+        1,
+    ),
+    LlmModel.MISTRAL_NEMO: ModelMetadata(
+        "open_router", 128000, 4096, "Mistral Nemo", "OpenRouter", "Mistral AI", 1
+    ),
+    LlmModel.COHERE_COMMAND_R_08_2024: ModelMetadata(
+        "open_router", 128000, 4096, "Command R 08.2024", "OpenRouter", "Cohere", 1
+    ),
+    LlmModel.COHERE_COMMAND_R_PLUS_08_2024: ModelMetadata(
+        "open_router", 128000, 4096, "Command R Plus 08.2024", "OpenRouter", "Cohere", 2
+    ),
+    LlmModel.DEEPSEEK_CHAT: ModelMetadata(
+        "open_router", 64000, 2048, "DeepSeek Chat", "OpenRouter", "DeepSeek", 1
+    ),
+    LlmModel.DEEPSEEK_R1_0528: ModelMetadata(
+        "open_router", 163840, 163840, "DeepSeek R1 0528", "OpenRouter", "DeepSeek", 1
+    ),
+    LlmModel.PERPLEXITY_SONAR: ModelMetadata(
+        "open_router", 127000, 8000, "Sonar", "OpenRouter", "Perplexity", 1
+    ),
+    LlmModel.PERPLEXITY_SONAR_PRO: ModelMetadata(
+        "open_router", 200000, 8000, "Sonar Pro", "OpenRouter", "Perplexity", 2
     ),
-    LlmModel.GEMINI_2_0_FLASH_LITE: ModelMetadata("open_router", 1048576, 8192),
-    LlmModel.MISTRAL_NEMO: ModelMetadata("open_router", 128000, 4096),
-    LlmModel.COHERE_COMMAND_R_08_2024: ModelMetadata("open_router", 128000, 4096),
-    LlmModel.COHERE_COMMAND_R_PLUS_08_2024: ModelMetadata("open_router", 128000, 4096),
-    LlmModel.DEEPSEEK_CHAT: ModelMetadata("open_router", 64000, 2048),
-    LlmModel.DEEPSEEK_R1_0528: ModelMetadata("open_router", 163840, 163840),
-    LlmModel.PERPLEXITY_SONAR: ModelMetadata("open_router", 127000, 8000),
-    LlmModel.PERPLEXITY_SONAR_PRO: ModelMetadata("open_router", 200000, 8000),
     LlmModel.PERPLEXITY_SONAR_DEEP_RESEARCH: ModelMetadata(
         "open_router",
         128000,
         16000,
+        "Sonar Deep Research",
+        "OpenRouter",
+        "Perplexity",
+        3,
     ),
     LlmModel.NOUSRESEARCH_HERMES_3_LLAMA_3_1_405B: ModelMetadata(
-        "open_router", 131000, 4096
+        "open_router",
+        131000,
+        4096,
+        "Hermes 3 Llama 3.1 405B",
+        "OpenRouter",
+        "Nous Research",
+        1,
     ),
     LlmModel.NOUSRESEARCH_HERMES_3_LLAMA_3_1_70B: ModelMetadata(
-        "open_router", 12288, 12288
+        "open_router",
+        12288,
+        12288,
+        "Hermes 3 Llama 3.1 70B",
+        "OpenRouter",
+        "Nous Research",
+        1,
+    ),
+    LlmModel.OPENAI_GPT_OSS_120B: ModelMetadata(
+        "open_router", 131072, 131072, "GPT-OSS 120B", "OpenRouter", "OpenAI", 1
+    ),
+    LlmModel.OPENAI_GPT_OSS_20B: ModelMetadata(
+        "open_router", 131072, 32768, "GPT-OSS 20B", "OpenRouter", "OpenAI", 1
+    ),
+    LlmModel.AMAZON_NOVA_LITE_V1: ModelMetadata(
+        "open_router", 300000, 5120, "Nova Lite V1", "OpenRouter", "Amazon", 1
+    ),
+    LlmModel.AMAZON_NOVA_MICRO_V1: ModelMetadata(
+        "open_router", 128000, 5120, "Nova Micro V1", "OpenRouter", "Amazon", 1
+    ),
+    LlmModel.AMAZON_NOVA_PRO_V1: ModelMetadata(
+        "open_router", 300000, 5120, "Nova Pro V1", "OpenRouter", "Amazon", 1
+    ),
+    LlmModel.MICROSOFT_WIZARDLM_2_8X22B: ModelMetadata(
+        "open_router", 65536, 4096, "WizardLM 2 8x22B", "OpenRouter", "Microsoft", 1
+    ),
+    LlmModel.GRYPHE_MYTHOMAX_L2_13B: ModelMetadata(
+        "open_router", 4096, 4096, "MythoMax L2 13B", "OpenRouter", "Gryphe", 1
+    ),
+    LlmModel.META_LLAMA_4_SCOUT: ModelMetadata(
+        "open_router", 131072, 131072, "Llama 4 Scout", "OpenRouter", "Meta", 1
+    ),
+    LlmModel.META_LLAMA_4_MAVERICK: ModelMetadata(
+        "open_router", 1048576, 1000000, "Llama 4 Maverick", "OpenRouter", "Meta", 1
+    ),
+    LlmModel.GROK_4: ModelMetadata(
+        "open_router", 256000, 256000, "Grok 4", "OpenRouter", "xAI", 3
+    ),
+    LlmModel.GROK_4_FAST: ModelMetadata(
+        "open_router", 2000000, 30000, "Grok 4 Fast", "OpenRouter", "xAI", 1
+    ),
+    LlmModel.GROK_4_1_FAST: ModelMetadata(
+        "open_router", 2000000, 30000, "Grok 4.1 Fast", "OpenRouter", "xAI", 1
+    ),
+    LlmModel.GROK_CODE_FAST_1: ModelMetadata(
+        "open_router", 256000, 10000, "Grok Code Fast 1", "OpenRouter", "xAI", 1
+    ),
+    LlmModel.KIMI_K2: ModelMetadata(
+        "open_router", 131000, 131000, "Kimi K2", "OpenRouter", "Moonshot AI", 1
+    ),
+    LlmModel.QWEN3_235B_A22B_THINKING: ModelMetadata(
+        "open_router",
+        262144,
+        262144,
+        "Qwen 3 235B A22B Thinking 2507",
+        "OpenRouter",
+        "Qwen",
+        1,
+    ),
+    LlmModel.QWEN3_CODER: ModelMetadata(
+        "open_router", 262144, 262144, "Qwen 3 Coder", "OpenRouter", "Qwen", 3
     ),
-    LlmModel.OPENAI_GPT_OSS_120B: ModelMetadata("open_router", 131072, 131072),
-    LlmModel.OPENAI_GPT_OSS_20B: ModelMetadata("open_router", 131072, 32768),
-    LlmModel.AMAZON_NOVA_LITE_V1: ModelMetadata("open_router", 300000, 5120),
-    LlmModel.AMAZON_NOVA_MICRO_V1: ModelMetadata("open_router", 128000, 5120),
-    LlmModel.AMAZON_NOVA_PRO_V1: ModelMetadata("open_router", 300000, 5120),
-    LlmModel.MICROSOFT_WIZARDLM_2_8X22B: ModelMetadata("open_router", 65536, 4096),
-    LlmModel.GRYPHE_MYTHOMAX_L2_13B: ModelMetadata("open_router", 4096, 4096),
-    LlmModel.META_LLAMA_4_SCOUT: ModelMetadata("open_router", 131072, 131072),
-    LlmModel.META_LLAMA_4_MAVERICK: ModelMetadata("open_router", 1048576, 1000000),
-    LlmModel.GROK_4: ModelMetadata("open_router", 256000, 256000),
-    LlmModel.GROK_4_FAST: ModelMetadata("open_router", 2000000, 30000),
-    LlmModel.GROK_4_1_FAST: ModelMetadata("open_router", 2000000, 30000),
-    LlmModel.GROK_CODE_FAST_1: ModelMetadata("open_router", 256000, 10000),
-    LlmModel.KIMI_K2: ModelMetadata("open_router", 131000, 131000),
-    LlmModel.QWEN3_235B_A22B_THINKING: ModelMetadata("open_router", 262144, 262144),
-    LlmModel.QWEN3_CODER: ModelMetadata("open_router", 262144, 262144),
     # Llama API models
-    LlmModel.LLAMA_API_LLAMA_4_SCOUT: ModelMetadata("llama_api", 128000, 4028),
-    LlmModel.LLAMA_API_LLAMA4_MAVERICK: ModelMetadata("llama_api", 128000, 4028),
-    LlmModel.LLAMA_API_LLAMA3_3_8B: ModelMetadata("llama_api", 128000, 4028),
-    LlmModel.LLAMA_API_LLAMA3_3_70B: ModelMetadata("llama_api", 128000, 4028),
+    LlmModel.LLAMA_API_LLAMA_4_SCOUT: ModelMetadata(
+        "llama_api",
+        128000,
+        4028,
+        "Llama 4 Scout 17B 16E Instruct FP8",
+        "Llama API",
+        "Meta",
+        1,
+    ),
+    LlmModel.LLAMA_API_LLAMA4_MAVERICK: ModelMetadata(
+        "llama_api",
+        128000,
+        4028,
+        "Llama 4 Maverick 17B 128E Instruct FP8",
+        "Llama API",
+        "Meta",
+        1,
+    ),
+    LlmModel.LLAMA_API_LLAMA3_3_8B: ModelMetadata(
+        "llama_api", 128000, 4028, "Llama 3.3 8B Instruct", "Llama API", "Meta", 1
+    ),
+    LlmModel.LLAMA_API_LLAMA3_3_70B: ModelMetadata(
+        "llama_api", 128000, 4028, "Llama 3.3 70B Instruct", "Llama API", "Meta", 1
+    ),
     # v0 by Vercel models
-    LlmModel.V0_1_5_MD: ModelMetadata("v0", 128000, 64000),
-    LlmModel.V0_1_5_LG: ModelMetadata("v0", 512000, 64000),
-    LlmModel.V0_1_0_MD: ModelMetadata("v0", 128000, 64000),
+    LlmModel.V0_1_5_MD: ModelMetadata("v0", 128000, 64000, "v0 1.5 MD", "V0", "V0", 1),
+    LlmModel.V0_1_5_LG: ModelMetadata("v0", 512000, 64000, "v0 1.5 LG", "V0", "V0", 1),
+    LlmModel.V0_1_0_MD: ModelMetadata("v0", 128000, 64000, "v0 1.0 MD", "V0", "V0", 1),
 }
 
 DEFAULT_LLM_MODEL = LlmModel.GPT5_2

autogpt_platform/backend/backend/blocks/test/test_smart_decision_maker.py

@@ -242,7 +242,7 @@ async def test_smart_decision_maker_tracks_llm_stats():
         outputs = {}
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -343,7 +343,7 @@ async def test_smart_decision_maker_parameter_validation():
 
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -409,7 +409,7 @@ async def test_smart_decision_maker_parameter_validation():
 
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -471,7 +471,7 @@ async def test_smart_decision_maker_parameter_validation():
         outputs = {}
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -535,7 +535,7 @@ async def test_smart_decision_maker_parameter_validation():
         outputs = {}
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -658,7 +658,7 @@ async def test_smart_decision_maker_raw_response_conversion():
         outputs = {}
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -730,7 +730,7 @@ async def test_smart_decision_maker_raw_response_conversion():
         outputs = {}
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -786,7 +786,7 @@ async def test_smart_decision_maker_raw_response_conversion():
         outputs = {}
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 
@@ -905,7 +905,7 @@ async def test_smart_decision_maker_agent_mode():
         # Create a mock execution context
 
         mock_execution_context = ExecutionContext(
-            safe_mode=False,
+            human_in_the_loop_safe_mode=False,
         )
 
         # Create a mock execution processor for agent mode tests
@@ -1027,7 +1027,7 @@ async def test_smart_decision_maker_traditional_mode_default():
 
         # Create execution context
 
-        mock_execution_context = ExecutionContext(safe_mode=False)
+        mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
 
         # Create a mock execution processor for tests
 

autogpt_platform/backend/backend/blocks/test/test_smart_decision_maker_dynamic_fields.py

@@ -386,7 +386,7 @@ async def test_output_yielding_with_dynamic_fields():
             outputs = {}
             from backend.data.execution import ExecutionContext
 
-            mock_execution_context = ExecutionContext(safe_mode=False)
+            mock_execution_context = ExecutionContext(human_in_the_loop_safe_mode=False)
             mock_execution_processor = MagicMock()
 
             async for output_name, output_value in block.run(
@@ -609,7 +609,9 @@ async def test_validation_errors_dont_pollute_conversation():
                 outputs = {}
                 from backend.data.execution import ExecutionContext
 
-                mock_execution_context = ExecutionContext(safe_mode=False)
+                mock_execution_context = ExecutionContext(
+                    human_in_the_loop_safe_mode=False
+                )
 
                 # Create a proper mock execution processor for agent mode
                 from collections import defaultdict

autogpt_platform/backend/backend/data/block.py

@@ -441,6 +441,7 @@ class Block(ABC, Generic[BlockSchemaInputType, BlockSchemaOutputType]):
         static_output: bool = False,
         block_type: BlockType = BlockType.STANDARD,
         webhook_config: Optional[BlockWebhookConfig | BlockManualWebhookConfig] = None,
+        is_sensitive_action: bool = False,
     ):
         """
         Initialize the block with the given schema.
@@ -473,8 +474,8 @@ class Block(ABC, Generic[BlockSchemaInputType, BlockSchemaOutputType]):
         self.static_output = static_output
         self.block_type = block_type
         self.webhook_config = webhook_config
+        self.is_sensitive_action = is_sensitive_action
         self.execution_stats: NodeExecutionStats = NodeExecutionStats()
-        self.requires_human_review: bool = False
 
         if self.webhook_config:
             if isinstance(self.webhook_config, BlockWebhookConfig):
@@ -622,6 +623,7 @@ class Block(ABC, Generic[BlockSchemaInputType, BlockSchemaOutputType]):
         input_data: BlockInput,
         *,
         user_id: str,
+        node_id: str,
         node_exec_id: str,
         graph_exec_id: str,
         graph_id: str,
@@ -637,8 +639,9 @@ class Block(ABC, Generic[BlockSchemaInputType, BlockSchemaOutputType]):
             - should_pause: True if execution should be paused for review
             - input_data_to_use: The input data to use (may be modified by reviewer)
         """
-        # Skip review if not required or safe mode is disabled
-        if not self.requires_human_review or not execution_context.safe_mode:
+        if not (
+            self.is_sensitive_action and execution_context.sensitive_action_safe_mode
+        ):
             return False, input_data
 
         from backend.blocks.helpers.review import HITLReviewHelper
@@ -647,11 +650,11 @@ class Block(ABC, Generic[BlockSchemaInputType, BlockSchemaOutputType]):
         decision = await HITLReviewHelper.handle_review_decision(
             input_data=input_data,
             user_id=user_id,
+            node_id=node_id,
             node_exec_id=node_exec_id,
             graph_exec_id=graph_exec_id,
             graph_id=graph_id,
             graph_version=graph_version,
-            execution_context=execution_context,
             block_name=self.name,
             editable=True,
         )

autogpt_platform/backend/backend/data/block_cost_config.py

@@ -99,10 +99,15 @@ MODEL_COST: dict[LlmModel, int] = {
     LlmModel.OPENAI_GPT_OSS_20B: 1,
     LlmModel.GEMINI_2_5_PRO: 4,
     LlmModel.GEMINI_3_PRO_PREVIEW: 5,
+    LlmModel.GEMINI_2_5_FLASH: 1,
+    LlmModel.GEMINI_2_0_FLASH: 1,
+    LlmModel.GEMINI_2_5_FLASH_LITE_PREVIEW: 1,
+    LlmModel.GEMINI_2_0_FLASH_LITE: 1,
     LlmModel.MISTRAL_NEMO: 1,
     LlmModel.COHERE_COMMAND_R_08_2024: 1,
     LlmModel.COHERE_COMMAND_R_PLUS_08_2024: 3,
     LlmModel.DEEPSEEK_CHAT: 2,
+    LlmModel.DEEPSEEK_R1_0528: 1,
     LlmModel.PERPLEXITY_SONAR: 1,
     LlmModel.PERPLEXITY_SONAR_PRO: 5,
     LlmModel.PERPLEXITY_SONAR_DEEP_RESEARCH: 10,
@@ -126,11 +131,6 @@ MODEL_COST: dict[LlmModel, int] = {
     LlmModel.KIMI_K2: 1,
     LlmModel.QWEN3_235B_A22B_THINKING: 1,
     LlmModel.QWEN3_CODER: 9,
-    LlmModel.GEMINI_2_5_FLASH: 1,
-    LlmModel.GEMINI_2_0_FLASH: 1,
-    LlmModel.GEMINI_2_5_FLASH_LITE_PREVIEW: 1,
-    LlmModel.GEMINI_2_0_FLASH_LITE: 1,
-    LlmModel.DEEPSEEK_R1_0528: 1,
     # v0 by Vercel models
     LlmModel.V0_1_5_MD: 1,
     LlmModel.V0_1_5_LG: 2,

autogpt_platform/backend/backend/data/db.py

@@ -38,20 +38,6 @@ POOL_TIMEOUT = os.getenv("DB_POOL_TIMEOUT")
 if POOL_TIMEOUT:
     DATABASE_URL = add_param(DATABASE_URL, "pool_timeout", POOL_TIMEOUT)
 
-# Add public schema to search_path for pgvector type access
-# The vector extension is in public schema, but search_path is determined by schema parameter
-# Extract the schema from DATABASE_URL or default to 'public' (matching get_database_schema())
-parsed_url = urlparse(DATABASE_URL)
-url_params = dict(parse_qsl(parsed_url.query))
-db_schema = url_params.get("schema", "public")
-# Build search_path, avoiding duplicates if db_schema is already 'public'
-search_path_schemas = list(
-    dict.fromkeys([db_schema, "public"])
-)  # Preserves order, removes duplicates
-search_path = ",".join(search_path_schemas)
-# This allows using ::vector without schema qualification
-DATABASE_URL = add_param(DATABASE_URL, "options", f"-c search_path={search_path}")
-
 HTTP_TIMEOUT = int(POOL_TIMEOUT) if POOL_TIMEOUT else None
 
 prisma = Prisma(
@@ -127,38 +113,48 @@ async def _raw_with_schema(
     *args,
     execute: bool = False,
     client: Prisma | None = None,
-    set_public_search_path: bool = False,
 ) -> list[dict] | int:
     """Internal: Execute raw SQL with proper schema handling.
 
     Use query_raw_with_schema() or execute_raw_with_schema() instead.
 
+    Supports placeholders:
+        - {schema_prefix}: Table/type prefix (e.g., "platform".)
+        - {schema}: Raw schema name for application tables (e.g., platform)
+
+    Note on pgvector types:
+        Use unqualified ::vector and <=> operator in queries. PostgreSQL resolves
+        these via search_path, which includes the schema where pgvector is installed
+        on all environments (local, CI, dev).
+
     Args:
-        query_template: SQL query with {schema_prefix} placeholder
+        query_template: SQL query with {schema_prefix} and/or {schema} placeholders
         *args: Query parameters
         execute: If False, executes SELECT query. If True, executes INSERT/UPDATE/DELETE.
         client: Optional Prisma client for transactions (only used when execute=True).
-        set_public_search_path: If True, sets search_path to include public schema.
-                                Needed for pgvector types and other public schema objects.
 
     Returns:
         - list[dict] if execute=False (query results)
         - int if execute=True (number of affected rows)
+
+    Example with vector type:
+        await execute_raw_with_schema(
+            'INSERT INTO {schema_prefix}"Embedding" (vec) VALUES ($1::vector)',
+            embedding_data
+        )
     """
     schema = get_database_schema()
     schema_prefix = f'"{schema}".' if schema != "public" else ""
-    formatted_query = query_template.format(schema_prefix=schema_prefix)
+
+    formatted_query = query_template.format(
+        schema_prefix=schema_prefix,
+        schema=schema,
+    )
 
     import prisma as prisma_module
 
     db_client = client if client else prisma_module.get_client()
 
-    # Set search_path to include public schema if requested
-    # Prisma doesn't support the 'options' connection parameter, so we set it per-session
-    # This is idempotent and safe to call multiple times
-    if set_public_search_path:
-        await db_client.execute_raw(f"SET search_path = {schema}, public")  # type: ignore
-
     if execute:
         result = await db_client.execute_raw(formatted_query, *args)  # type: ignore
     else:
@@ -167,16 +163,12 @@ async def _raw_with_schema(
     return result
 
 
-async def query_raw_with_schema(
-    query_template: str, *args, set_public_search_path: bool = False
-) -> list[dict]:
+async def query_raw_with_schema(query_template: str, *args) -> list[dict]:
     """Execute raw SQL SELECT query with proper schema handling.
 
