test: verify e2e tests on dev (no-op change)

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

@@ -3,8 +3,7 @@ import logging
 import time
 from asyncio import CancelledError
 from collections.abc import AsyncGenerator
-from dataclasses import dataclass
+from typing import Any
-from typing import Any, cast
 
 import openai
 import orjson
@@ -16,14 +15,7 @@ from openai import (
     PermissionDeniedError,
     RateLimitError,
 )
-from openai.types.chat import (
+from openai.types.chat import ChatCompletionChunk, ChatCompletionToolParam
-    ChatCompletionAssistantMessageParam,
-    ChatCompletionChunk,
-    ChatCompletionMessageParam,
-    ChatCompletionStreamOptionsParam,
-    ChatCompletionSystemMessageParam,
-    ChatCompletionToolParam,
-)
 
 from backend.data.redis_client import get_redis_async
 from backend.data.understanding import (
@@ -31,7 +23,6 @@ from backend.data.understanding import (
     get_business_understanding,
 )
 from backend.util.exceptions import NotFoundError
-from backend.util.prompt import estimate_token_count
 from backend.util.settings import Settings
 
 from . import db as chat_db
@@ -803,201 +794,6 @@ def _is_region_blocked_error(error: Exception) -> bool:
     return "not available in your region" in str(error).lower()
 