     Args:
-        query_template: SQL query with {schema_prefix} placeholder
+        query_template: SQL query with {schema_prefix} and/or {schema} placeholders
         *args: Query parameters
-        set_public_search_path: If True, sets search_path to include public schema.
-                                Needed for pgvector types and other public schema objects.
 
     Returns:
         List of result rows as dictionaries
@@ -187,23 +179,20 @@ async def query_raw_with_schema(
             user_id
         )
     """
-    return await _raw_with_schema(query_template, *args, execute=False, set_public_search_path=set_public_search_path)  # type: ignore
+    return await _raw_with_schema(query_template, *args, execute=False)  # type: ignore
 
 
 async def execute_raw_with_schema(
     query_template: str,
     *args,
     client: Prisma | None = None,
-    set_public_search_path: bool = False,
 ) -> int:
     """Execute raw SQL command (INSERT/UPDATE/DELETE) with proper schema handling.
 
     Args:
-        query_template: SQL query with {schema_prefix} placeholder
+        query_template: SQL query with {schema_prefix} and/or {schema} placeholders
         *args: Query parameters
         client: Optional Prisma client for transactions
-        set_public_search_path: If True, sets search_path to include public schema.
-                                Needed for pgvector types and other public schema objects.
 
     Returns:
         Number of affected rows
@@ -215,7 +204,7 @@ async def execute_raw_with_schema(
             client=tx  # Optional transaction client
         )
     """
-    return await _raw_with_schema(query_template, *args, execute=True, client=client, set_public_search_path=set_public_search_path)  # type: ignore
+    return await _raw_with_schema(query_template, *args, execute=True, client=client)  # type: ignore
 
 
 class BaseDbModel(BaseModel):

autogpt_platform/backend/backend/data/event_bus.py

@@ -103,8 +103,18 @@ class RedisEventBus(BaseRedisEventBus[M], ABC):
         return redis.get_redis()
 
     def publish_event(self, event: M, channel_key: str):
-        message, full_channel_name = self._serialize_message(event, channel_key)
-        self.connection.publish(full_channel_name, message)
+        """
+        Publish an event to Redis. Gracefully handles connection failures
+        by logging the error instead of raising exceptions.
+        """
+        try:
+            message, full_channel_name = self._serialize_message(event, channel_key)
+            self.connection.publish(full_channel_name, message)
+        except Exception:
+            logger.exception(
+                f"Failed to publish event to Redis channel {channel_key}. "
+                "Event bus operation will continue without Redis connectivity."
+            )
 
     def listen_events(self, channel_key: str) -> Generator[M, None, None]:
         pubsub, full_channel_name = self._get_pubsub_channel(
@@ -128,9 +138,19 @@ class AsyncRedisEventBus(BaseRedisEventBus[M], ABC):
         return await redis.get_redis_async()
 
     async def publish_event(self, event: M, channel_key: str):
-        message, full_channel_name = self._serialize_message(event, channel_key)
-        connection = await self.connection
-        await connection.publish(full_channel_name, message)
+        """
+        Publish an event to Redis. Gracefully handles connection failures
+        by logging the error instead of raising exceptions.
+        """
+        try:
+            message, full_channel_name = self._serialize_message(event, channel_key)
+            connection = await self.connection
+            await connection.publish(full_channel_name, message)
+        except Exception:
+            logger.exception(
+                f"Failed to publish event to Redis channel {channel_key}. "
+                "Event bus operation will continue without Redis connectivity."
+            )
 
     async def listen_events(self, channel_key: str) -> AsyncGenerator[M, None]:
         pubsub, full_channel_name = self._get_pubsub_channel(

autogpt_platform/backend/backend/data/event_bus_test.py

@@ -0,0 +1,56 @@
+"""
+Tests for event_bus graceful degradation when Redis is unavailable.
+"""
+
+from unittest.mock import AsyncMock, patch
+
+import pytest
+from pydantic import BaseModel
+
+from backend.data.event_bus import AsyncRedisEventBus
+
+
+class TestEvent(BaseModel):
+    """Test event model."""
+
+    message: str
+
+
+class TestNotificationBus(AsyncRedisEventBus[TestEvent]):
+    """Test implementation of AsyncRedisEventBus."""
+
+    Model = TestEvent
+
+    @property
+    def event_bus_name(self) -> str:
+        return "test_event_bus"
+
+
+@pytest.mark.asyncio
+async def test_publish_event_handles_connection_failure_gracefully():
+    """Test that publish_event logs exception instead of raising when Redis is unavailable."""
+    bus = TestNotificationBus()
+    event = TestEvent(message="test message")
+
+    # Mock get_redis_async to raise connection error
+    with patch(
+        "backend.data.event_bus.redis.get_redis_async",
+        side_effect=ConnectionError("Authentication required."),
+    ):
+        # Should not raise exception
+        await bus.publish_event(event, "test_channel")
+
+
+@pytest.mark.asyncio
+async def test_publish_event_works_with_redis_available():
+    """Test that publish_event works normally when Redis is available."""
+    bus = TestNotificationBus()
+    event = TestEvent(message="test message")
+
+    # Mock successful Redis connection
+    mock_redis = AsyncMock()
+    mock_redis.publish = AsyncMock()
+
+    with patch("backend.data.event_bus.redis.get_redis_async", return_value=mock_redis):
+        await bus.publish_event(event, "test_channel")
+        mock_redis.publish.assert_called_once()

autogpt_platform/backend/backend/data/execution.py

@@ -81,7 +81,10 @@ class ExecutionContext(BaseModel):
     This includes information needed by blocks, sub-graphs, and execution management.
     """
 
-    safe_mode: bool = True
+    model_config = {"extra": "ignore"}
+
+    human_in_the_loop_safe_mode: bool = True
+    sensitive_action_safe_mode: bool = False
     user_timezone: str = "UTC"
     root_execution_id: Optional[str] = None
     parent_execution_id: Optional[str] = None

autogpt_platform/backend/backend/data/graph.py

@@ -3,7 +3,7 @@ import logging
 import uuid
 from collections import defaultdict
 from datetime import datetime, timezone
-from typing import TYPE_CHECKING, Any, Literal, Optional, cast
+from typing import TYPE_CHECKING, Annotated, Any, Literal, Optional, cast
 
 from prisma.enums import SubmissionStatus
 from prisma.models import (
@@ -20,7 +20,7 @@ from prisma.types import (
     AgentNodeLinkCreateInput,
     StoreListingVersionWhereInput,
 )
-from pydantic import BaseModel, Field, create_model
+from pydantic import BaseModel, BeforeValidator, Field, create_model
 from pydantic.fields import computed_field
 
 from backend.blocks.agent import AgentExecutorBlock
@@ -62,7 +62,31 @@ logger = logging.getLogger(__name__)
 
 
 class GraphSettings(BaseModel):
-    human_in_the_loop_safe_mode: bool | None = None
+    # Use Annotated with BeforeValidator to coerce None to default values.
+    # This handles cases where the database has null values for these fields.
+    model_config = {"extra": "ignore"}
+
+    human_in_the_loop_safe_mode: Annotated[
+        bool, BeforeValidator(lambda v: v if v is not None else True)
+    ] = True
+    sensitive_action_safe_mode: Annotated[
+        bool, BeforeValidator(lambda v: v if v is not None else False)
+    ] = False
+
+    @classmethod
+    def from_graph(
+        cls,
+        graph: "GraphModel",
+        hitl_safe_mode: bool | None = None,
+        sensitive_action_safe_mode: bool = False,
+    ) -> "GraphSettings":
+        # Default to True if not explicitly set
+        if hitl_safe_mode is None:
+            hitl_safe_mode = True
+        return cls(
+            human_in_the_loop_safe_mode=hitl_safe_mode,
+            sensitive_action_safe_mode=sensitive_action_safe_mode,
+        )
 
 
 class Link(BaseDbModel):
@@ -244,10 +268,14 @@ class BaseGraph(BaseDbModel):
         return any(
             node.block_id
             for node in self.nodes
-            if (
-                node.block.block_type == BlockType.HUMAN_IN_THE_LOOP
-                or node.block.requires_human_review
-            )
+            if node.block.block_type == BlockType.HUMAN_IN_THE_LOOP
+        )
+
+    @computed_field
+    @property
+    def has_sensitive_action(self) -> bool:
+        return any(
+            node.block_id for node in self.nodes if node.block.is_sensitive_action
         )
 
     @property

autogpt_platform/backend/backend/data/human_review.py

@@ -17,6 +17,7 @@ from backend.api.features.executions.review.model import (
     PendingHumanReviewModel,
     SafeJsonData,
 )
+from backend.data.execution import get_graph_execution_meta
 from backend.util.json import SafeJson
 
 logger = logging.getLogger(__name__)
@@ -32,6 +33,125 @@ class ReviewResult(BaseModel):
     node_exec_id: str
 
 
+def get_auto_approve_key(graph_exec_id: str, node_id: str) -> str:
+    """Generate the special nodeExecId key for auto-approval records."""
+    return f"auto_approve_{graph_exec_id}_{node_id}"
+
+
+async def check_approval(
+    node_exec_id: str,
+    graph_exec_id: str,
+    node_id: str,
+    user_id: str,
+    input_data: SafeJsonData | None = None,
+) -> Optional[ReviewResult]:
+    """
+    Check if there's an existing approval for this node execution.
+
+    Checks both:
+    1. Normal approval by node_exec_id (previous run of the same node execution)
+    2. Auto-approval by special key pattern "auto_approve_{graph_exec_id}_{node_id}"
+
+    Args:
+        node_exec_id: ID of the node execution
+        graph_exec_id: ID of the graph execution
+        node_id: ID of the node definition (not execution)
+        user_id: ID of the user (for data isolation)
+        input_data: Current input data (used for auto-approvals to avoid stale data)
+
+    Returns:
+        ReviewResult if approval found (either normal or auto), None otherwise
+    """
+    auto_approve_key = get_auto_approve_key(graph_exec_id, node_id)
+
+    # Check for either normal approval or auto-approval in a single query
+    existing_review = await PendingHumanReview.prisma().find_first(
+        where={
+            "OR": [
+                {"nodeExecId": node_exec_id},
+                {"nodeExecId": auto_approve_key},
+            ],
+            "status": ReviewStatus.APPROVED,
+            "userId": user_id,
+        },
+    )
+
+    if existing_review:
+        is_auto_approval = existing_review.nodeExecId == auto_approve_key
+        logger.info(
+            f"Found {'auto-' if is_auto_approval else ''}approval for node {node_id} "
+            f"(exec: {node_exec_id}) in execution {graph_exec_id}"
+        )
+        # For auto-approvals, use current input_data to avoid replaying stale payload
+        # For normal approvals, use the stored payload (which may have been edited)
+        return ReviewResult(
+            data=(
+                input_data
+                if is_auto_approval and input_data is not None
+                else existing_review.payload
+            ),
+            status=ReviewStatus.APPROVED,
+            message=(
+                "Auto-approved (user approved all future actions for this node)"
+                if is_auto_approval
+                else existing_review.reviewMessage or ""
+            ),
+            processed=True,
+            node_exec_id=existing_review.nodeExecId,
+        )
+
+    return None
+
+
+async def create_auto_approval_record(
+    user_id: str,
+    graph_exec_id: str,
+    graph_id: str,
+    graph_version: int,
+    node_id: str,
+    payload: SafeJsonData,
+) -> None:
+    """
+    Create an auto-approval record for a node in this execution.
+
+    This is stored as a PendingHumanReview with a special nodeExecId pattern
+    and status=APPROVED, so future executions of the same node can skip review.
+
+    Raises:
+        ValueError: If the graph execution doesn't belong to the user
+    """
+    # Validate that the graph execution belongs to this user (defense in depth)
+    graph_exec = await get_graph_execution_meta(
+        user_id=user_id, execution_id=graph_exec_id
+    )
+    if not graph_exec:
+        raise ValueError(
+            f"Graph execution {graph_exec_id} not found or doesn't belong to user {user_id}"
+        )
+
+    auto_approve_key = get_auto_approve_key(graph_exec_id, node_id)
+
+    await PendingHumanReview.prisma().upsert(
+        where={"nodeExecId": auto_approve_key},
+        data={
+            "create": {
+                "nodeExecId": auto_approve_key,
+                "userId": user_id,
+                "graphExecId": graph_exec_id,
+                "graphId": graph_id,
+                "graphVersion": graph_version,
+                "payload": SafeJson(payload),
+                "instructions": "Auto-approval record",
+                "editable": False,
+                "status": ReviewStatus.APPROVED,
+                "processed": True,
+                "reviewedAt": datetime.now(timezone.utc),
+            },
+            "update": {},  # Already exists, no update needed
+        },
+    )
+
+
 async def get_or_create_human_review(
     user_id: str,
     node_exec_id: str,
@@ -108,6 +228,33 @@ async def get_or_create_human_review(
         )
 
 
+async def get_pending_review_by_node_exec_id(
+    node_exec_id: str, user_id: str
+) -> Optional["PendingHumanReviewModel"]:
+    """
+    Get a pending review by its node execution ID.
+
+    Args:
+        node_exec_id: The node execution ID to look up
+        user_id: User ID for authorization (only returns if review belongs to this user)
+
+    Returns:
+        The pending review if found and belongs to user, None otherwise
+    """
+    review = await PendingHumanReview.prisma().find_first(
+        where={
+            "nodeExecId": node_exec_id,
+            "userId": user_id,
+            "status": ReviewStatus.WAITING,
+        }
+    )
+
+    if not review:
+        return None
+
+    return PendingHumanReviewModel.from_db(review)
+
+
 async def has_pending_reviews_for_graph_exec(graph_exec_id: str) -> bool:
     """
     Check if a graph execution has any pending reviews.
@@ -256,3 +403,44 @@ async def update_review_processed_status(node_exec_id: str, processed: bool) ->
     await PendingHumanReview.prisma().update(
         where={"nodeExecId": node_exec_id}, data={"processed": processed}
     )
+
+
+async def cancel_pending_reviews_for_execution(graph_exec_id: str, user_id: str) -> int:
+    """
+    Cancel all pending reviews for a graph execution (e.g., when execution is stopped).
+
+    Marks all WAITING reviews as REJECTED with a message indicating the execution was stopped.
+
+    Args:
+        graph_exec_id: The graph execution ID
+        user_id: User ID who owns the execution (for security validation)
+
+    Returns:
+        Number of reviews cancelled
+
+    Raises:
+        ValueError: If the graph execution doesn't belong to the user
+    """
+    # Validate user ownership before cancelling reviews
+    graph_exec = await get_graph_execution_meta(
+        user_id=user_id, execution_id=graph_exec_id
+    )
+    if not graph_exec:
+        raise ValueError(
+            f"Graph execution {graph_exec_id} not found or doesn't belong to user {user_id}"
+        )
+
+    result = await PendingHumanReview.prisma().update_many(
+        where={
+            "graphExecId": graph_exec_id,
+            "userId": user_id,
+            "status": ReviewStatus.WAITING,
+        },
+        data={
+            "status": ReviewStatus.REJECTED,
+            "reviewMessage": "Execution was stopped by user",
+            "processed": True,
+            "reviewedAt": datetime.now(timezone.utc),
+        },
+    )
+    return result

autogpt_platform/backend/backend/data/human_review_test.py

@@ -46,8 +46,8 @@ async def test_get_or_create_human_review_new(
     sample_db_review.status = ReviewStatus.WAITING
     sample_db_review.processed = False
 
-    mock_upsert = mocker.patch("backend.data.human_review.PendingHumanReview.prisma")
-    mock_upsert.return_value.upsert = AsyncMock(return_value=sample_db_review)
+    mock_prisma = mocker.patch("backend.data.human_review.PendingHumanReview.prisma")
+    mock_prisma.return_value.upsert = AsyncMock(return_value=sample_db_review)
 
     result = await get_or_create_human_review(
         user_id="test-user-123",
@@ -75,8 +75,8 @@ async def test_get_or_create_human_review_approved(
     sample_db_review.processed = False
     sample_db_review.reviewMessage = "Looks good"
 
-    mock_upsert = mocker.patch("backend.data.human_review.PendingHumanReview.prisma")
-    mock_upsert.return_value.upsert = AsyncMock(return_value=sample_db_review)
+    mock_prisma = mocker.patch("backend.data.human_review.PendingHumanReview.prisma")
+    mock_prisma.return_value.upsert = AsyncMock(return_value=sample_db_review)
 
     result = await get_or_create_human_review(
         user_id="test-user-123",

autogpt_platform/backend/backend/data/understanding.py

@@ -328,6 +328,8 @@ async def clear_business_understanding(user_id: str) -> bool:
 
 def format_understanding_for_prompt(understanding: BusinessUnderstanding) -> str:
     """Format business understanding as text for system prompt injection."""
+    if not understanding:
+        return ""
     sections = []
 
     # User info section

autogpt_platform/backend/backend/executor/database.py

@@ -50,6 +50,8 @@ from backend.data.graph import (
     validate_graph_execution_permissions,
 )
 from backend.data.human_review import (
+    cancel_pending_reviews_for_execution,
+    check_approval,
     get_or_create_human_review,
     has_pending_reviews_for_graph_exec,
     update_review_processed_status,
@@ -190,6 +192,8 @@ class DatabaseManager(AppService):
     get_user_notification_preference = _(get_user_notification_preference)
 
     # Human In The Loop
+    cancel_pending_reviews_for_execution = _(cancel_pending_reviews_for_execution)
+    check_approval = _(check_approval)
     get_or_create_human_review = _(get_or_create_human_review)
     has_pending_reviews_for_graph_exec = _(has_pending_reviews_for_graph_exec)
     update_review_processed_status = _(update_review_processed_status)
@@ -313,6 +317,8 @@ class DatabaseManagerAsyncClient(AppServiceClient):
     set_execution_kv_data = d.set_execution_kv_data
 
     # Human In The Loop
+    cancel_pending_reviews_for_execution = d.cancel_pending_reviews_for_execution
+    check_approval = d.check_approval
     get_or_create_human_review = d.get_or_create_human_review
     update_review_processed_status = d.update_review_processed_status
 

autogpt_platform/backend/backend/executor/scheduler.py

@@ -309,7 +309,7 @@ def ensure_embeddings_coverage():
 
         # Process in batches until no more missing embeddings
         while True:
-            result = db_client.backfill_missing_embeddings(batch_size=10)
+            result = db_client.backfill_missing_embeddings(batch_size=100)
 
             total_processed += result["processed"]
             total_success += result["success"]

autogpt_platform/backend/backend/executor/utils.py

@@ -10,6 +10,7 @@ from pydantic import BaseModel, JsonValue, ValidationError
 
 from backend.data import execution as execution_db
 from backend.data import graph as graph_db
+from backend.data import human_review as human_review_db
 from backend.data import onboarding as onboarding_db
 from backend.data import user as user_db
 from backend.data.block import (
@@ -749,9 +750,27 @@ async def stop_graph_execution(
         if graph_exec.status in [
             ExecutionStatus.QUEUED,
             ExecutionStatus.INCOMPLETE,
+            ExecutionStatus.REVIEW,
         ]:
-            # If the graph is still on the queue, we can prevent them from being executed
-            # by setting the status to TERMINATED.
+            # If the graph is queued/incomplete/paused for review, terminate immediately
+            # No need to wait for executor since it's not actively running
+
+            # If graph is in REVIEW status, clean up pending reviews before terminating
+            if graph_exec.status == ExecutionStatus.REVIEW:
+                # Use human_review_db if Prisma connected, else database manager
+                review_db = (
+                    human_review_db
+                    if prisma.is_connected()
+                    else get_database_manager_async_client()
+                )
+                # Mark all pending reviews as rejected/cancelled
+                cancelled_count = await review_db.cancel_pending_reviews_for_execution(
+                    graph_exec_id, user_id
+                )
+                logger.info(
+                    f"Cancelled {cancelled_count} pending review(s) for stopped execution {graph_exec_id}"
+                )
+
             graph_exec.status = ExecutionStatus.TERMINATED
 
             await asyncio.gather(
@@ -873,11 +892,8 @@ async def add_graph_execution(
         settings = await gdb.get_graph_settings(user_id=user_id, graph_id=graph_id)
 
         execution_context = ExecutionContext(
-            safe_mode=(
-                settings.human_in_the_loop_safe_mode
-                if settings.human_in_the_loop_safe_mode is not None
-                else True
-            ),
+            human_in_the_loop_safe_mode=settings.human_in_the_loop_safe_mode,
+            sensitive_action_safe_mode=settings.sensitive_action_safe_mode,
             user_timezone=(
                 user.timezone if user.timezone != USER_TIMEZONE_NOT_SET else "UTC"
             ),
@@ -890,9 +906,28 @@ async def add_graph_execution(
             nodes_to_skip=nodes_to_skip,
             execution_context=execution_context,
         )
-        logger.info(f"Publishing execution {graph_exec.id} to execution queue")
+        logger.info(f"Queueing execution {graph_exec.id}")
+
+        # Update execution status to QUEUED BEFORE publishing to prevent race condition
+        # where two concurrent requests could both publish the same execution
+        updated_exec = await edb.update_graph_execution_stats(
+            graph_exec_id=graph_exec.id,
+            status=ExecutionStatus.QUEUED,
+        )
+
+        # Verify the status update succeeded (prevents duplicate queueing in race conditions)
+        # If another request already updated the status, this execution will not be QUEUED
+        if not updated_exec or updated_exec.status != ExecutionStatus.QUEUED:
+            logger.warning(
+                f"Skipping queue publish for execution {graph_exec.id} - "
+                f"status update failed or execution already queued by another request"
+            )
+            return graph_exec
+
+        graph_exec.status = ExecutionStatus.QUEUED
 
         # Publish to execution queue for executor to pick up
+        # This happens AFTER status update to ensure only one request publishes
         exec_queue = await get_async_execution_queue()
         await exec_queue.publish_message(
             routing_key=GRAPH_EXECUTION_ROUTING_KEY,
@@ -900,13 +935,6 @@ async def add_graph_execution(
             exchange=GRAPH_EXECUTION_EXCHANGE,
         )
         logger.info(f"Published execution {graph_exec.id} to RabbitMQ queue")
-
-        # Update execution status to QUEUED
-        graph_exec.status = ExecutionStatus.QUEUED
-        await edb.update_graph_execution_stats(
-            graph_exec_id=graph_exec.id,
-            status=graph_exec.status,
-        )
     except BaseException as e:
         err = str(e) or type(e).__name__
         if not graph_exec:

autogpt_platform/backend/backend/executor/utils_test.py

@@ -4,6 +4,7 @@ import pytest
 from pytest_mock import MockerFixture
 
 from backend.data.dynamic_fields import merge_execution_input, parse_execution_output
+from backend.data.execution import ExecutionStatus
 from backend.util.mock import MockObject
 
 
@@ -346,6 +347,7 @@ async def test_add_graph_execution_is_repeatable(mocker: MockerFixture):
     mock_graph_exec = mocker.MagicMock(spec=GraphExecutionWithNodes)
     mock_graph_exec.id = "execution-id-123"
     mock_graph_exec.node_executions = []  # Add this to avoid AttributeError
+    mock_graph_exec.status = ExecutionStatus.QUEUED  # Required for race condition check
     mock_graph_exec.to_graph_execution_entry.return_value = mocker.MagicMock()
 
     # Mock the queue and event bus
@@ -386,6 +388,7 @@ async def test_add_graph_execution_is_repeatable(mocker: MockerFixture):
     mock_user.timezone = "UTC"
     mock_settings = mocker.MagicMock()
     mock_settings.human_in_the_loop_safe_mode = True
+    mock_settings.sensitive_action_safe_mode = False
 
     mock_udb.get_user_by_id = mocker.AsyncMock(return_value=mock_user)
     mock_gdb.get_graph_settings = mocker.AsyncMock(return_value=mock_settings)
@@ -610,6 +613,7 @@ async def test_add_graph_execution_with_nodes_to_skip(mocker: MockerFixture):
     mock_graph_exec = mocker.MagicMock(spec=GraphExecutionWithNodes)
     mock_graph_exec.id = "execution-id-123"
     mock_graph_exec.node_executions = []
+    mock_graph_exec.status = ExecutionStatus.QUEUED  # Required for race condition check
 
     # Track what's passed to to_graph_execution_entry
     captured_kwargs = {}
@@ -651,6 +655,7 @@ async def test_add_graph_execution_with_nodes_to_skip(mocker: MockerFixture):
     mock_user.timezone = "UTC"
     mock_settings = mocker.MagicMock()
     mock_settings.human_in_the_loop_safe_mode = True
+    mock_settings.sensitive_action_safe_mode = False
 
     mock_udb.get_user_by_id = mocker.AsyncMock(return_value=mock_user)
     mock_gdb.get_graph_settings = mocker.AsyncMock(return_value=mock_settings)
@@ -668,3 +673,232 @@ async def test_add_graph_execution_with_nodes_to_skip(mocker: MockerFixture):
     # Verify nodes_to_skip was passed to to_graph_execution_entry
     assert "nodes_to_skip" in captured_kwargs
     assert captured_kwargs["nodes_to_skip"] == nodes_to_skip
+
+
+@pytest.mark.asyncio
+async def test_stop_graph_execution_in_review_status_cancels_pending_reviews(
+    mocker: MockerFixture,
+):
+    """Test that stopping an execution in REVIEW status cancels pending reviews."""
+    from backend.data.execution import ExecutionStatus, GraphExecutionMeta
+    from backend.executor.utils import stop_graph_execution
+
+    user_id = "test-user"
+    graph_exec_id = "test-exec-123"
+
+    # Mock graph execution in REVIEW status
+    mock_graph_exec = mocker.MagicMock(spec=GraphExecutionMeta)
+    mock_graph_exec.id = graph_exec_id
+    mock_graph_exec.status = ExecutionStatus.REVIEW
+
+    # Mock dependencies
+    mock_get_queue = mocker.patch("backend.executor.utils.get_async_execution_queue")
+    mock_queue_client = mocker.AsyncMock()
+    mock_get_queue.return_value = mock_queue_client
+
+    mock_prisma = mocker.patch("backend.executor.utils.prisma")
+    mock_prisma.is_connected.return_value = True
+
+    mock_human_review_db = mocker.patch("backend.executor.utils.human_review_db")
+    mock_human_review_db.cancel_pending_reviews_for_execution = mocker.AsyncMock(
+        return_value=2  # 2 reviews cancelled
+    )
+
+    mock_execution_db = mocker.patch("backend.executor.utils.execution_db")
+    mock_execution_db.get_graph_execution_meta = mocker.AsyncMock(
+        return_value=mock_graph_exec
+    )
+    mock_execution_db.update_graph_execution_stats = mocker.AsyncMock()
+
+    mock_get_event_bus = mocker.patch(
+        "backend.executor.utils.get_async_execution_event_bus"
+    )
+    mock_event_bus = mocker.MagicMock()
+    mock_event_bus.publish = mocker.AsyncMock()
+    mock_get_event_bus.return_value = mock_event_bus
+
+    mock_get_child_executions = mocker.patch(
+        "backend.executor.utils._get_child_executions"
+    )
+    mock_get_child_executions.return_value = []  # No children
+
+    # Call stop_graph_execution with timeout to allow status check
+    await stop_graph_execution(
+        user_id=user_id,
+        graph_exec_id=graph_exec_id,
+        wait_timeout=1.0,  # Wait to allow status check
+        cascade=True,
+    )
+
+    # Verify pending reviews were cancelled
+    mock_human_review_db.cancel_pending_reviews_for_execution.assert_called_once_with(
+        graph_exec_id, user_id
+    )
+
+    # Verify execution status was updated to TERMINATED
+    mock_execution_db.update_graph_execution_stats.assert_called_once()
+    call_kwargs = mock_execution_db.update_graph_execution_stats.call_args[1]
+    assert call_kwargs["graph_exec_id"] == graph_exec_id
+    assert call_kwargs["status"] == ExecutionStatus.TERMINATED
+
+
+@pytest.mark.asyncio
+async def test_stop_graph_execution_with_database_manager_when_prisma_disconnected(
+    mocker: MockerFixture,
+):
+    """Test that stop uses database manager when Prisma is not connected."""
+    from backend.data.execution import ExecutionStatus, GraphExecutionMeta
+    from backend.executor.utils import stop_graph_execution
+
+    user_id = "test-user"
+    graph_exec_id = "test-exec-456"
+
+    # Mock graph execution in REVIEW status
+    mock_graph_exec = mocker.MagicMock(spec=GraphExecutionMeta)
+    mock_graph_exec.id = graph_exec_id
+    mock_graph_exec.status = ExecutionStatus.REVIEW
+
+    # Mock dependencies
+    mock_get_queue = mocker.patch("backend.executor.utils.get_async_execution_queue")
+    mock_queue_client = mocker.AsyncMock()
+    mock_get_queue.return_value = mock_queue_client
+
+    # Prisma is NOT connected
+    mock_prisma = mocker.patch("backend.executor.utils.prisma")
+    mock_prisma.is_connected.return_value = False
+
+    # Mock database manager client
+    mock_get_db_manager = mocker.patch(
+        "backend.executor.utils.get_database_manager_async_client"
+    )
+    mock_db_manager = mocker.AsyncMock()
+    mock_db_manager.get_graph_execution_meta = mocker.AsyncMock(
+        return_value=mock_graph_exec
+    )
+    mock_db_manager.cancel_pending_reviews_for_execution = mocker.AsyncMock(
+        return_value=3  # 3 reviews cancelled
+    )
+    mock_db_manager.update_graph_execution_stats = mocker.AsyncMock()
+    mock_get_db_manager.return_value = mock_db_manager
+
+    mock_get_event_bus = mocker.patch(
+        "backend.executor.utils.get_async_execution_event_bus"
+    )
+    mock_event_bus = mocker.MagicMock()
+    mock_event_bus.publish = mocker.AsyncMock()
+    mock_get_event_bus.return_value = mock_event_bus
+
+    mock_get_child_executions = mocker.patch(
+        "backend.executor.utils._get_child_executions"
+    )
+    mock_get_child_executions.return_value = []  # No children
+
+    # Call stop_graph_execution with timeout
+    await stop_graph_execution(
+        user_id=user_id,
+        graph_exec_id=graph_exec_id,
+        wait_timeout=1.0,
+        cascade=True,
+    )
+
+    # Verify database manager was used for cancel_pending_reviews
+    mock_db_manager.cancel_pending_reviews_for_execution.assert_called_once_with(
+        graph_exec_id, user_id
+    )
+
+    # Verify execution status was updated via database manager
+    mock_db_manager.update_graph_execution_stats.assert_called_once()
+
+
+@pytest.mark.asyncio
+async def test_stop_graph_execution_cascades_to_child_with_reviews(
+    mocker: MockerFixture,
+):
+    """Test that stopping parent execution cascades to children and cancels their reviews."""
+    from backend.data.execution import ExecutionStatus, GraphExecutionMeta
+    from backend.executor.utils import stop_graph_execution
+
+    user_id = "test-user"
+    parent_exec_id = "parent-exec"
+    child_exec_id = "child-exec"
+
+    # Mock parent execution in RUNNING status
+    mock_parent_exec = mocker.MagicMock(spec=GraphExecutionMeta)
+    mock_parent_exec.id = parent_exec_id
+    mock_parent_exec.status = ExecutionStatus.RUNNING
+
+    # Mock child execution in REVIEW status
+    mock_child_exec = mocker.MagicMock(spec=GraphExecutionMeta)
+    mock_child_exec.id = child_exec_id
+    mock_child_exec.status = ExecutionStatus.REVIEW
+
+    # Mock dependencies
+    mock_get_queue = mocker.patch("backend.executor.utils.get_async_execution_queue")
+    mock_queue_client = mocker.AsyncMock()
+    mock_get_queue.return_value = mock_queue_client
+
+    mock_prisma = mocker.patch("backend.executor.utils.prisma")
+    mock_prisma.is_connected.return_value = True
+
+    mock_human_review_db = mocker.patch("backend.executor.utils.human_review_db")
+    mock_human_review_db.cancel_pending_reviews_for_execution = mocker.AsyncMock(
+        return_value=1  # 1 child review cancelled
+    )
+
+    # Mock execution_db to return different status based on which execution is queried
+    mock_execution_db = mocker.patch("backend.executor.utils.execution_db")
+
+    # Track call count to simulate status transition
+    call_count = {"count": 0}
+
+    async def get_exec_meta_side_effect(execution_id, user_id):
+        call_count["count"] += 1
+        if execution_id == parent_exec_id:
+            # After a few calls (child processing happens), transition parent to TERMINATED
+            # This simulates the executor service processing the stop request
+            if call_count["count"] > 3:
+                mock_parent_exec.status = ExecutionStatus.TERMINATED
+            return mock_parent_exec
+        elif execution_id == child_exec_id:
+            return mock_child_exec
+        return None
+
+    mock_execution_db.get_graph_execution_meta = mocker.AsyncMock(
+        side_effect=get_exec_meta_side_effect
+    )
+    mock_execution_db.update_graph_execution_stats = mocker.AsyncMock()
+
+    mock_get_event_bus = mocker.patch(
+        "backend.executor.utils.get_async_execution_event_bus"
+    )
+    mock_event_bus = mocker.MagicMock()
+    mock_event_bus.publish = mocker.AsyncMock()
+    mock_get_event_bus.return_value = mock_event_bus
+
+    # Mock _get_child_executions to return the child
+    mock_get_child_executions = mocker.patch(
+        "backend.executor.utils._get_child_executions"
+    )
+
+    def get_children_side_effect(parent_id):
+        if parent_id == parent_exec_id:
+            return [mock_child_exec]
+        return []
+
+    mock_get_child_executions.side_effect = get_children_side_effect
+
+    # Call stop_graph_execution on parent with cascade=True
+    await stop_graph_execution(
+        user_id=user_id,
+        graph_exec_id=parent_exec_id,
+        wait_timeout=1.0,
+        cascade=True,
+    )
+
+    # Verify child reviews were cancelled
+    mock_human_review_db.cancel_pending_reviews_for_execution.assert_called_once_with(
+        child_exec_id, user_id
+    )
+
+    # Verify both parent and child status updates
+    assert mock_execution_db.update_graph_execution_stats.call_count >= 1

autogpt_platform/backend/migrations/20260109181714_add_docs_embedding/migration.sql

@@ -1,9 +1,10 @@
 -- CreateExtension
 -- Supabase: pgvector must be enabled via Dashboard → Database → Extensions first
--- Create in public schema so vector type is available across all schemas
+-- Creates extension in current schema (determined by search_path from DATABASE_URL ?schema= param)
+-- This ensures vector type is in the same schema as tables, making ::vector work without explicit qualification
 DO $$
 BEGIN
-    CREATE EXTENSION IF NOT EXISTS "vector" WITH SCHEMA "public";
+    CREATE EXTENSION IF NOT EXISTS "vector";
 EXCEPTION WHEN OTHERS THEN
     RAISE NOTICE 'vector extension not available or already exists, skipping';
 END $$;
@@ -19,7 +20,7 @@ CREATE TABLE "UnifiedContentEmbedding" (
     "contentType" "ContentType" NOT NULL,
     "contentId" TEXT NOT NULL,
     "userId" TEXT,
-    "embedding" public.vector(1536) NOT NULL,
+    "embedding" vector(1536) NOT NULL,
     "searchableText" TEXT NOT NULL,
     "metadata" JSONB NOT NULL DEFAULT '{}',
 
@@ -45,4 +46,4 @@ CREATE UNIQUE INDEX "UnifiedContentEmbedding_contentType_contentId_userId_key" O
 -- Uses cosine distance operator (<=>), which matches the query in hybrid_search.py
 -- Note: Drop first in case Prisma created a btree index (Prisma doesn't support HNSW)
 DROP INDEX IF EXISTS "UnifiedContentEmbedding_embedding_idx";
-CREATE INDEX "UnifiedContentEmbedding_embedding_idx" ON "UnifiedContentEmbedding" USING hnsw ("embedding" public.vector_cosine_ops);
+CREATE INDEX "UnifiedContentEmbedding_embedding_idx" ON "UnifiedContentEmbedding" USING hnsw ("embedding" vector_cosine_ops);

autogpt_platform/backend/migrations/20260121200000_remove_node_execution_fk_from_pending_human_review/migration.sql

@@ -0,0 +1,7 @@
+-- Remove NodeExecution foreign key from PendingHumanReview
+-- The nodeExecId column remains as the primary key, but we remove the FK constraint
+-- to AgentNodeExecution since PendingHumanReview records can persist after node
+-- execution records are deleted.
+
+-- Drop foreign key constraint that linked PendingHumanReview.nodeExecId to AgentNodeExecution.id
+ALTER TABLE "PendingHumanReview" DROP CONSTRAINT IF EXISTS "PendingHumanReview_nodeExecId_fkey";

autogpt_platform/backend/schema.prisma

@@ -517,8 +517,6 @@ model AgentNodeExecution {
 
   stats Json?
 
-  PendingHumanReview PendingHumanReview?
-
   @@index([agentGraphExecutionId, agentNodeId, executionStatus])
   @@index([agentNodeId, executionStatus])
   @@index([addedTime, queuedTime])
@@ -567,6 +565,7 @@ enum ReviewStatus {
 }
 
 // Pending human reviews for Human-in-the-loop blocks
+// Also stores auto-approval records with special nodeExecId patterns (e.g., "auto_approve_{graph_exec_id}_{node_id}")
 model PendingHumanReview {
   nodeExecId    String       @id
   userId        String
@@ -585,7 +584,6 @@ model PendingHumanReview {
   reviewedAt    DateTime?
 