 
-# Context window management constants
-TOKEN_THRESHOLD = 120_000
-KEEP_RECENT_MESSAGES = 15
-
-
-@dataclass
-class ContextWindowResult:
-    """Result of context window management."""
-
-    messages: list[dict[str, Any]]
-    token_count: int
-    was_compacted: bool
-    error: str | None = None
-
-
-def _messages_to_dicts(messages: list) -> list[dict[str, Any]]:
-    """Convert message objects to dicts, filtering None values.
-
-    Handles both TypedDict (dict-like) and other message formats.
-    """
-    result = []
-    for msg in messages:
-        if msg is None:
-            continue
-        if isinstance(msg, dict):
-            msg_dict = {k: v for k, v in msg.items() if v is not None}
-        else:
-            msg_dict = dict(msg)
-        result.append(msg_dict)
-    return result
-
-
-async def _manage_context_window(
-    messages: list,
-    model: str,
-    api_key: str | None = None,
-    base_url: str | None = None,
-) -> ContextWindowResult:
-    """
-    Manage context window by summarizing old messages if token count exceeds threshold.
-
-    This function handles context compaction for LLM calls by:
-    1. Counting tokens in the message list
-    2. If over threshold, summarizing old messages while keeping recent ones
-    3. Ensuring tool_call/tool_response pairs stay intact
-    4. Progressively reducing message count if still over limit
-
-    Args:
-        messages: List of messages in OpenAI format (with system prompt if present)
-        model: Model name for token counting
-        api_key: API key for summarization calls
-        base_url: Base URL for summarization calls
-
-    Returns:
-        ContextWindowResult with compacted messages and metadata
-    """
-    if not messages:
-        return ContextWindowResult([], 0, False, "No messages to compact")
-
-    messages_dict = _messages_to_dicts(messages)
-
-    # Normalize model name for token counting (tiktoken only supports OpenAI models)
-    token_count_model = model.split("/")[-1] if "/" in model else model
-    if "claude" in token_count_model.lower() or not any(
-        known in token_count_model.lower()
-        for known in ["gpt", "o1", "chatgpt", "text-"]
-    ):
-        token_count_model = "gpt-4o"
-
-    try:
-        token_count = estimate_token_count(messages_dict, model=token_count_model)
-    except Exception as e:
-        logger.warning(f"Token counting failed: {e}. Using gpt-4o approximation.")
-        token_count_model = "gpt-4o"
-        token_count = estimate_token_count(messages_dict, model=token_count_model)
-
-    if token_count <= TOKEN_THRESHOLD:
-        return ContextWindowResult(messages, token_count, False)
-
-    has_system_prompt = messages[0].get("role") == "system"
-    slice_start = max(0, len(messages_dict) - KEEP_RECENT_MESSAGES)
-    recent_messages = _ensure_tool_pairs_intact(
-        messages_dict[-KEEP_RECENT_MESSAGES:], messages_dict, slice_start
-    )
-
-    # Determine old messages to summarize (explicit bounds to avoid slice edge cases)
-    system_msg = messages[0] if has_system_prompt else None
-    if has_system_prompt:
-        old_messages_dict = (
-            messages_dict[1:-KEEP_RECENT_MESSAGES]
-            if len(messages_dict) > KEEP_RECENT_MESSAGES + 1
-            else []
-        )
-    else:
-        old_messages_dict = (
-            messages_dict[:-KEEP_RECENT_MESSAGES]
-            if len(messages_dict) > KEEP_RECENT_MESSAGES
-            else []
-        )
-
-    # Try to summarize old messages, fall back to truncation on failure
-    summary_msg = None
-    if old_messages_dict:
-        try:
-            summary_text = await _summarize_messages(
-                old_messages_dict, model=model, api_key=api_key, base_url=base_url
-            )
-            summary_msg = ChatCompletionAssistantMessageParam(
-                role="assistant",
-                content=f"[Previous conversation summary — for context only]: {summary_text}",
-            )
-            base = [system_msg, summary_msg] if has_system_prompt else [summary_msg]
-            messages = base + recent_messages
-            logger.info(
-                f"Context summarized: {token_count} tokens, "
-                f"summarized {len(old_messages_dict)} msgs, kept {KEEP_RECENT_MESSAGES}"
-            )
-        except Exception as e:
-            logger.warning(f"Summarization failed, falling back to truncation: {e}")
-            messages = (
-                [system_msg] + recent_messages if has_system_prompt else recent_messages
-            )
-    else:
-        logger.warning(
-            f"Token count {token_count} exceeds threshold but no old messages to summarize"
-        )
-
-    new_token_count = estimate_token_count(
-        _messages_to_dicts(messages), model=token_count_model
-    )
-
-    # Progressive truncation if still over limit
-    if new_token_count > TOKEN_THRESHOLD:
-        logger.warning(
-            f"Still over limit: {new_token_count} tokens. Reducing messages."
-        )
-        base_msgs = (
-            recent_messages
-            if old_messages_dict
-            else (messages_dict[1:] if has_system_prompt else messages_dict)
-        )
-
-        def build_messages(recent: list) -> list:
-            """Build message list with optional system prompt and summary."""
-            prefix = []
-            if has_system_prompt and system_msg:
-                prefix.append(system_msg)
-            if summary_msg:
-                prefix.append(summary_msg)
-            return prefix + recent
-
-        for keep_count in [12, 10, 8, 5, 3, 2, 1, 0]:
-            if keep_count == 0:
-                messages = build_messages([])
-                if not messages:
-                    continue
-            elif len(base_msgs) < keep_count:
-                continue
-            else:
-                reduced = _ensure_tool_pairs_intact(
-                    base_msgs[-keep_count:],
-                    base_msgs,
-                    max(0, len(base_msgs) - keep_count),
-                )
-                messages = build_messages(reduced)
-
-            new_token_count = estimate_token_count(
-                _messages_to_dicts(messages), model=token_count_model
-            )
-            if new_token_count <= TOKEN_THRESHOLD:
-                logger.info(
-                    f"Reduced to {keep_count} messages, {new_token_count} tokens"
-                )
-                break
-        else:
-            logger.error(
-                f"Cannot reduce below threshold. Final: {new_token_count} tokens"
-            )
-            if has_system_prompt and len(messages) > 1:
-                messages = messages[1:]
-                logger.critical("Dropped system prompt as last resort")
-                return ContextWindowResult(
-                    messages, new_token_count, True, "System prompt dropped"
-                )
-            # No system prompt to drop - return error so callers don't proceed with oversized context
-            return ContextWindowResult(
-                messages,
-                new_token_count,
-                True,
-                "Unable to reduce context below token limit",
-            )
-
-    return ContextWindowResult(messages, new_token_count, True)
-
-
 async def _summarize_messages(
     messages: list,
     model: str,
@@ -1226,8 +1022,11 @@ async def _stream_chat_chunks(
 
     logger.info("Starting pure chat stream")
 
+    # Build messages with system prompt prepended
     messages = session.to_openai_messages()
     if system_prompt:
+        from openai.types.chat import ChatCompletionSystemMessageParam
+
         system_message = ChatCompletionSystemMessageParam(
             role="system",
             content=system_prompt,
@@ -1235,38 +1034,314 @@ async def _stream_chat_chunks(
         messages = [system_message] + messages
 