   User           User                @relation(fields: [userId], references: [id], onDelete: Cascade)
-  NodeExecution  AgentNodeExecution  @relation(fields: [nodeExecId], references: [id], onDelete: Cascade)
   GraphExecution AgentGraphExecution @relation(fields: [graphExecId], references: [id], onDelete: Cascade)
 
   @@unique([nodeExecId]) // One pending review per node execution

autogpt_platform/backend/scripts/generate_block_docs.py

@@ -366,12 +366,12 @@ def generate_block_markdown(
     lines.append("")
 
     # What it is (full description)
-    lines.append(f"### What it is")
+    lines.append("### What it is")
     lines.append(block.description or "No description available.")
     lines.append("")
 
     # How it works (manual section)
-    lines.append(f"### How it works")
+    lines.append("### How it works")
     how_it_works = manual_content.get(
         "how_it_works", "_Add technical explanation here._"
     )
@@ -383,7 +383,7 @@ def generate_block_markdown(
     # Inputs table (auto-generated)
     visible_inputs = [f for f in block.inputs if not f.hidden]
     if visible_inputs:
-        lines.append(f"### Inputs")
+        lines.append("### Inputs")
         lines.append("")
         lines.append("| Input | Description | Type | Required |")
         lines.append("|-------|-------------|------|----------|")
@@ -400,7 +400,7 @@ def generate_block_markdown(
     # Outputs table (auto-generated)
     visible_outputs = [f for f in block.outputs if not f.hidden]
     if visible_outputs:
-        lines.append(f"### Outputs")
+        lines.append("### Outputs")
         lines.append("")
         lines.append("| Output | Description | Type |")
         lines.append("|--------|-------------|------|")
@@ -414,7 +414,7 @@ def generate_block_markdown(
         lines.append("")
 
     # Possible use case (manual section)
-    lines.append(f"### Possible use case")
+    lines.append("### Possible use case")
     use_case = manual_content.get("use_case", "_Add practical use case examples here._")
     lines.append("<!-- MANUAL: use_case -->")
     lines.append(use_case)

autogpt_platform/backend/snapshots/grph_single

@@ -11,6 +11,7 @@
   "forked_from_version": null,
   "has_external_trigger": false,
   "has_human_in_the_loop": false,
+  "has_sensitive_action": false,
   "id": "graph-123",
   "input_schema": {
     "properties": {},

autogpt_platform/backend/snapshots/grphs_all

@@ -11,6 +11,7 @@
     "forked_from_version": null,
     "has_external_trigger": false,
     "has_human_in_the_loop": false,
+    "has_sensitive_action": false,
     "id": "graph-123",
     "input_schema": {
       "properties": {},

autogpt_platform/backend/snapshots/lib_agts_search

@@ -27,6 +27,8 @@
         "properties": {}
       },
       "has_external_trigger": false,
+      "has_human_in_the_loop": false,
+      "has_sensitive_action": false,
       "trigger_setup_info": null,
       "new_output": false,
       "can_access_graph": true,
@@ -34,7 +36,8 @@
       "is_favorite": false,
       "recommended_schedule_cron": null,
       "settings": {
-        "human_in_the_loop_safe_mode": null
+        "human_in_the_loop_safe_mode": true,
+        "sensitive_action_safe_mode": false
       },
       "marketplace_listing": null
     },
@@ -65,6 +68,8 @@
         "properties": {}
       },
       "has_external_trigger": false,
+      "has_human_in_the_loop": false,
+      "has_sensitive_action": false,
       "trigger_setup_info": null,
       "new_output": false,
       "can_access_graph": false,
@@ -72,7 +77,8 @@
       "is_favorite": false,
       "recommended_schedule_cron": null,
       "settings": {
-        "human_in_the_loop_safe_mode": null
+        "human_in_the_loop_safe_mode": true,
+        "sensitive_action_safe_mode": false
       },
       "marketplace_listing": null
     }

autogpt_platform/frontend/src/app/(platform)/build/components/BuilderActions/components/RunGraph/RunGraph.tsx

@@ -5,10 +5,11 @@ import {
   TooltipContent,
   TooltipTrigger,
 } from "@/components/atoms/Tooltip/BaseTooltip";
-import { PlayIcon, StopIcon } from "@phosphor-icons/react";
+import { CircleNotchIcon, PlayIcon, StopIcon } from "@phosphor-icons/react";
 import { useShallow } from "zustand/react/shallow";
 import { RunInputDialog } from "../RunInputDialog/RunInputDialog";
 import { useRunGraph } from "./useRunGraph";
+import { cn } from "@/lib/utils";
 
 export const RunGraph = ({ flowID }: { flowID: string | null }) => {
   const {
@@ -24,6 +25,31 @@ export const RunGraph = ({ flowID }: { flowID: string | null }) => {
     useShallow((state) => state.isGraphRunning),
   );
 
+  const isLoading = isExecutingGraph || isTerminatingGraph || isSaving;
+
+  // Determine which icon to show with proper animation
+  const renderIcon = () => {
+    const iconClass = cn(
+      "size-4 transition-transform duration-200 ease-out",
+      !isLoading && "group-hover:scale-110",
+    );
+
+    if (isLoading) {
+      return (
+        <CircleNotchIcon
+          className={cn(iconClass, "animate-spin")}
+          weight="bold"
+        />
+      );
+    }
+
+    if (isGraphRunning) {
+      return <StopIcon className={iconClass} weight="fill" />;
+    }
+
+    return <PlayIcon className={iconClass} weight="fill" />;
+  };
+
   return (
     <>
       <Tooltip>
@@ -33,18 +59,18 @@ export const RunGraph = ({ flowID }: { flowID: string | null }) => {
             variant={isGraphRunning ? "destructive" : "primary"}
             data-id={isGraphRunning ? "stop-graph-button" : "run-graph-button"}
             onClick={isGraphRunning ? handleStopGraph : handleRunGraph}
-            disabled={!flowID || isExecutingGraph || isTerminatingGraph}
-            loading={isExecutingGraph || isTerminatingGraph || isSaving}
+            disabled={!flowID || isLoading}
+            className="group"
           >
-            {!isGraphRunning ? (
-              <PlayIcon className="size-4" />
-            ) : (
-              <StopIcon className="size-4" />
-            )}
+            {renderIcon()}
           </Button>
         </TooltipTrigger>
         <TooltipContent>
-          {isGraphRunning ? "Stop agent" : "Run agent"}
+          {isLoading
+            ? "Processing..."
+            : isGraphRunning
+              ? "Stop agent"
+              : "Run agent"}
         </TooltipContent>
       </Tooltip>
       <RunInputDialog

autogpt_platform/frontend/src/app/(platform)/build/components/BuilderActions/components/RunInputDialog/RunInputDialog.tsx

@@ -10,6 +10,7 @@ import { useRunInputDialog } from "./useRunInputDialog";
 import { CronSchedulerDialog } from "../CronSchedulerDialog/CronSchedulerDialog";
 import { useTutorialStore } from "@/app/(platform)/build/stores/tutorialStore";
 import { useEffect } from "react";
+import { CredentialsGroupedView } from "@/components/contextual/CredentialsInput/components/CredentialsGroupedView/CredentialsGroupedView";
 
 export const RunInputDialog = ({
   isOpen,
@@ -23,19 +24,17 @@ export const RunInputDialog = ({
   const hasInputs = useGraphStore((state) => state.hasInputs);
   const hasCredentials = useGraphStore((state) => state.hasCredentials);
   const inputSchema = useGraphStore((state) => state.inputSchema);
-  const credentialsSchema = useGraphStore(
-    (state) => state.credentialsInputSchema,
-  );
 
   const {
-    credentialsUiSchema,
+    credentialFields,
+    requiredCredentials,
     handleManualRun,
     handleInputChange,
     openCronSchedulerDialog,
     setOpenCronSchedulerDialog,
     inputValues,
     credentialValues,
-    handleCredentialChange,
+    handleCredentialFieldChange,
     isExecutingGraph,
   } = useRunInputDialog({ setIsOpen });
 
@@ -62,67 +61,67 @@ export const RunInputDialog = ({
           isOpen,
           set: setIsOpen,
         }}
-        styling={{ maxWidth: "600px", minWidth: "600px" }}
+        styling={{ maxWidth: "700px", minWidth: "700px" }}
       >
         <Dialog.Content>
-          <div className="space-y-6 p-1" data-id="run-input-dialog-content">
-            {/* Credentials Section */}
-            {hasCredentials() && (
-              <div data-id="run-input-credentials-section">
-                <div className="mb-4">
-                  <Text variant="h4" className="text-gray-900">
-                    Credentials
-                  </Text>
+          <div
+            className="grid grid-cols-[1fr_auto] gap-10 p-1"
+            data-id="run-input-dialog-content"
+          >
+            <div className="space-y-6">
+              {/* Credentials Section */}
+              {hasCredentials() && credentialFields.length > 0 && (
+                <div data-id="run-input-credentials-section">
+                  <div className="mb-4">
+                    <Text variant="h4" className="text-gray-900">
+                      Credentials
+                    </Text>
+                  </div>
+                  <div className="px-2" data-id="run-input-credentials-form">
+                    <CredentialsGroupedView
+                      credentialFields={credentialFields}
+                      requiredCredentials={requiredCredentials}
+                      inputCredentials={credentialValues}
+                      inputValues={inputValues}
+                      onCredentialChange={handleCredentialFieldChange}
+                    />
+                  </div>
                 </div>
-                <div className="px-2" data-id="run-input-credentials-form">
-                  <FormRenderer
-                    jsonSchema={credentialsSchema as RJSFSchema}
-                    handleChange={(v) => handleCredentialChange(v.formData)}
-                    uiSchema={credentialsUiSchema}
-                    initialValues={{}}
-                    formContext={{
-                      showHandles: false,
-                      size: "large",
-                      showOptionalToggle: false,
-                    }}
-                  />
-                </div>
-              </div>
-            )}
+              )}
 
-            {/* Inputs Section */}
-            {hasInputs() && (
-              <div data-id="run-input-inputs-section">
-                <div className="mb-4">
-                  <Text variant="h4" className="text-gray-900">
-                    Inputs
-                  </Text>
+              {/* Inputs Section */}
+              {hasInputs() && (
+                <div data-id="run-input-inputs-section">
+                  <div className="mb-4">
+                    <Text variant="h4" className="text-gray-900">
+                      Inputs
+                    </Text>
+                  </div>
+                  <div data-id="run-input-inputs-form">
+                    <FormRenderer
+                      jsonSchema={inputSchema as RJSFSchema}
+                      handleChange={(v) => handleInputChange(v.formData)}
+                      uiSchema={uiSchema}
+                      initialValues={{}}
+                      formContext={{
+                        showHandles: false,
+                        size: "large",
+                      }}
+                    />
+                  </div>
                 </div>
-                <div data-id="run-input-inputs-form">
-                  <FormRenderer
-                    jsonSchema={inputSchema as RJSFSchema}
-                    handleChange={(v) => handleInputChange(v.formData)}
-                    uiSchema={uiSchema}
-                    initialValues={{}}
-                    formContext={{
-                      showHandles: false,
-                      size: "large",
-                    }}
-                  />
-                </div>
-              </div>
-            )}
+              )}
+            </div>
 
-            {/* Action Button */}
             <div
-              className="flex justify-end pt-2"
+              className="flex flex-col items-end justify-start"
               data-id="run-input-actions-section"
             >
               {purpose === "run" && (
                 <Button
                   variant="primary"
                   size="large"
-                  className="group h-fit min-w-0 gap-2"
+                  className="group h-fit min-w-0 gap-2 px-10"
                   onClick={handleManualRun}
                   loading={isExecutingGraph}
                   data-id="run-input-manual-run-button"
@@ -137,7 +136,7 @@ export const RunInputDialog = ({
                 <Button
                   variant="primary"
                   size="large"
-                  className="group h-fit min-w-0 gap-2"
+                  className="group h-fit min-w-0 gap-2 px-10"
                   onClick={() => setOpenCronSchedulerDialog(true)}
                   data-id="run-input-schedule-button"
                 >

autogpt_platform/frontend/src/app/(platform)/build/components/BuilderActions/components/RunInputDialog/useRunInputDialog.ts

@@ -7,12 +7,11 @@ import {
   GraphExecutionMeta,
 } from "@/lib/autogpt-server-api";
 import { parseAsInteger, parseAsString, useQueryStates } from "nuqs";
-import { useMemo, useState } from "react";
-import { uiSchema } from "../../../FlowEditor/nodes/uiSchema";
-import { isCredentialFieldSchema } from "@/components/renderers/InputRenderer/custom/CredentialField/helpers";
+import { useCallback, useMemo, useState } from "react";
 import { useNodeStore } from "@/app/(platform)/build/stores/nodeStore";
 import { useToast } from "@/components/molecules/Toast/use-toast";
 import { useReactFlow } from "@xyflow/react";
+import type { CredentialField } from "@/components/contextual/CredentialsInput/components/CredentialsGroupedView/helpers";
 
 export const useRunInputDialog = ({
   setIsOpen,
@@ -120,27 +119,32 @@ export const useRunInputDialog = ({
       },
     });
 
-  // We are rendering the credentials field differently compared to other fields.
-  // In the node, we have the field name as "credential" - so our library catches it and renders it differently.
-  // But here we have a different name, something like `Firecrawl credentials`, so here we are telling the library that this field is a credential field type.
+  // Convert credentials schema to credential fields array for CredentialsGroupedView
+  const credentialFields: CredentialField[] = useMemo(() => {
+    if (!credentialsSchema?.properties) return [];
+    return Object.entries(credentialsSchema.properties);
+  }, [credentialsSchema]);
 
-  const credentialsUiSchema = useMemo(() => {
-    const dynamicUiSchema: any = { ...uiSchema };
+  // Get required credentials as a Set
+  const requiredCredentials = useMemo(() => {
+    return new Set<string>(credentialsSchema?.required || []);
+  }, [credentialsSchema]);
 
-    if (credentialsSchema?.properties) {
-      Object.keys(credentialsSchema.properties).forEach((fieldName) => {
-        const fieldSchema = credentialsSchema.properties[fieldName];
-        if (isCredentialFieldSchema(fieldSchema)) {
-          dynamicUiSchema[fieldName] = {
-            ...dynamicUiSchema[fieldName],
-            "ui:field": "custom/credential_field",
-          };
+  // Handler for individual credential changes
+  const handleCredentialFieldChange = useCallback(
+    (key: string, value?: CredentialsMetaInput) => {
+      setCredentialValues((prev) => {
+        if (value) {
+          return { ...prev, [key]: value };
+        } else {
+          const next = { ...prev };
+          delete next[key];
+          return next;
         }
       });
-    }
-
-    return dynamicUiSchema;
-  }, [credentialsSchema]);
+    },
+    [],
+  );
 
   const handleManualRun = async () => {
     // Filter out incomplete credentials (those without a valid id)
@@ -173,12 +177,14 @@ export const useRunInputDialog = ({
   };
 
   return {
-    credentialsUiSchema,
+    credentialFields,
+    requiredCredentials,
     inputValues,
     credentialValues,
     isExecutingGraph,
     handleInputChange,
     handleCredentialChange,
+    handleCredentialFieldChange,
     handleManualRun,
     openCronSchedulerDialog,
     setOpenCronSchedulerDialog,

autogpt_platform/frontend/src/app/(platform)/build/components/FloatingSafeModeToogle.tsx

@@ -18,69 +18,110 @@ interface Props {
   fullWidth?: boolean;
 }
 
+interface SafeModeButtonProps {
+  isEnabled: boolean;
+  label: string;
+  tooltipEnabled: string;
+  tooltipDisabled: string;
+  onToggle: () => void;
+  isPending: boolean;
+  fullWidth?: boolean;
+}
+
+function SafeModeButton({
+  isEnabled,
+  label,
+  tooltipEnabled,
+  tooltipDisabled,
+  onToggle,
+  isPending,
+  fullWidth = false,
+}: SafeModeButtonProps) {
+  return (
+    <Tooltip delayDuration={100}>
+      <TooltipTrigger asChild>
+        <Button
+          variant={isEnabled ? "primary" : "outline"}
+          size="small"
+          onClick={onToggle}
+          disabled={isPending}
+          className={cn("justify-start", fullWidth ? "w-full" : "")}
+        >
+          {isEnabled ? (
+            <>
+              <ShieldCheckIcon weight="bold" size={16} />
+              <Text variant="body" className="text-zinc-200">
+                {label}: ON
+              </Text>
+            </>
+          ) : (
+            <>
+              <ShieldIcon weight="bold" size={16} />
+              <Text variant="body" className="text-zinc-600">
+                {label}: OFF
+              </Text>
+            </>
+          )}
+        </Button>
+      </TooltipTrigger>
+      <TooltipContent>
+        <div className="text-center">
+          <div className="font-medium">
+            {label}: {isEnabled ? "ON" : "OFF"}
+          </div>
+          <div className="mt-1 text-xs text-muted-foreground">
+            {isEnabled ? tooltipEnabled : tooltipDisabled}
+          </div>
+        </div>
+      </TooltipContent>
+    </Tooltip>
+  );
+}
+
 export function FloatingSafeModeToggle({
   graph,
   className,
   fullWidth = false,
 }: Props) {
   const {
-    currentSafeMode,
+    currentHITLSafeMode,
+    showHITLToggle,
+    handleHITLToggle,
+    currentSensitiveActionSafeMode,
+    showSensitiveActionToggle,
+    handleSensitiveActionToggle,
     isPending,
     shouldShowToggle,
-    isStateUndetermined,
-    handleToggle,
   } = useAgentSafeMode(graph);
 
-  if (!shouldShowToggle || isStateUndetermined || isPending) {
+  if (!shouldShowToggle || isPending) {
     return null;
   }
 
   return (
-    <div className={cn("fixed z-50", className)}>
-      <Tooltip delayDuration={100}>
-        <TooltipTrigger asChild>
-          <Button
-            variant={currentSafeMode! ? "primary" : "outline"}
-            key={graph.id}
-            size="small"
-            title={
-              currentSafeMode!
-                ? "Safe Mode: ON. Human in the loop blocks require manual review"
-                : "Safe Mode: OFF. Human in the loop blocks proceed automatically"
-            }
-            onClick={handleToggle}
-            className={cn(fullWidth ? "w-full" : "")}
-          >
-            {currentSafeMode! ? (
-              <>
-                <ShieldCheckIcon weight="bold" size={16} />
-                <Text variant="body" className="text-zinc-200">
-                  Safe Mode: ON
-                </Text>
-              </>
-            ) : (
-              <>
-                <ShieldIcon weight="bold" size={16} />
-                <Text variant="body" className="text-zinc-600">
-                  Safe Mode: OFF
-                </Text>
-              </>
-            )}
-          </Button>
-        </TooltipTrigger>
-        <TooltipContent>
-          <div className="text-center">
-            <div className="font-medium">
-              Safe Mode: {currentSafeMode! ? "ON" : "OFF"}
-            </div>
-            <div className="mt-1 text-xs text-muted-foreground">
-              {currentSafeMode!
-                ? "Human in the loop blocks require manual review"
-                : "Human in the loop blocks proceed automatically"}
-            </div>
-          </div>
-        </TooltipContent>
-      </Tooltip>
+    <div className={cn("fixed z-50 flex flex-col gap-2", className)}>
+      {showHITLToggle && (
+        <SafeModeButton
+          isEnabled={currentHITLSafeMode}
+          label="Human in the loop block approval"
+          tooltipEnabled="The agent will pause at human-in-the-loop blocks and wait for your approval"
+          tooltipDisabled="Human in the loop blocks will proceed automatically"
+          onToggle={handleHITLToggle}
+          isPending={isPending}
+          fullWidth={fullWidth}
+        />
+      )}
+      {showSensitiveActionToggle && (
+        <SafeModeButton
+          isEnabled={currentSensitiveActionSafeMode}
+          label="Sensitive actions blocks approval"
+          tooltipEnabled="The agent will pause at sensitive action blocks and wait for your approval"
+          tooltipDisabled="Sensitive action blocks will proceed automatically"
+          onToggle={handleSensitiveActionToggle}
+          isPending={isPending}
+          fullWidth={fullWidth}
+        />
+      )}
     </div>
   );
 }

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/Flow/components/CustomControl.tsx

@@ -53,14 +53,14 @@ export const CustomControls = memo(
     const controls = [
       {
         id: "zoom-in-button",
-        icon: <PlusIcon className="size-4" />,
+        icon: <PlusIcon className="size-3.5 text-zinc-600" />,
         label: "Zoom In",
         onClick: () => zoomIn(),
         className: "h-10 w-10 border-none",
       },
       {
         id: "zoom-out-button",
-        icon: <MinusIcon className="size-4" />,
+        icon: <MinusIcon className="size-3.5 text-zinc-600" />,
         label: "Zoom Out",
         onClick: () => zoomOut(),
         className: "h-10 w-10 border-none",
@@ -68,9 +68,9 @@ export const CustomControls = memo(
       {
         id: "tutorial-button",
         icon: isTutorialLoading ? (
-          <CircleNotchIcon className="size-4 animate-spin" />
+          <CircleNotchIcon className="size-3.5 animate-spin text-zinc-600" />
         ) : (
-          <ChalkboardIcon className="size-4" />
+          <ChalkboardIcon className="size-3.5 text-zinc-600" />
         ),
         label: isTutorialLoading ? "Loading Tutorial..." : "Start Tutorial",
         onClick: handleTutorialClick,
@@ -79,7 +79,7 @@ export const CustomControls = memo(
       },
       {
         id: "fit-view-button",
-        icon: <FrameCornersIcon className="size-4" />,
+        icon: <FrameCornersIcon className="size-3.5 text-zinc-600" />,
         label: "Fit View",
         onClick: () => fitView({ padding: 0.2, duration: 800, maxZoom: 1 }),
         className: "h-10 w-10 border-none",
@@ -87,9 +87,9 @@ export const CustomControls = memo(
       {
         id: "lock-button",
         icon: !isLocked ? (
-          <LockOpenIcon className="size-4" />
+          <LockOpenIcon className="size-3.5 text-zinc-600" />
         ) : (
-          <LockIcon className="size-4" />
+          <LockIcon className="size-3.5 text-zinc-600" />
         ),
         label: "Toggle Lock",
         onClick: () => setIsLocked(!isLocked),

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/Flow/useFlow.ts

@@ -139,14 +139,6 @@ export const useFlow = () => {
       useNodeStore.getState().setNodes([]);
       useNodeStore.getState().clearResolutionState();
       addNodes(customNodes);
-
-      // Sync hardcoded values with handle IDs.
-      // If a key–value field has a key without a value, the backend omits it from hardcoded values.
-      // But if a handleId exists for that key, it causes inconsistency.
-      // This ensures hardcoded values stay in sync with handle IDs.
-      customNodes.forEach((node) => {
-        useNodeStore.getState().syncHardcodedValuesWithHandleIds(node.id);
-      });
     }
   }, [customNodes, addNodes]);
 
@@ -158,6 +150,14 @@ export const useFlow = () => {
     }
   }, [graph?.links, addLinks]);
 
+  useEffect(() => {
+    if (customNodes.length > 0 && graph?.links) {
+      customNodes.forEach((node) => {
+        useNodeStore.getState().syncHardcodedValuesWithHandleIds(node.id);
+      });
+    }
+  }, [customNodes, graph?.links]);
+
   // update node execution status in nodes
   useEffect(() => {
     if (

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/edges/CustomEdge.tsx

@@ -19,6 +19,8 @@ export type CustomEdgeData = {
   beadUp?: number;
   beadDown?: number;
   beadData?: Map<string, NodeExecutionResult["status"]>;
+  edgeColorClass?: string;
+  edgeHexColor?: string;
 };
 
 export type CustomEdge = XYEdge<CustomEdgeData, "custom">;
@@ -36,7 +38,6 @@ const CustomEdge = ({
   selected,
 }: EdgeProps<CustomEdge>) => {
   const removeConnection = useEdgeStore((state) => state.removeEdge);
-  // Subscribe to the brokenEdgeIDs map and check if this edge is broken across any node
   const isBroken = useNodeStore((state) => state.isEdgeBroken(id));
   const [isHovered, setIsHovered] = useState(false);
 
@@ -52,6 +53,7 @@ const CustomEdge = ({
   const isStatic = data?.isStatic ?? false;
   const beadUp = data?.beadUp ?? 0;
   const beadDown = data?.beadDown ?? 0;
+  const edgeColorClass = data?.edgeColorClass;
 
   const handleRemoveEdge = () => {
     removeConnection(id);
@@ -70,7 +72,9 @@ const CustomEdge = ({
             ? "!stroke-red-500 !stroke-[2px] [stroke-dasharray:4]"
             : selected
               ? "stroke-zinc-800"
-              : "stroke-zinc-500/50 hover:stroke-zinc-500",
+              : edgeColorClass
+                ? cn(edgeColorClass, "opacity-70 hover:opacity-100")
+                : "stroke-zinc-500/50 hover:stroke-zinc-500",
         )}
       />
       <JSBeads

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/edges/useCustomEdge.ts

@@ -8,6 +8,7 @@ import { useCallback } from "react";
 import { useNodeStore } from "../../../stores/nodeStore";
 import { useHistoryStore } from "../../../stores/historyStore";
 import { CustomEdge } from "./CustomEdge";
+import { getEdgeColorFromOutputType } from "../nodes/helpers";
 
 export const useCustomEdge = () => {
   const edges = useEdgeStore((s) => s.edges);
@@ -34,8 +35,13 @@ export const useCustomEdge = () => {
       if (exists) return;
 
       const nodes = useNodeStore.getState().nodes;
-      const isStatic = nodes.find((n) => n.id === conn.source)?.data
-        ?.staticOutput;
+      const sourceNode = nodes.find((n) => n.id === conn.source);
+      const isStatic = sourceNode?.data?.staticOutput;
+
+      const { colorClass, hexColor } = getEdgeColorFromOutputType(
+        sourceNode?.data?.outputSchema,
+        conn.sourceHandle,
+      );
 
       addEdge({
         source: conn.source,
@@ -44,6 +50,8 @@ export const useCustomEdge = () => {
         targetHandle: conn.targetHandle,
         data: {
           isStatic,
+          edgeColorClass: colorClass,
+          edgeHexColor: hexColor,
         },
       });
     },

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/nodes/CustomNode/components/NodeOutput/NodeOutput.tsx

@@ -1,22 +1,21 @@
 import { Button } from "@/components/atoms/Button/Button";
 import { Text } from "@/components/atoms/Text/Text";
+import {
+  Accordion,
+  AccordionContent,
+  AccordionItem,
+  AccordionTrigger,
+} from "@/components/molecules/Accordion/Accordion";
 import { beautifyString, cn } from "@/lib/utils";
-import { CaretDownIcon, CopyIcon, CheckIcon } from "@phosphor-icons/react";
+import { CopyIcon, CheckIcon } from "@phosphor-icons/react";
 import { NodeDataViewer } from "./components/NodeDataViewer/NodeDataViewer";
 import { ContentRenderer } from "./components/ContentRenderer";
 import { useNodeOutput } from "./useNodeOutput";
 import { ViewMoreData } from "./components/ViewMoreData";
 
 export const NodeDataRenderer = ({ nodeId }: { nodeId: string }) => {
-  const {
-    outputData,
-    isExpanded,
-    setIsExpanded,
-    copiedKey,
-    handleCopy,
-    executionResultId,
-    inputData,
-  } = useNodeOutput(nodeId);
+  const { outputData, copiedKey, handleCopy, executionResultId, inputData } =
+    useNodeOutput(nodeId);
 
   if (Object.keys(outputData).length === 0) {
     return null;
@@ -25,122 +24,117 @@ export const NodeDataRenderer = ({ nodeId }: { nodeId: string }) => {
   return (
     <div
       data-tutorial-id={`node-output`}
-      className="flex flex-col gap-3 rounded-b-xl border-t border-zinc-200 px-4 py-4"
+      className="rounded-b-xl border-t border-zinc-200 px-4 py-2"
     >
-      <div className="flex items-center justify-between">
-        <Text variant="body-medium" className="!font-semibold text-slate-700">
-          Node Output
-        </Text>
-        <Button
-          variant="ghost"
-          size="small"
-          onClick={() => setIsExpanded(!isExpanded)}
-          className="h-fit min-w-0 p-1 text-slate-600 hover:text-slate-900"
-        >
-          <CaretDownIcon
-            size={16}
-            weight="bold"
-            className={`transition-transform ${isExpanded ? "rotate-180" : ""}`}
-          />
-        </Button>
-      </div>
+      <Accordion type="single" collapsible defaultValue="node-output">
+        <AccordionItem value="node-output" className="border-none">
+          <AccordionTrigger className="py-2 hover:no-underline">
+            <Text
+              variant="body-medium"
+              className="!font-semibold text-slate-700"
+            >
+              Node Output
+            </Text>
+          </AccordionTrigger>
+          <AccordionContent className="pt-2">
+            <div className="flex max-w-[350px] flex-col gap-4">
+              <div className="space-y-2">
+                <Text variant="small-medium">Input</Text>
 
-      {isExpanded && (
-        <>
-          <div className="flex max-w-[350px] flex-col gap-4">
-            <div className="space-y-2">
-              <Text variant="small-medium">Input</Text>
+                <ContentRenderer value={inputData} shortContent={false} />
 
-              <ContentRenderer value={inputData} shortContent={false} />
-
-              <div className="mt-1 flex justify-end gap-1">
-                <NodeDataViewer
-                  data={inputData}
-                  pinName="Input"
-                  execId={executionResultId}
-                />
-                <Button
-                  variant="secondary"
-                  size="small"
-                  onClick={() => handleCopy("input", inputData)}
-                  className={cn(
-                    "h-fit min-w-0 gap-1.5 border border-zinc-200 p-2 text-black hover:text-slate-900",
-                    copiedKey === "input" &&
-                      "border-green-400 bg-green-100 hover:border-green-400 hover:bg-green-200",
-                  )}
-                >
-                  {copiedKey === "input" ? (
-                    <CheckIcon size={12} className="text-green-600" />
-                  ) : (
-                    <CopyIcon size={12} />
-                  )}
-                </Button>
+                <div className="mt-1 flex justify-end gap-1">
+                  <NodeDataViewer
+                    data={inputData}
+                    pinName="Input"
+                    execId={executionResultId}
+                  />
+                  <Button
+                    variant="secondary"
+                    size="small"
+                    onClick={() => handleCopy("input", inputData)}
+                    className={cn(
+                      "h-fit min-w-0 gap-1.5 border border-zinc-200 p-2 text-black hover:text-slate-900",
+                      copiedKey === "input" &&
+                        "border-green-400 bg-green-100 hover:border-green-400 hover:bg-green-200",
+                    )}
+                  >
+                    {copiedKey === "input" ? (
+                      <CheckIcon size={12} className="text-green-600" />
+                    ) : (
+                      <CopyIcon size={12} />
+                    )}
+                  </Button>
+                </div>
               </div>
-            </div>
 
-            {Object.entries(outputData)
-              .slice(0, 2)
-              .map(([key, value]) => (
-                <div key={key} className="flex flex-col gap-2">
-                  <div className="flex items-center gap-2">
-                    <Text
-                      variant="small-medium"
-                      className="!font-semibold text-slate-600"
-                    >
-                      Pin:
-                    </Text>
-                    <Text variant="small" className="text-slate-700">
-                      {beautifyString(key)}
-                    </Text>
-                  </div>
-                  <div className="w-full space-y-2">
-                    <Text
-                      variant="small"
-                      className="!font-semibold text-slate-600"
-                    >
-                      Data:
-                    </Text>
-                    <div className="relative space-y-2">
-                      {value.map((item, index) => (
-                        <div key={index}>
-                          <ContentRenderer value={item} shortContent={true} />
+              {Object.entries(outputData)
+                .slice(0, 2)
+                .map(([key, value]) => (
+                  <div key={key} className="flex flex-col gap-2">
+                    <div className="flex items-center gap-2">
+                      <Text
+                        variant="small-medium"
+                        className="!font-semibold text-slate-600"
+                      >
+                        Pin:
+                      </Text>
+                      <Text variant="small" className="text-slate-700">
+                        {beautifyString(key)}
+                      </Text>
+                    </div>
+                    <div className="w-full space-y-2">
+                      <Text
+                        variant="small"
+                        className="!font-semibold text-slate-600"
+                      >
+                        Data:
+                      </Text>
+                      <div className="relative space-y-2">
+                        {value.map((item, index) => (
+                          <div key={index}>
+                            <ContentRenderer value={item} shortContent={true} />
+                          </div>
+                        ))}
+
+                        <div className="mt-1 flex justify-end gap-1">
+                          <NodeDataViewer
+                            data={value}
+                            pinName={key}
+                            execId={executionResultId}
+                          />
+                          <Button
+                            variant="secondary"
+                            size="small"
+                            onClick={() => handleCopy(key, value)}
+                            className={cn(
+                              "h-fit min-w-0 gap-1.5 border border-zinc-200 p-2 text-black hover:text-slate-900",
+                              copiedKey === key &&
+                                "border-green-400 bg-green-100 hover:border-green-400 hover:bg-green-200",
+                            )}
+                          >
+                            {copiedKey === key ? (
+                              <CheckIcon size={12} className="text-green-600" />
+                            ) : (
+                              <CopyIcon size={12} />
+                            )}
+                          </Button>
                         </div>
-                      ))}
-
-                      <div className="mt-1 flex justify-end gap-1">
-                        <NodeDataViewer
-                          data={value}
-                          pinName={key}
-                          execId={executionResultId}
-                        />
-                        <Button
-                          variant="secondary"
-                          size="small"
-                          onClick={() => handleCopy(key, value)}
-                          className={cn(
-                            "h-fit min-w-0 gap-1.5 border border-zinc-200 p-2 text-black hover:text-slate-900",
-                            copiedKey === key &&
-                              "border-green-400 bg-green-100 hover:border-green-400 hover:bg-green-200",
-                          )}
-                        >
-                          {copiedKey === key ? (
-                            <CheckIcon size={12} className="text-green-600" />
-                          ) : (
-                            <CopyIcon size={12} />
-                          )}
-                        </Button>
                       </div>
                     </div>
                   </div>
-                </div>
-              ))}
-          </div>
+                ))}
+            </div>
 
-          {Object.keys(outputData).length > 2 && (
-            <ViewMoreData outputData={outputData} execId={executionResultId} />
-          )}
-        </>
-      )}
+            {Object.keys(outputData).length > 2 && (
+              <ViewMoreData
+                outputData={outputData}
+                execId={executionResultId}
+              />
+            )}
+          </AccordionContent>
+        </AccordionItem>
+      </Accordion>
     </div>
   );
 };

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/nodes/CustomNode/components/NodeOutput/useNodeOutput.tsx

@@ -4,7 +4,6 @@ import { useShallow } from "zustand/react/shallow";
 import { useState } from "react";
 
 export const useNodeOutput = (nodeId: string) => {
-  const [isExpanded, setIsExpanded] = useState(true);
   const [copiedKey, setCopiedKey] = useState<string | null>(null);
   const { toast } = useToast();
 
@@ -37,13 +36,10 @@ export const useNodeOutput = (nodeId: string) => {
     }
   };
   return {
-    outputData: outputData,
-    inputData: inputData,
-    isExpanded: isExpanded,
-    setIsExpanded: setIsExpanded,
-    copiedKey: copiedKey,
-    setCopiedKey: setCopiedKey,
-    handleCopy: handleCopy,
+    outputData,
+    inputData,
+    copiedKey,
+    handleCopy,
     executionResultId: nodeExecutionResult?.node_exec_id,
   };
 };

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/nodes/helpers.ts

@@ -187,3 +187,38 @@ export const getTypeDisplayInfo = (schema: any) => {
     hexColor,
   };
 };
+
+export function getEdgeColorFromOutputType(
+  outputSchema: RJSFSchema | undefined,
+  sourceHandle: string,
+): { colorClass: string; hexColor: string } {
+  const defaultColor = {
+    colorClass: "stroke-zinc-500/50",
+    hexColor: "#6b7280",
+  };
+
+  if (!outputSchema?.properties) return defaultColor;
+
+  const properties = outputSchema.properties as Record<string, unknown>;
+  const handleParts = sourceHandle.split("_#_");
+  let currentSchema: Record<string, unknown> = properties;
+
+  for (let i = 0; i < handleParts.length; i++) {
+    const part = handleParts[i];
+    const fieldSchema = currentSchema[part] as Record<string, unknown>;
+    if (!fieldSchema) return defaultColor;
+
+    if (i === handleParts.length - 1) {
+      const { hexColor, colorClass } = getTypeDisplayInfo(fieldSchema);
+      return { colorClass: colorClass.replace("!text-", "stroke-"), hexColor };
+    }
+
+    if (fieldSchema.properties) {
+      currentSchema = fieldSchema.properties as Record<string, unknown>;
+    } else {
+      return defaultColor;
+    }
+  }
+
+  return defaultColor;
+}

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/tutorial/icons.ts

@@ -1,7 +1,32 @@
-// These are SVG Phosphor icons
+type IconOptions = {
+  size?: number;
+  color?: string;
+};
+
+const DEFAULT_SIZE = 16;
+const DEFAULT_COLOR = "#52525b"; // zinc-600
+
+const iconPaths = {
+  ClickIcon: `M88,24V16a8,8,0,0,1,16,0v8a8,8,0,0,1-16,0ZM16,104h8a8,8,0,0,0,0-16H16a8,8,0,0,0,0,16ZM124.42,39.16a8,8,0,0,0,10.74-3.58l8-16a8,8,0,0,0-14.31-7.16l-8,16A8,8,0,0,0,124.42,39.16Zm-96,81.69-16,8a8,8,0,0,0,7.16,14.31l16-8a8,8,0,1,0-7.16-14.31ZM219.31,184a16,16,0,0,1,0,22.63l-12.68,12.68a16,16,0,0,1-22.63,0L132.7,168,115,214.09c0,.1-.08.21-.13.32a15.83,15.83,0,0,1-14.6,9.59l-.79,0a15.83,15.83,0,0,1-14.41-11L32.8,52.92A16,16,0,0,1,52.92,32.8L213,85.07a16,16,0,0,1,1.41,29.8l-.32.13L168,132.69ZM208,195.31,156.69,144h0a16,16,0,0,1,4.93-26l.32-.14,45.95-17.64L48,48l52.2,159.86,17.65-46c0-.11.08-.22.13-.33a16,16,0,0,1,11.69-9.34,16.72,16.72,0,0,1,3-.28,16,16,0,0,1,11.3,4.69L195.31,208Z`,
+  Keyboard: `M224,48H32A16,16,0,0,0,16,64V192a16,16,0,0,0,16,16H224a16,16,0,0,0,16-16V64A16,16,0,0,0,224,48Zm0,144H32V64H224V192Zm-16-64a8,8,0,0,1-8,8H56a8,8,0,0,1,0-16H200A8,8,0,0,1,208,128Zm0-32a8,8,0,0,1-8,8H56a8,8,0,0,1,0-16H200A8,8,0,0,1,208,96ZM72,160a8,8,0,0,1-8,8H56a8,8,0,0,1,0-16h8A8,8,0,0,1,72,160Zm96,0a8,8,0,0,1-8,8H96a8,8,0,0,1,0-16h64A8,8,0,0,1,168,160Zm40,0a8,8,0,0,1-8,8h-8a8,8,0,0,1,0-16h8A8,8,0,0,1,208,160Z`,
+  Drag: `M188,80a27.79,27.79,0,0,0-13.36,3.4,28,28,0,0,0-46.64-11A28,28,0,0,0,80,92v20H68a28,28,0,0,0-28,28v12a88,88,0,0,0,176,0V108A28,28,0,0,0,188,80Zm12,72a72,72,0,0,1-144,0V140a12,12,0,0,1,12-12H80v24a8,8,0,0,0,16,0V92a12,12,0,0,1,24,0v28a8,8,0,0,0,16,0V92a12,12,0,0,1,24,0v28a8,8,0,0,0,16,0V108a12,12,0,0,1,24,0Z`,
+};
+
+function createIcon(path: string, options: IconOptions = {}): string {
+  const size = options.size ?? DEFAULT_SIZE;
+  const color = options.color ?? DEFAULT_COLOR;
+  return `<svg xmlns="http://www.w3.org/2000/svg" width="${size}" height="${size}" fill="${color}" viewBox="0 0 256 256"><path d="${path}"></path></svg>`;
+}
 
 export const ICONS = {
-  ClickIcon: `<svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="#000000" viewBox="0 0 256 256"><path d="M88,24V16a8,8,0,0,1,16,0v8a8,8,0,0,1-16,0ZM16,104h8a8,8,0,0,0,0-16H16a8,8,0,0,0,0,16ZM124.42,39.16a8,8,0,0,0,10.74-3.58l8-16a8,8,0,0,0-14.31-7.16l-8,16A8,8,0,0,0,124.42,39.16Zm-96,81.69-16,8a8,8,0,0,0,7.16,14.31l16-8a8,8,0,1,0-7.16-14.31ZM219.31,184a16,16,0,0,1,0,22.63l-12.68,12.68a16,16,0,0,1-22.63,0L132.7,168,115,214.09c0,.1-.08.21-.13.32a15.83,15.83,0,0,1-14.6,9.59l-.79,0a15.83,15.83,0,0,1-14.41-11L32.8,52.92A16,16,0,0,1,52.92,32.8L213,85.07a16,16,0,0,1,1.41,29.8l-.32.13L168,132.69ZM208,195.31,156.69,144h0a16,16,0,0,1,4.93-26l.32-.14,45.95-17.64L48,48l52.2,159.86,17.65-46c0-.11.08-.22.13-.33a16,16,0,0,1,11.69-9.34,16.72,16.72,0,0,1,3-.28,16,16,0,0,1,11.3,4.69L195.31,208Z"></path></svg>`,
-  Keyboard: `<svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="#000000" viewBox="0 0 256 256"><path d="M224,48H32A16,16,0,0,0,16,64V192a16,16,0,0,0,16,16H224a16,16,0,0,0,16-16V64A16,16,0,0,0,224,48Zm0,144H32V64H224V192Zm-16-64a8,8,0,0,1-8,8H56a8,8,0,0,1,0-16H200A8,8,0,0,1,208,128Zm0-32a8,8,0,0,1-8,8H56a8,8,0,0,1,0-16H200A8,8,0,0,1,208,96ZM72,160a8,8,0,0,1-8,8H56a8,8,0,0,1,0-16h8A8,8,0,0,1,72,160Zm96,0a8,8,0,0,1-8,8H96a8,8,0,0,1,0-16h64A8,8,0,0,1,168,160Zm40,0a8,8,0,0,1-8,8h-8a8,8,0,0,1,0-16h8A8,8,0,0,1,208,160Z"></path></svg>`,
-  Drag: `<svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="#000000" viewBox="0 0 256 256"><path d="M188,80a27.79,27.79,0,0,0-13.36,3.4,28,28,0,0,0-46.64-11A28,28,0,0,0,80,92v20H68a28,28,0,0,0-28,28v12a88,88,0,0,0,176,0V108A28,28,0,0,0,188,80Zm12,72a72,72,0,0,1-144,0V140a12,12,0,0,1,12-12H80v24a8,8,0,0,0,16,0V92a12,12,0,0,1,24,0v28a8,8,0,0,0,16,0V92a12,12,0,0,1,24,0v28a8,8,0,0,0,16,0V108a12,12,0,0,1,24,0Z"></path></svg>`,
+  ClickIcon: createIcon(iconPaths.ClickIcon),
+  Keyboard: createIcon(iconPaths.Keyboard),
+  Drag: createIcon(iconPaths.Drag),
 };
+
+export function getIcon(
+  name: keyof typeof iconPaths,
+  options?: IconOptions,
+): string {
+  return createIcon(iconPaths[name], options);
+}

autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/tutorial/index.ts

@@ -11,6 +11,7 @@ import {
 } from "./helpers";
 import { useNodeStore } from "../../../stores/nodeStore";
 import { useEdgeStore } from "../../../stores/edgeStore";
+import { useTutorialStore } from "../../../stores/tutorialStore";
 
 let isTutorialLoading = false;
 let tutorialLoadingCallback: ((loading: boolean) => void) | null = null;
@@ -60,12 +61,14 @@ export const startTutorial = async () => {
     handleTutorialComplete();
     removeTutorialStyles();
     clearPrefetchedBlocks();
+    useTutorialStore.getState().setIsTutorialRunning(false);
   });
 
   tour.on("cancel", () => {
     handleTutorialCancel(tour);
     removeTutorialStyles();
     clearPrefetchedBlocks();
+    useTutorialStore.getState().setIsTutorialRunning(false);
   });
 
   for (const step of tour.steps) {

autogpt_platform/frontend/src/app/(platform)/build/components/helper.ts

@@ -61,12 +61,18 @@ export const convertNodesPlusBlockInfoIntoCustomNodes = (
   return customNode;
 };
 