     # Apply context window management
-    context_result = await _manage_context_window(
+    token_count = 0  # Initialize for exception handler
-        messages=messages,
+    try:
-        model=model,
+        from backend.util.prompt import estimate_token_count
-        api_key=config.api_key,
-        base_url=config.base_url,
-    )
 
-    if context_result.error:
+        # Convert to dict for token counting
-        if "System prompt dropped" in context_result.error:
+        # OpenAI message types are TypedDicts, so they're already dict-like
-            # Warning only - continue with reduced context
+        messages_dict = []
-            yield StreamError(
+        for msg in messages:
-                errorText=(
+            # TypedDict objects are already dicts, just filter None values
-                    "Warning: System prompt dropped due to size constraints. "
+            if isinstance(msg, dict):
-                    "Assistant behavior may be affected."
+                msg_dict = {k: v for k, v in msg.items() if v is not None}
-                )
+            else:
+                # Fallback for unexpected types
+                msg_dict = dict(msg)
+            messages_dict.append(msg_dict)
+
+        # Estimate tokens using appropriate tokenizer
+        # Normalize model name for token counting (tiktoken only supports OpenAI models)
+        token_count_model = model
+        if "/" in model:
+            # Strip provider prefix (e.g., "anthropic/claude-opus-4.5" -> "claude-opus-4.5")
+            token_count_model = model.split("/")[-1]
+
+        # For Claude and other non-OpenAI models, approximate with gpt-4o tokenizer
+        # Most modern LLMs have similar tokenization (~1 token per 4 chars)
+        if "claude" in token_count_model.lower() or not any(
+            known in token_count_model.lower()
+            for known in ["gpt", "o1", "chatgpt", "text-"]
+        ):
+            token_count_model = "gpt-4o"
+
+        # Attempt token counting with error handling
+        try:
+            token_count = estimate_token_count(messages_dict, model=token_count_model)
+        except Exception as token_error:
+            # If token counting fails, use gpt-4o as fallback approximation
+            logger.warning(
+                f"Token counting failed for model {token_count_model}: {token_error}. "
+                "Using gpt-4o approximation."
             )
-        else:
+            token_count = estimate_token_count(messages_dict, model="gpt-4o")
-            # Any other error - abort to prevent failed LLM calls
+
+        # If over threshold, summarize old messages
+        if token_count > 120_000:
+            KEEP_RECENT = 15
+
+            # Check if we have a system prompt at the start
+            has_system_prompt = (
+                len(messages) > 0 and messages[0].get("role") == "system"
+            )
+
+            # Always attempt mitigation when over limit, even with few messages
+            if messages:
+                # Split messages based on whether system prompt exists
+                # Calculate start index for the slice
+                slice_start = max(0, len(messages_dict) - KEEP_RECENT)
+                recent_messages = messages_dict[-KEEP_RECENT:]
+
+                # Ensure tool_call/tool_response pairs stay together
+                # This prevents API errors from orphan tool responses
+                recent_messages = _ensure_tool_pairs_intact(
+                    recent_messages, messages_dict, slice_start
+                )
+
+                if has_system_prompt:
+                    # Keep system prompt separate, summarize everything between system and recent
+                    system_msg = messages[0]
+                    old_messages_dict = messages_dict[1:-KEEP_RECENT]
+                else:
+                    # No system prompt, summarize everything except recent
+                    system_msg = None
+                    old_messages_dict = messages_dict[:-KEEP_RECENT]
+
+                # Summarize any non-empty old messages (no minimum threshold)
+                # If we're over the token limit, we need to compress whatever we can
+                if old_messages_dict:
+                    # Summarize old messages using the same model as chat
+                    summary_text = await _summarize_messages(
+                        old_messages_dict,
+                        model=model,
+                        api_key=config.api_key,
+                        base_url=config.base_url,
+                    )
+
+                    # Build new message list
+                    # Use assistant role (not system) to prevent privilege escalation
+                    # of user-influenced content to instruction-level authority
+                    from openai.types.chat import ChatCompletionAssistantMessageParam
+
+                    summary_msg = ChatCompletionAssistantMessageParam(
+                        role="assistant",
+                        content=(
+                            "[Previous conversation summary — for context only]: "
+                            f"{summary_text}"
+                        ),
+                    )
+
+                    # Rebuild messages based on whether we have a system prompt
+                    if has_system_prompt:
+                        # system_prompt + summary + recent_messages
+                        messages = [system_msg, summary_msg] + recent_messages
+                    else:
+                        # summary + recent_messages (no original system prompt)
+                        messages = [summary_msg] + recent_messages
+
+                    logger.info(
+                        f"Context summarized: {token_count} tokens, "