+const isToolSourceName = (sourceName: string): boolean =>
+  sourceName.startsWith("tools_^_");
+
+const cleanupSourceName = (sourceName: string): string =>
+  isToolSourceName(sourceName) ? "tools" : sourceName;
+
 export const linkToCustomEdge = (link: Link): CustomEdge => ({
   id: link.id ?? "",
   type: "custom" as const,
   source: link.source_id,
   target: link.sink_id,
-  sourceHandle: link.source_name,
+  sourceHandle: cleanupSourceName(link.source_name),
   targetHandle: link.sink_name,
   data: {
     isStatic: link.is_static,

autogpt_platform/frontend/src/app/(platform)/chat/components/Chat/components/ChatContainer/helpers.ts

@@ -267,23 +267,34 @@ export function extractCredentialsNeeded(
       | undefined;
     if (missingCreds && Object.keys(missingCreds).length > 0) {
       const agentName = (setupInfo?.agent_name as string) || "this block";
-      const credentials = Object.values(missingCreds).map((credInfo) => ({
-        provider: (credInfo.provider as string) || "unknown",
-        providerName:
-          (credInfo.provider_name as string) ||
-          (credInfo.provider as string) ||
-          "Unknown Provider",
-        credentialType:
+      const credentials = Object.values(missingCreds).map((credInfo) => {
+        // Normalize to array at boundary - prefer 'types' array, fall back to single 'type'
+        const typesArray = credInfo.types as
+          | Array<"api_key" | "oauth2" | "user_password" | "host_scoped">
+          | undefined;
+        const singleType =
           (credInfo.type as
             | "api_key"
             | "oauth2"
             | "user_password"
-            | "host_scoped") || "api_key",
-        title:
-          (credInfo.title as string) ||
-          `${(credInfo.provider_name as string) || (credInfo.provider as string)} credentials`,
-        scopes: credInfo.scopes as string[] | undefined,
-      }));
+            | "host_scoped"
+            | undefined) || "api_key";
+        const credentialTypes =
+          typesArray && typesArray.length > 0 ? typesArray : [singleType];
+
+        return {
+          provider: (credInfo.provider as string) || "unknown",
+          providerName:
+            (credInfo.provider_name as string) ||
+            (credInfo.provider as string) ||
+            "Unknown Provider",
+          credentialTypes,
+          title:
+            (credInfo.title as string) ||
+            `${(credInfo.provider_name as string) || (credInfo.provider as string)} credentials`,
+          scopes: credInfo.scopes as string[] | undefined,
+        };
+      });
       return {
         type: "credentials_needed",
         toolName,
@@ -358,11 +369,14 @@ export function extractInputsNeeded(
       credentials.forEach((cred) => {
         const id = cred.id as string;
         if (id) {
+          const credentialTypes = Array.isArray(cred.types)
+            ? cred.types
+            : [(cred.type as string) || "api_key"];
           credentialsSchema[id] = {
             type: "object",
             properties: {},
             credentials_provider: [cred.provider as string],
-            credentials_types: [(cred.type as string) || "api_key"],
+            credentials_types: credentialTypes,
             credentials_scopes: cred.scopes as string[] | undefined,
           };
         }

autogpt_platform/frontend/src/app/(platform)/chat/components/Chat/components/ChatCredentialsSetup/ChatCredentialsSetup.tsx

@@ -9,7 +9,9 @@ import { useChatCredentialsSetup } from "./useChatCredentialsSetup";
 export interface CredentialInfo {
   provider: string;
   providerName: string;
-  credentialType: "api_key" | "oauth2" | "user_password" | "host_scoped";
+  credentialTypes: Array<
+    "api_key" | "oauth2" | "user_password" | "host_scoped"
+  >;
   title: string;
   scopes?: string[];
 }
@@ -30,7 +32,7 @@ function createSchemaFromCredentialInfo(
     type: "object",
     properties: {},
     credentials_provider: [credential.provider],
-    credentials_types: [credential.credentialType],
+    credentials_types: credential.credentialTypes,
     credentials_scopes: credential.scopes,
     discriminator: undefined,
     discriminator_mapping: undefined,

autogpt_platform/frontend/src/app/(platform)/chat/components/Chat/components/ChatMessage/useChatMessage.ts

@@ -41,7 +41,9 @@ export type ChatMessageData =
       credentials: Array<{
         provider: string;
         providerName: string;
-        credentialType: "api_key" | "oauth2" | "user_password" | "host_scoped";
+        credentialTypes: Array<
+          "api_key" | "oauth2" | "user_password" | "host_scoped"
+        >;
         title: string;
         scopes?: string[];
       }>;

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/NewAgentLibraryView/components/modals/AgentSettingsModal/AgentSettingsModal.tsx

@@ -31,10 +31,18 @@ export function AgentSettingsModal({
     }
   }
 
-  const { currentSafeMode, isPending, hasHITLBlocks, handleToggle } =
-    useAgentSafeMode(agent);
+  const {
+    currentHITLSafeMode,
+    showHITLToggle,
+    handleHITLToggle,
+    currentSensitiveActionSafeMode,
+    showSensitiveActionToggle,
+    handleSensitiveActionToggle,
+    isPending,
+    shouldShowToggle,
+  } = useAgentSafeMode(agent);
 
-  if (!hasHITLBlocks) return null;
+  if (!shouldShowToggle) return null;
 
   return (
     <Dialog
@@ -57,23 +65,48 @@ export function AgentSettingsModal({
       )}
       <Dialog.Content>
         <div className="space-y-6">
-          <div className="flex w-full flex-col items-start gap-4 rounded-xl border border-zinc-100 bg-white p-6">
-            <div className="flex w-full items-start justify-between gap-4">
-              <div className="flex-1">
-                <Text variant="large-semibold">Require human approval</Text>
-                <Text variant="large" className="mt-1 text-zinc-900">
-                  The agent will pause and wait for your review before
-                  continuing
-                </Text>
+          {showHITLToggle && (
+            <div className="flex w-full flex-col items-start gap-4 rounded-xl border border-zinc-100 bg-white p-6">
+              <div className="flex w-full items-start justify-between gap-4">
+                <div className="flex-1">
+                  <Text variant="large-semibold">
+                    Human-in-the-loop approval
+                  </Text>
+                  <Text variant="large" className="mt-1 text-zinc-900">
+                    The agent will pause at human-in-the-loop blocks and wait
+                    for your review before continuing
+                  </Text>
+                </div>
+                <Switch
+                  checked={currentHITLSafeMode || false}
+                  onCheckedChange={handleHITLToggle}
+                  disabled={isPending}
+                  className="mt-1"
+                />
               </div>
-              <Switch
-                checked={currentSafeMode || false}
-                onCheckedChange={handleToggle}
-                disabled={isPending}
-                className="mt-1"
-              />
             </div>
-          </div>
+          )}
+          {showSensitiveActionToggle && (
+            <div className="flex w-full flex-col items-start gap-4 rounded-xl border border-zinc-100 bg-white p-6">
+              <div className="flex w-full items-start justify-between gap-4">
+                <div className="flex-1">
+                  <Text variant="large-semibold">
+                    Sensitive action approval
+                  </Text>
+                  <Text variant="large" className="mt-1 text-zinc-900">
+                    The agent will pause at sensitive action blocks and wait for
+                    your review before continuing
+                  </Text>
+                </div>
+                <Switch
+                  checked={currentSensitiveActionSafeMode}
+                  onCheckedChange={handleSensitiveActionToggle}
+                  disabled={isPending}
+                  className="mt-1"
+                />
+              </div>
+            </div>
+          )}
         </div>
       </Dialog.Content>
     </Dialog>

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/NewAgentLibraryView/components/modals/RunAgentModal/RunAgentModal.tsx

@@ -14,6 +14,10 @@ import {
 import { Dialog } from "@/components/molecules/Dialog/Dialog";
 import { useEffect, useRef, useState } from "react";
 import { ScheduleAgentModal } from "../ScheduleAgentModal/ScheduleAgentModal";
+import {
+  AIAgentSafetyPopup,
+  useAIAgentSafetyPopup,
+} from "./components/AIAgentSafetyPopup/AIAgentSafetyPopup";
 import { ModalHeader } from "./components/ModalHeader/ModalHeader";
 import { ModalRunSection } from "./components/ModalRunSection/ModalRunSection";
 import { RunActions } from "./components/RunActions/RunActions";
@@ -83,8 +87,18 @@ export function RunAgentModal({
 
   const [isScheduleModalOpen, setIsScheduleModalOpen] = useState(false);
   const [hasOverflow, setHasOverflow] = useState(false);
+  const [isSafetyPopupOpen, setIsSafetyPopupOpen] = useState(false);
+  const [pendingRunAction, setPendingRunAction] = useState<(() => void) | null>(
+    null,
+  );
   const contentRef = useRef<HTMLDivElement>(null);
 
+  const { shouldShowPopup, dismissPopup } = useAIAgentSafetyPopup(
+    agent.id,
+    agent.has_sensitive_action,
+    agent.has_human_in_the_loop,
+  );
+
   const hasAnySetupFields =
     Object.keys(agentInputFields || {}).length > 0 ||
     Object.keys(agentCredentialsInputFields || {}).length > 0;
@@ -165,6 +179,24 @@ export function RunAgentModal({
     onScheduleCreated?.(schedule);
   }
 
+  function handleRunWithSafetyCheck() {
+    if (shouldShowPopup) {
+      setPendingRunAction(() => handleRun);
+      setIsSafetyPopupOpen(true);
+    } else {
+      handleRun();
+    }
+  }
+
+  function handleSafetyPopupAcknowledge() {
+    setIsSafetyPopupOpen(false);
+    dismissPopup();
+    if (pendingRunAction) {
+      pendingRunAction();
+      setPendingRunAction(null);
+    }
+  }
+
   return (
     <>
       <Dialog
@@ -248,7 +280,7 @@ export function RunAgentModal({
                 )}
                 <RunActions
                   defaultRunType={defaultRunType}
-                  onRun={handleRun}
+                  onRun={handleRunWithSafetyCheck}
                   isExecuting={isExecuting}
                   isSettingUpTrigger={isSettingUpTrigger}
                   isRunReady={allRequiredInputsAreSet}
@@ -266,6 +298,12 @@ export function RunAgentModal({
           </div>
         </Dialog.Content>
       </Dialog>
+
+      <AIAgentSafetyPopup
+        agentId={agent.id}
+        isOpen={isSafetyPopupOpen}
+        onAcknowledge={handleSafetyPopupAcknowledge}
+      />
     </>
   );
 }

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/NewAgentLibraryView/components/modals/RunAgentModal/components/AIAgentSafetyPopup/AIAgentSafetyPopup.tsx

@@ -0,0 +1,108 @@
+"use client";
+
+import { Button } from "@/components/atoms/Button/Button";
+import { Text } from "@/components/atoms/Text/Text";
+import { Dialog } from "@/components/molecules/Dialog/Dialog";
+import { Key, storage } from "@/services/storage/local-storage";
+import { ShieldCheckIcon } from "@phosphor-icons/react";
+import { useCallback, useEffect, useState } from "react";
+
+interface Props {
+  agentId: string;
+  onAcknowledge: () => void;
+  isOpen: boolean;
+}
+
+export function AIAgentSafetyPopup({ agentId, onAcknowledge, isOpen }: Props) {
+  function handleAcknowledge() {
+    // Add this agent to the list of agents for which popup has been shown
+    const seenAgentsJson = storage.get(Key.AI_AGENT_SAFETY_POPUP_SHOWN);
+    const seenAgents: string[] = seenAgentsJson
+      ? JSON.parse(seenAgentsJson)
+      : [];
+
+    if (!seenAgents.includes(agentId)) {
+      seenAgents.push(agentId);
+      storage.set(Key.AI_AGENT_SAFETY_POPUP_SHOWN, JSON.stringify(seenAgents));
+    }
+
+    onAcknowledge();
+  }
+
+  if (!isOpen) return null;
+
+  return (
+    <Dialog
+      controlled={{ isOpen, set: () => {} }}
+      styling={{ maxWidth: "480px" }}
+    >
+      <Dialog.Content>
+        <div className="flex flex-col items-center p-6 text-center">
+          <div className="mb-6 flex h-16 w-16 items-center justify-center rounded-full bg-blue-50">
+            <ShieldCheckIcon
+              weight="fill"
+              size={32}
+              className="text-blue-600"
+            />
+          </div>
+
+          <Text variant="h3" className="mb-4">
+            Safety Checks Enabled
+          </Text>
+
+          <Text variant="body" className="mb-2 text-zinc-700">
+            AI-generated agents may take actions that affect your data or
+            external systems.
+          </Text>
+
+          <Text variant="body" className="mb-8 text-zinc-700">
+            AutoGPT includes safety checks so you&apos;ll always have the
+            opportunity to review and approve sensitive actions before they
+            happen.
+          </Text>
+
+          <Button
+            variant="primary"
+            size="large"
+            className="w-full"
+            onClick={handleAcknowledge}
+          >
+            Got it
+          </Button>
+        </div>
+      </Dialog.Content>
+    </Dialog>
+  );
+}
+
+export function useAIAgentSafetyPopup(
+  agentId: string,
+  hasSensitiveAction: boolean,
+  hasHumanInTheLoop: boolean,
+) {
+  const [shouldShowPopup, setShouldShowPopup] = useState(false);
+  const [hasChecked, setHasChecked] = useState(false);
+
+  useEffect(() => {
+    if (hasChecked) return;
+
+    const seenAgentsJson = storage.get(Key.AI_AGENT_SAFETY_POPUP_SHOWN);
+    const seenAgents: string[] = seenAgentsJson
+      ? JSON.parse(seenAgentsJson)
+      : [];
+    const hasSeenPopupForThisAgent = seenAgents.includes(agentId);
+    const isRelevantAgent = hasSensitiveAction || hasHumanInTheLoop;
+
+    setShouldShowPopup(!hasSeenPopupForThisAgent && isRelevantAgent);
+    setHasChecked(true);
+  }, [agentId, hasSensitiveAction, hasHumanInTheLoop, hasChecked]);
+
+  const dismissPopup = useCallback(() => {
+    setShouldShowPopup(false);
+  }, []);
+
+  return {
+    shouldShowPopup,
+    dismissPopup,
+  };
+}

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/NewAgentLibraryView/components/modals/RunAgentModal/components/ModalRunSection/ModalRunSection.tsx

@@ -1,9 +1,9 @@
 import { Input } from "@/components/atoms/Input/Input";
+import { CredentialsGroupedView } from "@/components/contextual/CredentialsInput/components/CredentialsGroupedView/CredentialsGroupedView";
 import { InformationTooltip } from "@/components/molecules/InformationTooltip/InformationTooltip";
 import { useMemo } from "react";
 import { RunAgentInputs } from "../../../RunAgentInputs/RunAgentInputs";
 import { useRunAgentModalContext } from "../../context";
-import { CredentialsGroupedView } from "../CredentialsGroupedView/CredentialsGroupedView";
 import { ModalSection } from "../ModalSection/ModalSection";
 import { WebhookTriggerBanner } from "../WebhookTriggerBanner/WebhookTriggerBanner";
 
@@ -19,6 +19,8 @@ export function ModalRunSection() {
     setInputValue,
     agentInputFields,
     agentCredentialsInputFields,
+    inputCredentials,
+    setInputCredentialsValue,
   } = useRunAgentModalContext();
 
   const inputFields = Object.entries(agentInputFields || {});
@@ -102,6 +104,9 @@ export function ModalRunSection() {
           <CredentialsGroupedView
             credentialFields={credentialFields}
             requiredCredentials={requiredCredentials}
+            inputCredentials={inputCredentials}
+            inputValues={inputValues}
+            onCredentialChange={setInputCredentialsValue}
           />
         </ModalSection>
       ) : null}

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/NewAgentLibraryView/components/selected-views/SelectedRunView/components/SafeModeToggle.tsx

@@ -5,48 +5,104 @@ import { Graph } from "@/lib/autogpt-server-api/types";
 import { cn } from "@/lib/utils";
 import { ShieldCheckIcon, ShieldIcon } from "@phosphor-icons/react";
 import { useAgentSafeMode } from "@/hooks/useAgentSafeMode";
+import {
+  Tooltip,
+  TooltipContent,
+  TooltipTrigger,
+} from "@/components/atoms/Tooltip/BaseTooltip";
 
 interface Props {
   graph: GraphModel | LibraryAgent | Graph;
   className?: string;
-  fullWidth?: boolean;
 }
 
-export function SafeModeToggle({ graph }: Props) {
+interface SafeModeIconButtonProps {
+  isEnabled: boolean;
+  label: string;
+  tooltipEnabled: string;
+  tooltipDisabled: string;
+  onToggle: () => void;
+  isPending: boolean;
+}
+
+function SafeModeIconButton({
+  isEnabled,
+  label,
+  tooltipEnabled,
+  tooltipDisabled,
+  onToggle,
+  isPending,
+}: SafeModeIconButtonProps) {
+  return (
+    <Tooltip delayDuration={100}>
+      <TooltipTrigger asChild>
+        <Button
+          variant="icon"
+          size="icon"
+          aria-label={`${label}: ${isEnabled ? "ON" : "OFF"}. ${isEnabled ? tooltipEnabled : tooltipDisabled}`}
+          onClick={onToggle}
+          disabled={isPending}
+          className={cn(isPending ? "opacity-0" : "opacity-100")}
+        >
+          {isEnabled ? (
+            <ShieldCheckIcon weight="bold" size={16} />
+          ) : (
+            <ShieldIcon weight="bold" size={16} />
+          )}
+        </Button>
+      </TooltipTrigger>
+      <TooltipContent>
+        <div className="text-center">
+          <div className="font-medium">
+            {label}: {isEnabled ? "ON" : "OFF"}
+          </div>
+          <div className="mt-1 text-xs text-muted-foreground">
+            {isEnabled ? tooltipEnabled : tooltipDisabled}
+          </div>
+        </div>
+      </TooltipContent>
+    </Tooltip>
+  );
+}
+
+export function SafeModeToggle({ graph, className }: Props) {
   const {
-    currentSafeMode,
+    currentHITLSafeMode,
+    showHITLToggle,
+    handleHITLToggle,
+    currentSensitiveActionSafeMode,
+    showSensitiveActionToggle,
+    handleSensitiveActionToggle,
     isPending,
     shouldShowToggle,
-    isStateUndetermined,
-    handleToggle,
   } = useAgentSafeMode(graph);
 