+                        f"summarized {len(old_messages_dict)} old messages, "
+                        f"kept last {KEEP_RECENT} messages"
+                    )
+
+                    # Fallback: If still over limit after summarization, progressively drop recent messages
+                    # This handles edge cases where recent messages are extremely large
+                    new_messages_dict = []
+                    for msg in messages:
+                        if isinstance(msg, dict):
+                            msg_dict = {k: v for k, v in msg.items() if v is not None}
+                        else:
+                            msg_dict = dict(msg)
+                        new_messages_dict.append(msg_dict)
+
+                    new_token_count = estimate_token_count(
+                        new_messages_dict, model=token_count_model
+                    )
+
+                    if new_token_count > 120_000:
+                        # Still over limit - progressively reduce KEEP_RECENT
+                        logger.warning(
+                            f"Still over limit after summarization: {new_token_count} tokens. "
+                            "Reducing number of recent messages kept."
+                        )
+
+                        for keep_count in [12, 10, 8, 5, 3, 2, 1, 0]:
+                            if keep_count == 0:
+                                # Try with just system prompt + summary (no recent messages)
+                                if has_system_prompt:
+                                    messages = [system_msg, summary_msg]
+                                else:
+                                    messages = [summary_msg]
+                                logger.info(
+                                    "Trying with 0 recent messages (system + summary only)"
+                                )
+                            else:
+                                # Slice from ORIGINAL recent_messages to avoid duplicating summary
+                                reduced_recent = (
+                                    recent_messages[-keep_count:]
+                                    if len(recent_messages) >= keep_count
+                                    else recent_messages
+                                )
+                                # Ensure tool pairs stay intact in the reduced slice
+                                reduced_slice_start = max(
+                                    0, len(recent_messages) - keep_count
+                                )
+                                reduced_recent = _ensure_tool_pairs_intact(
+                                    reduced_recent, recent_messages, reduced_slice_start
+                                )
+                                if has_system_prompt:
+                                    messages = [
+                                        system_msg,
+                                        summary_msg,
+                                    ] + reduced_recent
+                                else:
+                                    messages = [summary_msg] + reduced_recent
+
+                            new_messages_dict = []
+                            for msg in messages:
+                                if isinstance(msg, dict):
+                                    msg_dict = {
+                                        k: v for k, v in msg.items() if v is not None
+                                    }
+                                else:
+                                    msg_dict = dict(msg)
+                                new_messages_dict.append(msg_dict)
+
+                            new_token_count = estimate_token_count(
+                                new_messages_dict, model=token_count_model
+                            )
+
+                            if new_token_count <= 120_000:
+                                logger.info(
+                                    f"Reduced to {keep_count} recent messages, "
+                                    f"now {new_token_count} tokens"
+                                )
+                                break
+                        else:
+                            logger.error(
+                                f"Unable to reduce token count below threshold even with 0 messages. "
+                                f"Final count: {new_token_count} tokens"
+                            )
+                            # ABSOLUTE LAST RESORT: Drop system prompt
+                            # This should only happen if summary itself is massive
+                            if has_system_prompt and len(messages) > 1:
+                                messages = messages[1:]  # Drop system prompt
+                                logger.critical(
+                                    "CRITICAL: Dropped system prompt as absolute last resort. "
+                                    "Behavioral consistency may be affected."
+                                )
+                                # Yield error to user
+                                yield StreamError(
+                                    errorText=(
+                                        "Warning: System prompt dropped due to size constraints. "
+                                        "Assistant behavior may be affected."
+                                    )
+                                )
+                else:
+                    # No old messages to summarize - all messages are "recent"
+                    # Apply progressive truncation to reduce token count
+                    logger.warning(
+                        f"Token count {token_count} exceeds threshold but no old messages to summarize. "
+                        f"Applying progressive truncation to recent messages."
+                    )
+