-  if (!shouldShowToggle || isStateUndetermined) {
+  if (!shouldShowToggle) {
     return null;
   }
 
   return (
-    <Button
-      variant="icon"
-      key={graph.id}
-      size="icon"
-      aria-label={
-        currentSafeMode!
-          ? "Safe Mode: ON. Human in the loop blocks require manual review"
-          : "Safe Mode: OFF. Human in the loop blocks proceed automatically"
-      }
-      onClick={handleToggle}
-      className={cn(isPending ? "opacity-0" : "opacity-100")}
-    >
-      {currentSafeMode! ? (
-        <>
-          <ShieldCheckIcon weight="bold" size={16} />
-        </>
-      ) : (
-        <>
-          <ShieldIcon weight="bold" size={16} />
-        </>
+    <div className={cn("flex gap-1", className)}>
+      {showHITLToggle && (
+        <SafeModeIconButton
+          isEnabled={currentHITLSafeMode}
+          label="Human-in-the-loop"
+          tooltipEnabled="The agent will pause at human-in-the-loop blocks and wait for your approval"
+          tooltipDisabled="Human-in-the-loop blocks will proceed automatically"
+          onToggle={handleHITLToggle}
+          isPending={isPending}
+        />
       )}
-    </Button>
+      {showSensitiveActionToggle && (
+        <SafeModeIconButton
+          isEnabled={currentSensitiveActionSafeMode}
+          label="Sensitive actions"
+          tooltipEnabled="The agent will pause at sensitive action blocks and wait for your approval"
+          tooltipDisabled="Sensitive action blocks will proceed automatically"
+          onToggle={handleSensitiveActionToggle}
+          isPending={isPending}
+        />
+      )}
+    </div>
   );
 }

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/NewAgentLibraryView/components/selected-views/SelectedSettingsView/SelectedSettingsView.tsx

@@ -13,8 +13,16 @@ interface Props {
 }
 
 export function SelectedSettingsView({ agent, onClearSelectedRun }: Props) {
-  const { currentSafeMode, isPending, hasHITLBlocks, handleToggle } =
-    useAgentSafeMode(agent);
+  const {
+    currentHITLSafeMode,
+    showHITLToggle,
+    handleHITLToggle,
+    currentSensitiveActionSafeMode,
+    showSensitiveActionToggle,
+    handleSensitiveActionToggle,
+    isPending,
+    shouldShowToggle,
+  } = useAgentSafeMode(agent);
 
   return (
     <SelectedViewLayout agent={agent}>
@@ -34,24 +42,51 @@ export function SelectedSettingsView({ agent, onClearSelectedRun }: Props) {
         </div>
 
         <div className={`${AGENT_LIBRARY_SECTION_PADDING_X} space-y-6`}>
-          {hasHITLBlocks ? (
-            <div className="flex w-full max-w-2xl flex-col items-start gap-4 rounded-xl border border-zinc-100 bg-white p-6">
-              <div className="flex w-full items-start justify-between gap-4">
-                <div className="flex-1">
-                  <Text variant="large-semibold">Require human approval</Text>
-                  <Text variant="large" className="mt-1 text-zinc-900">
-                    The agent will pause and wait for your review before
-                    continuing
-                  </Text>
+          {shouldShowToggle ? (
+            <>
+              {showHITLToggle && (
+                <div className="flex w-full max-w-2xl flex-col items-start gap-4 rounded-xl border border-zinc-100 bg-white p-6">
+                  <div className="flex w-full items-start justify-between gap-4">
+                    <div className="flex-1">
+                      <Text variant="large-semibold">
+                        Human-in-the-loop approval
+                      </Text>
+                      <Text variant="large" className="mt-1 text-zinc-900">
+                        The agent will pause at human-in-the-loop blocks and
+                        wait for your review before continuing
+                      </Text>
+                    </div>
+                    <Switch
+                      checked={currentHITLSafeMode || false}
+                      onCheckedChange={handleHITLToggle}
+                      disabled={isPending}
+                      className="mt-1"
+                    />
+                  </div>
                 </div>
-                <Switch
-                  checked={currentSafeMode || false}
-                  onCheckedChange={handleToggle}
-                  disabled={isPending}
-                  className="mt-1"
-                />
-              </div>
-            </div>
+              )}
+              {showSensitiveActionToggle && (
+                <div className="flex w-full max-w-2xl flex-col items-start gap-4 rounded-xl border border-zinc-100 bg-white p-6">
+                  <div className="flex w-full items-start justify-between gap-4">
+                    <div className="flex-1">
+                      <Text variant="large-semibold">
+                        Sensitive action approval
+                      </Text>
+                      <Text variant="large" className="mt-1 text-zinc-900">
+                        The agent will pause at sensitive action blocks and wait
+                        for your review before continuing
+                      </Text>
+                    </div>
+                    <Switch
+                      checked={currentSensitiveActionSafeMode}
+                      onCheckedChange={handleSensitiveActionToggle}
+                      disabled={isPending}
+                      className="mt-1"
+                    />
+                  </div>
+                </div>
+              )}
+            </>
           ) : (
             <div className="rounded-xl border border-zinc-100 bg-white p-6">
               <Text variant="body" className="text-muted-foreground">

autogpt_platform/frontend/src/app/(platform)/library/agents/[id]/components/OldAgentLibraryView/components/agent-run-draft-view.tsx

@@ -1,8 +1,15 @@
 "use client";
-import React, { useCallback, useEffect, useMemo, useState } from "react";
+import React, {
+  useCallback,
+  useContext,
+  useEffect,
+  useMemo,
+  useState,
+} from "react";
 
 import {
   CredentialsMetaInput,
+  CredentialsType,
   GraphExecutionID,
   GraphMeta,
   LibraryAgentPreset,
@@ -29,7 +36,11 @@ import {
 } from "@/components/__legacy__/ui/icons";
 import { Input } from "@/components/__legacy__/ui/input";
 import { Button } from "@/components/atoms/Button/Button";
-import { CredentialsInput } from "@/components/contextual/CredentialsInput/CredentialsInput";
+import { CredentialsGroupedView } from "@/components/contextual/CredentialsInput/components/CredentialsGroupedView/CredentialsGroupedView";
+import {
+  findSavedCredentialByProviderAndType,
+  findSavedUserCredentialByProviderAndType,
+} from "@/components/contextual/CredentialsInput/components/CredentialsGroupedView/helpers";
 import { InformationTooltip } from "@/components/molecules/InformationTooltip/InformationTooltip";
 import {
   useToast,
@@ -37,6 +48,7 @@ import {
 } from "@/components/molecules/Toast/use-toast";
 import { humanizeCronExpression } from "@/lib/cron-expression-utils";
 import { cn, isEmpty } from "@/lib/utils";
+import { CredentialsProvidersContext } from "@/providers/agent-credentials/credentials-provider";
 import { ClockIcon, CopyIcon, InfoIcon } from "@phosphor-icons/react";
 import { CalendarClockIcon, Trash2Icon } from "lucide-react";
 
@@ -90,6 +102,7 @@ export function AgentRunDraftView({
   const api = useBackendAPI();
   const { toast } = useToast();
   const toastOnFail = useToastOnFail();
+  const allProviders = useContext(CredentialsProvidersContext);
 
   const [inputValues, setInputValues] = useState<Record<string, any>>({});
   const [inputCredentials, setInputCredentials] = useState<
@@ -128,6 +141,77 @@ export function AgentRunDraftView({
     () => graph.credentials_input_schema.properties,
     [graph],
   );
+  const credentialFields = useMemo(
+    function getCredentialFields() {
+      return Object.entries(agentCredentialsInputFields);
+    },
+    [agentCredentialsInputFields],
+  );
+  const requiredCredentials = useMemo(
+    function getRequiredCredentials() {
+      return new Set(
+        (graph.credentials_input_schema?.required as string[]) || [],
+      );
+    },
+    [graph.credentials_input_schema?.required],
+  );
+
+  useEffect(
+    function initializeDefaultCredentials() {
+      if (!allProviders) return;
+      if (!graph.credentials_input_schema?.properties) return;
+      if (requiredCredentials.size === 0) return;
+
+      setInputCredentials(function updateCredentials(currentCreds) {
+        const next = { ...currentCreds };
+        let didAdd = false;
+
+        for (const key of requiredCredentials) {
+          if (next[key]) continue;
+          const schema = graph.credentials_input_schema.properties[key];
+          if (!schema) continue;
+
+          const providerNames = schema.credentials_provider || [];
+          const credentialTypes = schema.credentials_types || [];
+          const requiredScopes = schema.credentials_scopes;
+
+          const userCredential = findSavedUserCredentialByProviderAndType(
+            providerNames,
+            credentialTypes,
+            requiredScopes,
+            allProviders,
+          );
+
+          const savedCredential =
+            userCredential ||
+            findSavedCredentialByProviderAndType(
+              providerNames,
+              credentialTypes,
+              requiredScopes,
+              allProviders,
+            );
+
+          if (!savedCredential) continue;
+
+          next[key] = {
+            id: savedCredential.id,
+            provider: savedCredential.provider,
+            type: savedCredential.type as CredentialsType,
+            title: savedCredential.title,
+          };
+          didAdd = true;
+        }
+
+        if (!didAdd) return currentCreds;
+        return next;
+      });
+    },
+    [
+      allProviders,
+      graph.credentials_input_schema?.properties,
+      requiredCredentials,
+    ],
+  );
 
   const [allRequiredInputsAreSet, missingInputs] = useMemo(() => {
     const nonEmptyInputs = new Set(
@@ -145,18 +229,35 @@ export function AgentRunDraftView({
     );
     return [isSuperset, difference];
   }, [agentInputSchema.required, inputValues]);
-  const [allCredentialsAreSet, missingCredentials] = useMemo(() => {
-    const availableCredentials = new Set(Object.keys(inputCredentials));
-    const allCredentials = new Set(Object.keys(agentCredentialsInputFields));
-    // Backwards-compatible implementation of isSupersetOf and difference
-    const isSuperset = Array.from(allCredentials).every((item) =>
-      availableCredentials.has(item),
-    );
-    const difference = Array.from(allCredentials).filter(
-      (item) => !availableCredentials.has(item),
-    );
-    return [isSuperset, difference];
-  }, [agentCredentialsInputFields, inputCredentials]);
+  const [allCredentialsAreSet, missingCredentials] = useMemo(
+    function getCredentialStatus() {
+      const missing = Array.from(requiredCredentials).filter((key) => {
+        const cred = inputCredentials[key];
+        return !cred || !cred.id;
+      });
+      return [missing.length === 0, missing];
+    },
+    [requiredCredentials, inputCredentials],
+  );
+  function addChangedCredentials(prev: Set<keyof LibraryAgentPresetUpdatable>) {
+    const next = new Set(prev);
+    next.add("credentials");
+    return next;
+  }
+
+  function handleCredentialChange(key: string, value?: CredentialsMetaInput) {
+    setInputCredentials(function updateInputCredentials(currentCreds) {
+      const next = { ...currentCreds };
+      if (value === undefined) {
+        delete next[key];
+        return next;
+      }
+      next[key] = value;
+      return next;
+    });
+    setChangedPresetAttributes(addChangedCredentials);
+  }
+
   const notifyMissingInputs = useCallback(
     (needPresetName: boolean = true) => {
       const allMissingFields = (
@@ -649,35 +750,6 @@ export function AgentRunDraftView({
               </>
             )}
 
-            {/* Credentials inputs */}
-            {Object.entries(agentCredentialsInputFields).map(
-              ([key, inputSubSchema]) => (
-                <CredentialsInput
-                  key={key}
-                  schema={{ ...inputSubSchema, discriminator: undefined }}
-                  selectedCredentials={
-                    inputCredentials[key] ?? inputSubSchema.default
-                  }
-                  onSelectCredentials={(value) => {
-                    setInputCredentials((obj) => {
-                      const newObj = { ...obj };
-                      if (value === undefined) {
-                        delete newObj[key];
-                        return newObj;
-                      }
-                      return {
-                        ...obj,
-                        [key]: value,
-                      };
-                    });
-                    setChangedPresetAttributes((prev) =>
-                      prev.add("credentials"),
-                    );
-                  }}
-                />
-              ),
-            )}
-
             {/* Regular inputs */}
             {Object.entries(agentInputFields).map(([key, inputSubSchema]) => (
               <RunAgentInputs
@@ -695,6 +767,17 @@ export function AgentRunDraftView({
                 data-testid={`agent-input-${key}`}
               />
             ))}
+
+            {/* Credentials inputs */}
+            {credentialFields.length > 0 && (
+              <CredentialsGroupedView
+                credentialFields={credentialFields}
+                requiredCredentials={requiredCredentials}
+                inputCredentials={inputCredentials}
+                inputValues={inputValues}
+                onCredentialChange={handleCredentialChange}
+              />
+            )}
           </CardContent>
         </Card>
       </div>

autogpt_platform/frontend/src/app/(platform)/library/components/LibraryUploadAgentDialog/LibraryUploadAgentDialog.tsx

@@ -2,6 +2,7 @@
 import { Button } from "@/components/atoms/Button/Button";
 import { FileInput } from "@/components/atoms/FileInput/FileInput";
 import { Input } from "@/components/atoms/Input/Input";
+import { LoadingSpinner } from "@/components/atoms/LoadingSpinner/LoadingSpinner";
 import { Dialog } from "@/components/molecules/Dialog/Dialog";
 import {
   Form,
@@ -120,7 +121,7 @@ export default function LibraryUploadAgentDialog() {
           >
             {isUploading ? (
               <div className="flex items-center gap-2">
-                <div className="h-4 w-4 animate-spin rounded-full border-b-2 border-t-2 border-white"></div>
+                <LoadingSpinner size="small" className="text-white" />
                 <span>Uploading...</span>
               </div>
             ) : (

autogpt_platform/frontend/src/app/api/openapi.json

@@ -6383,6 +6383,11 @@
             "title": "Has Human In The Loop",
             "readOnly": true
           },
+          "has_sensitive_action": {
+            "type": "boolean",
+            "title": "Has Sensitive Action",
+            "readOnly": true
+          },
           "trigger_setup_info": {
             "anyOf": [
               { "$ref": "#/components/schemas/GraphTriggerInfo" },
@@ -6399,6 +6404,7 @@
           "output_schema",
           "has_external_trigger",
           "has_human_in_the_loop",
+          "has_sensitive_action",
           "trigger_setup_info"
         ],
         "title": "BaseGraph"
@@ -7629,6 +7635,11 @@
             "title": "Has Human In The Loop",
             "readOnly": true
           },
+          "has_sensitive_action": {
+            "type": "boolean",
+            "title": "Has Sensitive Action",
+            "readOnly": true
+          },
           "trigger_setup_info": {
             "anyOf": [
               { "$ref": "#/components/schemas/GraphTriggerInfo" },
@@ -7652,6 +7663,7 @@
           "output_schema",
           "has_external_trigger",
           "has_human_in_the_loop",
+          "has_sensitive_action",
           "trigger_setup_info",
           "credentials_input_schema"
         ],
@@ -7730,6 +7742,11 @@
             "title": "Has Human In The Loop",
             "readOnly": true
           },
+          "has_sensitive_action": {
+            "type": "boolean",
+            "title": "Has Sensitive Action",
+            "readOnly": true
+          },
           "trigger_setup_info": {
             "anyOf": [
               { "$ref": "#/components/schemas/GraphTriggerInfo" },
@@ -7754,6 +7771,7 @@
           "output_schema",
           "has_external_trigger",
           "has_human_in_the_loop",
+          "has_sensitive_action",
           "trigger_setup_info",
           "credentials_input_schema"
         ],
@@ -7762,8 +7780,14 @@
       "GraphSettings": {
         "properties": {
           "human_in_the_loop_safe_mode": {
-            "anyOf": [{ "type": "boolean" }, { "type": "null" }],
-            "title": "Human In The Loop Safe Mode"
+            "type": "boolean",
+            "title": "Human In The Loop Safe Mode",
+            "default": true
+          },
+          "sensitive_action_safe_mode": {
+            "type": "boolean",
+            "title": "Sensitive Action Safe Mode",
+            "default": false
           }
         },
         "type": "object",
@@ -7921,6 +7945,16 @@
             "title": "Has External Trigger",
             "description": "Whether the agent has an external trigger (e.g. webhook) node"
           },
+          "has_human_in_the_loop": {
+            "type": "boolean",
+            "title": "Has Human In The Loop",
+            "description": "Whether the agent has human-in-the-loop blocks"
+          },
+          "has_sensitive_action": {
+            "type": "boolean",
+            "title": "Has Sensitive Action",
+            "description": "Whether the agent has sensitive action blocks"
+          },
           "trigger_setup_info": {
             "anyOf": [
               { "$ref": "#/components/schemas/GraphTriggerInfo" },
@@ -7967,6 +8001,8 @@
           "output_schema",
           "credentials_input_schema",
           "has_external_trigger",
+          "has_human_in_the_loop",
+          "has_sensitive_action",
           "new_output",
           "can_access_graph",
           "is_latest_version",
@@ -9375,6 +9411,12 @@
             ],
             "title": "Reviewed Data",
             "description": "Optional edited data (ignored if approved=False)"
+          },
+          "auto_approve_future": {
+            "type": "boolean",
+            "title": "Auto Approve Future",
+            "description": "If true and this review is approved, future executions of this same block (node) will be automatically approved. This only affects approved reviews.",
+            "default": false
           }
         },
         "type": "object",
@@ -9394,7 +9436,7 @@
         "type": "object",
         "required": ["reviews"],
         "title": "ReviewRequest",
-        "description": "Request model for processing ALL pending reviews for an execution.\n\nThis request must include ALL pending reviews for a graph execution.\nEach review will be either approved (with optional data modifications)\nor rejected (data ignored). The execution will resume only after ALL reviews are processed."
+        "description": "Request model for processing ALL pending reviews for an execution.\n\nThis request must include ALL pending reviews for a graph execution.\nEach review will be either approved (with optional data modifications)\nor rejected (data ignored). The execution will resume only after ALL reviews are processed.\n\nEach review item can individually specify whether to auto-approve future executions\nof the same block via the `auto_approve_future` field on ReviewItem."
       },
       "ReviewResponse": {
         "properties": {

autogpt_platform/frontend/src/components/contextual/CredentialsInput/components/CredentialsGroupedView/CredentialsGroupedView.tsx

@@ -5,30 +5,37 @@ import {
   AccordionItem,
   AccordionTrigger,
 } from "@/components/molecules/Accordion/Accordion";
+import {
+  CredentialsMetaInput,
+  CredentialsType,
+} from "@/lib/autogpt-server-api/types";
 import { CredentialsProvidersContext } from "@/providers/agent-credentials/credentials-provider";
-import { SlidersHorizontal } from "@phosphor-icons/react";
+import { SlidersHorizontalIcon } from "@phosphor-icons/react";
 import { useContext, useEffect, useMemo, useRef } from "react";
-import { useRunAgentModalContext } from "../../context";
 import {
   areSystemCredentialProvidersLoading,
   CredentialField,
   findSavedCredentialByProviderAndType,
   hasMissingRequiredSystemCredentials,
   splitCredentialFieldsBySystem,
-} from "../helpers";
+} from "./helpers";
 
 type Props = {
   credentialFields: CredentialField[];
   requiredCredentials: Set<string>;
+  inputCredentials: Record<string, CredentialsMetaInput | undefined>;
+  inputValues: Record<string, any>;
+  onCredentialChange: (key: string, value?: CredentialsMetaInput) => void;
 };
 
 export function CredentialsGroupedView({
   credentialFields,
   requiredCredentials,
+  inputCredentials,
+  inputValues,
+  onCredentialChange,
 }: Props) {
   const allProviders = useContext(CredentialsProvidersContext);
-  const { inputCredentials, setInputCredentialsValue, inputValues } =
-    useRunAgentModalContext();
 
   const { userCredentialFields, systemCredentialFields } = useMemo(
     () =>
@@ -87,11 +94,11 @@ export function CredentialsGroupedView({
       );
 
       if (savedCredential) {
-        setInputCredentialsValue(key, {
+        onCredentialChange(key, {
           id: savedCredential.id,
           provider: savedCredential.provider,
-          type: savedCredential.type,
-          title: (savedCredential as { title?: string }).title,
+          type: savedCredential.type as CredentialsType,
+          title: savedCredential.title,
         });
       }
     }
@@ -103,7 +110,7 @@ export function CredentialsGroupedView({
     systemCredentialFields,
     requiredCredentials,
     inputCredentials,
-    setInputCredentialsValue,
+    onCredentialChange,
     isLoadingProviders,
   ]);
 
@@ -123,7 +130,7 @@ export function CredentialsGroupedView({
                   }
                   selectedCredentials={selectedCred}
                   onSelectCredentials={(value) => {
-                    setInputCredentialsValue(key, value);
+                    onCredentialChange(key, value);
                   }}
                   siblingInputs={inputValues}
                   isOptional={!requiredCredentials.has(key)}
@@ -143,7 +150,8 @@ export function CredentialsGroupedView({
           <AccordionItem value="system-credentials" className="border-none">
             <AccordionTrigger className="py-2 text-sm text-muted-foreground hover:no-underline">
               <div className="flex items-center gap-1">
-                <SlidersHorizontal size={16} weight="bold" /> System credentials
+                <SlidersHorizontalIcon size={16} weight="bold" /> System
+                credentials
                 {hasMissingSystemCredentials && (
                   <span className="text-destructive">(missing)</span>
                 )}
@@ -163,7 +171,7 @@ export function CredentialsGroupedView({
                         }
                         selectedCredentials={selectedCred}
                         onSelectCredentials={(value) => {
-                          setInputCredentialsValue(key, value);
+                          onCredentialChange(key, value);
                         }}
                         siblingInputs={inputValues}
                         isOptional={!requiredCredentials.has(key)}

autogpt_platform/frontend/src/components/contextual/CredentialsInput/components/CredentialsGroupedView/helpers.ts

@@ -1,5 +1,5 @@
 import { CredentialsProvidersContextType } from "@/providers/agent-credentials/credentials-provider";
-import { getSystemCredentials } from "../../../../../../../../../../../components/contextual/CredentialsInput/helpers";
+import { filterSystemCredentials, getSystemCredentials } from "../../helpers";
 
 export type CredentialField = [string, any];
 
@@ -208,3 +208,42 @@ export function findSavedCredentialByProviderAndType(
 
   return undefined;
 }
+
+export function findSavedUserCredentialByProviderAndType(
+  providerNames: string[],
+  credentialTypes: string[],
+  requiredScopes: string[] | undefined,
+  allProviders: CredentialsProvidersContextType | null,
+): SavedCredential | undefined {
+  for (const providerName of providerNames) {
+    const providerData = allProviders?.[providerName];
+    if (!providerData) continue;
+
+    const userCredentials = filterSystemCredentials(
+      providerData.savedCredentials ?? [],
+    );
+
+    const matchingCredentials: SavedCredential[] = [];
+
+    for (const credential of userCredentials) {
+      const typeMatches =
+        credentialTypes.length === 0 ||
+        credentialTypes.includes(credential.type);
+      const scopesMatch = hasRequiredScopes(credential, requiredScopes);
+
+      if (!typeMatches) continue;
+      if (!scopesMatch) continue;
+
+      matchingCredentials.push(credential as SavedCredential);
+    }
+
+    if (matchingCredentials.length === 1) {
+      return matchingCredentials[0];
+    }
+    if (matchingCredentials.length > 1) {
+      return undefined;
+    }
+  }
+
+  return undefined;
+}

autogpt_platform/frontend/src/components/contextual/CredentialsInput/useCredentialsInput.ts

@@ -98,24 +98,20 @@ export function useCredentialsInput({
 
   // Auto-select the first available credential on initial mount
   // Once a user has made a selection, we don't override it
-  useEffect(() => {
-    if (readOnly) return;
-    if (!credentials || !("savedCredentials" in credentials)) return;
+  useEffect(
+    function autoSelectCredential() {
+      if (readOnly) return;
+      if (!credentials || !("savedCredentials" in credentials)) return;
+      if (selectedCredential?.id) return;
 
-    // If already selected, don't auto-select
-    if (selectedCredential?.id) return;
+      const savedCreds = credentials.savedCredentials;
+      if (savedCreds.length === 0) return;
 
-    // Only attempt auto-selection once
-    if (hasAttemptedAutoSelect.current) return;
-    hasAttemptedAutoSelect.current = true;
+      if (hasAttemptedAutoSelect.current) return;
+      hasAttemptedAutoSelect.current = true;
 
-    // If optional, don't auto-select (user can choose "None")
-    if (isOptional) return;
+      if (isOptional) return;
 
-    const savedCreds = credentials.savedCredentials;
-
-    // Auto-select the first credential if any are available
-    if (savedCreds.length > 0) {
       const cred = savedCreds[0];
       onSelectCredential({
         id: cred.id,
@@ -123,14 +119,15 @@ export function useCredentialsInput({
         provider: credentials.provider,
         title: (cred as any).title,
       });
-    }
-  }, [
-    credentials,
-    selectedCredential?.id,
-    readOnly,
-    isOptional,
-    onSelectCredential,
-  ]);
+    },
+    [
+      credentials,
+      selectedCredential?.id,
+      readOnly,
+      isOptional,
+      onSelectCredential,
+    ],
+  );
 
   if (
     !credentials ||

autogpt_platform/frontend/src/components/molecules/Popover/Popover.tsx

@@ -0,0 +1,33 @@
+"use client";
+
+import * as PopoverPrimitive from "@radix-ui/react-popover";
+import * as React from "react";
+
+import { cn } from "@/lib/utils";
+
+const Popover = PopoverPrimitive.Root;
+
+const PopoverTrigger = PopoverPrimitive.Trigger;
+
+const PopoverAnchor = PopoverPrimitive.Anchor;
+
+const PopoverContent = React.forwardRef<
+  React.ElementRef<typeof PopoverPrimitive.Content>,
+  React.ComponentPropsWithoutRef<typeof PopoverPrimitive.Content>
+>(({ className, align = "center", sideOffset = 4, ...props }, ref) => (
+  <PopoverPrimitive.Portal>
+    <PopoverPrimitive.Content
+      ref={ref}
+      align={align}
+      sideOffset={sideOffset}
+      className={cn(
+        "z-50 w-72 rounded-lg border border-zinc-200 bg-white p-4 text-zinc-900 shadow-md outline-none data-[state=open]:animate-in data-[state=closed]:animate-out data-[state=closed]:fade-out-0 data-[state=open]:fade-in-0 data-[state=closed]:zoom-out-95 data-[state=open]:zoom-in-95 data-[side=bottom]:slide-in-from-top-2 data-[side=left]:slide-in-from-right-2 data-[side=right]:slide-in-from-left-2 data-[side=top]:slide-in-from-bottom-2",
+        className,
+      )}
+      {...props}
+    />
+  </PopoverPrimitive.Portal>
+));
+PopoverContent.displayName = PopoverPrimitive.Content.displayName;
+
+export { Popover, PopoverAnchor, PopoverContent, PopoverTrigger };

autogpt_platform/frontend/src/components/organisms/FloatingReviewsPanel/FloatingReviewsPanel.tsx

@@ -31,6 +31,29 @@ export function FloatingReviewsPanel({
       query: {
         enabled: !!(graphId && executionId),
         select: okData,
+        // Poll while execution is in progress to detect status changes
+        refetchInterval: (q) => {
+          // Note: refetchInterval callback receives raw data before select transform
+          const rawData = q.state.data as
+            | { status: number; data?: { status?: string } }
+            | undefined;
+          if (rawData?.status !== 200) return false;
+
+          const status = rawData?.data?.status;
+          if (!status) return false;
+
+          // Poll every 2 seconds while running or in review
+          if (
+            status === AgentExecutionStatus.RUNNING ||
+            status === AgentExecutionStatus.QUEUED ||
+            status === AgentExecutionStatus.INCOMPLETE ||
+            status === AgentExecutionStatus.REVIEW
+          ) {
+            return 2000;
+          }
+          return false;
+        },
+        refetchIntervalInBackground: true,
       },
     },
   );
@@ -40,28 +63,47 @@ export function FloatingReviewsPanel({
     useShallow((state) => state.graphExecutionStatus),
   );
 
+  // Determine if we should poll for pending reviews
+  const isInReviewStatus =
+    executionDetails?.status === AgentExecutionStatus.REVIEW ||
+    graphExecutionStatus === AgentExecutionStatus.REVIEW;
+
   const { pendingReviews, isLoading, refetch } = usePendingReviewsForExecution(
     executionId || "",
+    {
+      enabled: !!executionId,
+      // Poll every 2 seconds when in REVIEW status to catch new reviews
+      refetchInterval: isInReviewStatus ? 2000 : false,
+    },
   );
 
+  // Refetch pending reviews when execution status changes
   useEffect(() => {
-    if (executionId) {
+    if (executionId && executionDetails?.status) {
       refetch();
     }
   }, [executionDetails?.status, executionId, refetch]);
 
-  // Refetch when graph execution status changes to REVIEW
-  useEffect(() => {
-    if (graphExecutionStatus === AgentExecutionStatus.REVIEW && executionId) {
-      refetch();
-    }
-  }, [graphExecutionStatus, executionId, refetch]);
+  // Hide panel if:
+  // 1. No execution ID
+  // 2. No pending reviews and not in REVIEW status
+  // 3. Execution is RUNNING or QUEUED (hasn't paused for review yet)
+  if (!executionId) {
+    return null;
+  }
 
   if (
-    !executionId ||
-    (!isLoading &&
-      pendingReviews.length === 0 &&
-      executionDetails?.status !== AgentExecutionStatus.REVIEW)
+    !isLoading &&
+    pendingReviews.length === 0 &&
+    executionDetails?.status !== AgentExecutionStatus.REVIEW
+  ) {
+    return null;
+  }
+
+  // Don't show panel while execution is still running/queued (not paused for review)
+  if (
+    executionDetails?.status === AgentExecutionStatus.RUNNING ||
+    executionDetails?.status === AgentExecutionStatus.QUEUED
   ) {
     return null;
   }

autogpt_platform/frontend/src/components/organisms/PendingReviewCard/PendingReviewCard.tsx

@@ -1,9 +1,7 @@
 import { PendingHumanReviewModel } from "@/app/api/__generated__/models/pendingHumanReviewModel";
 import { Text } from "@/components/atoms/Text/Text";
-import { Button } from "@/components/atoms/Button/Button";
 import { Input } from "@/components/atoms/Input/Input";
 import { Switch } from "@/components/atoms/Switch/Switch";
-import { TrashIcon, EyeSlashIcon } from "@phosphor-icons/react";
 import { useState } from "react";
 
 interface StructuredReviewPayload {
@@ -40,19 +38,15 @@ function extractReviewData(payload: unknown): {
 interface PendingReviewCardProps {
   review: PendingHumanReviewModel;
   onReviewDataChange: (nodeExecId: string, data: string) => void;
-  reviewMessage?: string;
-  onReviewMessageChange?: (nodeExecId: string, message: string) => void;
-  isDisabled?: boolean;
-  onToggleDisabled?: (nodeExecId: string) => void;
+  autoApproveFuture?: boolean;
+  onAutoApproveFutureChange?: (nodeExecId: string, enabled: boolean) => void;
 }
 
 export function PendingReviewCard({
   review,
   onReviewDataChange,
-  reviewMessage = "",
-  onReviewMessageChange,
-  isDisabled = false,
-  onToggleDisabled,
+  autoApproveFuture = false,
+  onAutoApproveFutureChange,
 }: PendingReviewCardProps) {
   const extractedData = extractReviewData(review.payload);
   const isDataEditable = review.editable;
@@ -64,13 +58,6 @@ export function PendingReviewCard({
     onReviewDataChange(review.node_exec_id, JSON.stringify(newValue, null, 2));
   };
 
-  const handleMessageChange = (newMessage: string) => {
-    onReviewMessageChange?.(review.node_exec_id, newMessage);
-  };
-
-  // Show simplified view when no toggle functionality is provided (Screenshot 1 mode)
-  const showSimplified = !onToggleDisabled;
-
   const renderDataInput = () => {
     const data = currentData;
 
@@ -147,35 +134,13 @@ export function PendingReviewCard({
   // Use the existing HITL review interface
   return (
     <div className="space-y-4">
-      {!showSimplified && (
-        <div className="flex items-start justify-between">
-          <div className="flex-1">
-            {isDisabled && (
-              <Text variant="small" className="text-muted-foreground">
-                This item will be rejected
-              </Text>
-            )}
-          </div>
-          <Button
-            onClick={() => onToggleDisabled!(review.node_exec_id)}
-            variant={isDisabled ? "primary" : "secondary"}
-            size="small"
-            leftIcon={
-              isDisabled ? <EyeSlashIcon size={14} /> : <TrashIcon size={14} />
-            }
-          >
-            {isDisabled ? "Include" : "Exclude"}
-          </Button>
-        </div>
-      )}
-
       {/* Show instructions as field label */}
       {instructions && (
         <div className="space-y-3">
           <Text variant="body" className="font-semibold text-gray-900">
             {getFieldLabel(instructions)}
           </Text>
-          {isDataEditable && !isDisabled ? (
+          {isDataEditable && !autoApproveFuture ? (
             renderDataInput()
           ) : (
             <div className="rounded-lg border border-gray-200 bg-white p-3">
@@ -198,7 +163,7 @@ export function PendingReviewCard({
               </span>
             )}
           </Text>
-          {isDataEditable && !isDisabled ? (
+          {isDataEditable && !autoApproveFuture ? (
             renderDataInput()
           ) : (
             <div className="rounded-lg border border-gray-200 bg-white p-3">
@@ -210,22 +175,26 @@ export function PendingReviewCard({
         </div>
       )}
 
-      {!showSimplified && isDisabled && (
-        <div>
-          <Text variant="body" className="mb-2 font-semibold">
-            Rejection Reason (Optional):
-          </Text>
-          <Input
-            id="rejection-reason"
-            label="Rejection Reason"
-            hideLabel
-            size="small"
-            type="textarea"
-            rows={3}
-            value={reviewMessage}
-            onChange={(e) => handleMessageChange(e.target.value)}
-            placeholder="Add any notes about why you're rejecting this..."
-          />
+      {/* Auto-approve toggle for this review */}
+      {onAutoApproveFutureChange && (
+        <div className="space-y-2 pt-2">
+          <div className="flex items-center gap-3">
+            <Switch
+              checked={autoApproveFuture}
+              onCheckedChange={(enabled: boolean) =>
+                onAutoApproveFutureChange(review.node_exec_id, enabled)
+              }
+            />
+            <Text variant="small" className="text-gray-700">
+              Auto-approve future executions of this block
+            </Text>
+          </div>
+          {autoApproveFuture && (
+            <Text variant="small" className="pl-11 text-gray-500">
+              Original data will be used for this and all future reviews from
+              this block.
+            </Text>
+          )}
         </div>
       )}
     </div>

autogpt_platform/frontend/src/components/organisms/PendingReviewsList/PendingReviewsList.tsx

@@ -32,14 +32,15 @@ export function PendingReviewsList({
     },
   );
 
-  const [reviewMessageMap, setReviewMessageMap] = useState<
-    Record<string, string>
-  >({});
-
   const [pendingAction, setPendingAction] = useState<
     "approve" | "reject" | null
   >(null);
 
+  // Track per-review auto-approval state
+  const [autoApproveFutureMap, setAutoApproveFutureMap] = useState<
+    Record<string, boolean>
+  >({});
+
   const { toast } = useToast();
 
   const reviewActionMutation = usePostV2ProcessReviewAction({
@@ -88,8 +89,23 @@ export function PendingReviewsList({
     setReviewDataMap((prev) => ({ ...prev, [nodeExecId]: data }));
   }
 
-  function handleReviewMessageChange(nodeExecId: string, message: string) {
-    setReviewMessageMap((prev) => ({ ...prev, [nodeExecId]: message }));
+  // Handle per-review auto-approval toggle
+  function handleAutoApproveFutureToggle(nodeExecId: string, enabled: boolean) {
+    setAutoApproveFutureMap((prev) => ({
+      ...prev,
+      [nodeExecId]: enabled,
+    }));
+
+    if (enabled) {
+      // Reset this review's data to original value
+      const review = reviews.find((r) => r.node_exec_id === nodeExecId);
+      if (review) {
+        setReviewDataMap((prev) => ({
+          ...prev,
+          [nodeExecId]: JSON.stringify(review.payload, null, 2),
+        }));
+      }
+    }
   }
 
   function processReviews(approved: boolean) {
@@ -107,30 +123,39 @@ export function PendingReviewsList({
 
     for (const review of reviews) {
       const reviewData = reviewDataMap[review.node_exec_id];
-      const reviewMessage = reviewMessageMap[review.node_exec_id];
+      const autoApproveThisReview = autoApproveFutureMap[review.node_exec_id];
 
-      let parsedData: any = review.payload; // Default to original payload
+      // When auto-approving future actions for this review, send undefined (use original data)
+      // Otherwise, parse and send the edited data if available
+      let parsedData: any = undefined;
 
-      // Parse edited data if available and editable
-      if (review.editable && reviewData) {
-        try {
-          parsedData = JSON.parse(reviewData);
-        } catch (error) {
-          toast({
-            title: "Invalid JSON",
-            description: `Please fix the JSON format in review for node ${review.node_exec_id}: ${error instanceof Error ? error.message : "Invalid syntax"}`,
-            variant: "destructive",
-          });
-          setPendingAction(null);
-          return;
+      if (!autoApproveThisReview) {
+        // For regular approve/reject, use edited data if available
+        if (review.editable && reviewData) {
+          try {
+            parsedData = JSON.parse(reviewData);
+          } catch (error) {
+            toast({
+              title: "Invalid JSON",
+              description: `Please fix the JSON format in review for node ${review.node_exec_id}: ${error instanceof Error ? error.message : "Invalid syntax"}`,
+              variant: "destructive",
+            });
+            setPendingAction(null);
+            return;
+          }
+        } else {
+          // No edits, use original payload
+          parsedData = review.payload;
         }
       }
+      // When autoApproveThisReview is true, parsedData stays undefined
+      // Backend will use the original payload stored in the database
 
       reviewItems.push({
         node_exec_id: review.node_exec_id,
         approved,
         reviewed_data: parsedData,
-        message: reviewMessage || undefined,
+        auto_approve_future: autoApproveThisReview && approved,
       });
     }
 
@@ -182,21 +207,19 @@ export function PendingReviewsList({
       <div className="space-y-7">
         {reviews.map((review) => (
           <PendingReviewCard
-            key={review.node_exec_id}
+            key={`${review.node_exec_id}`}
             review={review}
             onReviewDataChange={handleReviewDataChange}
-            onReviewMessageChange={handleReviewMessageChange}
-            reviewMessage={reviewMessageMap[review.node_exec_id] || ""}
+            autoApproveFuture={
+              autoApproveFutureMap[review.node_exec_id] || false
+            }
+            onAutoApproveFutureChange={handleAutoApproveFutureToggle}
           />
         ))}
       </div>
 
-      <div className="space-y-7">
-        <Text variant="body" className="text-textGrey">
-          Note: Changes you make here apply only to this task
-        </Text>
-
-        <div className="flex gap-2">
+      <div className="space-y-4">
+        <div className="flex flex-wrap gap-2">
           <Button
             onClick={() => processReviews(true)}
             disabled={reviewActionMutation.isPending || reviews.length === 0}
@@ -220,6 +243,11 @@ export function PendingReviewsList({
             Reject
           </Button>
         </div>
+
+        <Text variant="small" className="text-textGrey">
+          You can turn auto-approval on or off anytime in this agent&apos;s
+          settings.
+        </Text>
       </div>
     </div>
   );

autogpt_platform/frontend/src/components/renderers/InputRenderer/custom/CredentialField/components/CredentialFieldTitle.tsx

@@ -35,12 +35,13 @@ export const CredentialFieldTitle = (props: {
     uiOptions,
   );
 
-  const credentialProvider = toDisplayName(
-    getCredentialProviderFromSchema(
-      useNodeStore.getState().getHardCodedValues(nodeId),
-      schema as BlockIOCredentialsSubSchema,
-    ) ?? "",
+  const provider = getCredentialProviderFromSchema(
+    useNodeStore.getState().getHardCodedValues(nodeId),
+    schema as BlockIOCredentialsSubSchema,
   );
+  const credentialProvider = provider
+    ? `${toDisplayName(provider)} credential`
+    : "credential";
 
   const updatedUiSchema = updateUiOption(uiSchema, {
     showHandles: false,