+                    # Create a base list excluding system prompt to avoid duplication
+                    # This is the pool of messages we'll slice from in the loop
+                    # Use messages_dict for type consistency with _ensure_tool_pairs_intact
+                    base_msgs = (
+                        messages_dict[1:] if has_system_prompt else messages_dict
+                    )
+
+                    # Try progressively smaller keep counts
+                    new_token_count = token_count  # Initialize with current count
+                    for keep_count in [12, 10, 8, 5, 3, 2, 1, 0]:
+                        if keep_count == 0:
+                            # Try with just system prompt (no recent messages)
+                            if has_system_prompt:
+                                messages = [system_msg]
+                                logger.info(
+                                    "Trying with 0 recent messages (system prompt only)"
+                                )
+                            else:
+                                # No system prompt and no recent messages = empty messages list
+                                # This is invalid, skip this iteration
+                                continue
+                        else:
+                            if len(base_msgs) < keep_count:
+                                continue  # Skip if we don't have enough messages
+
+                            # Slice from base_msgs to get recent messages (without system prompt)
+                            recent_messages = base_msgs[-keep_count:]
+
+                            # Ensure tool pairs stay intact in the reduced slice
+                            reduced_slice_start = max(0, len(base_msgs) - keep_count)
+                            recent_messages = _ensure_tool_pairs_intact(
+                                recent_messages, base_msgs, reduced_slice_start
+                            )
+
+                            if has_system_prompt:
+                                messages = [system_msg] + recent_messages
+                            else:
+                                messages = recent_messages
+
+                        new_messages_dict = []
+                        for msg in messages:
+                            if msg is None:
+                                continue  # Skip None messages (type safety)
+                            if isinstance(msg, dict):
+                                msg_dict = {
+                                    k: v for k, v in msg.items() if v is not None
+                                }
+                            else:
+                                msg_dict = dict(msg)
+                            new_messages_dict.append(msg_dict)
+
+                        new_token_count = estimate_token_count(
+                            new_messages_dict, model=token_count_model
+                        )
+
+                        if new_token_count <= 120_000:
+                            logger.info(
+                                f"Reduced to {keep_count} recent messages, "
+                                f"now {new_token_count} tokens"
+                            )
+                            break
+                    else:
+                        # Even with 0 messages still over limit
+                        logger.error(
+                            f"Unable to reduce token count below threshold even with 0 messages. "
+                            f"Final count: {new_token_count} tokens. Messages may be extremely large."
+                        )
+                        # ABSOLUTE LAST RESORT: Drop system prompt
+                        if has_system_prompt and len(messages) > 1:
+                            messages = messages[1:]  # Drop system prompt
+                            logger.critical(
+                                "CRITICAL: Dropped system prompt as absolute last resort. "
+                                "Behavioral consistency may be affected."
+                            )
+                            # Yield error to user
+                            yield StreamError(
+                                errorText=(
+                                    "Warning: System prompt dropped due to size constraints. "
+                                    "Assistant behavior may be affected."
+                                )
+                            )
+
+    except Exception as e:
+        logger.error(f"Context summarization failed: {e}", exc_info=True)
+        # If we were over the token limit, yield error to user
+        # Don't silently continue with oversized messages that will fail
+        if token_count > 120_000:
             yield StreamError(
                 errorText=(
-                    f"Context window management failed: {context_result.error}. "
+                    f"Unable to manage context window (token limit exceeded: {token_count} tokens). "
-                    "Please start a new conversation."
+                    "Context summarization failed. Please start a new conversation."
                 )
             )
             yield StreamFinish()
             return
-
+        # Otherwise, continue with original messages (under limit)
-    messages = context_result.messages
-    if context_result.was_compacted:
-        logger.info(
-            f"Context compacted for streaming: {context_result.token_count} tokens"
-        )
 
     # Loop to handle tool calls and continue conversation
     while True:
@@ -1294,6 +1369,14 @@ async def _stream_chat_chunks(
                         :128
                     ]  # OpenRouter limit
 
+                # Create the stream with proper types
+                from typing import cast
+
+                from openai.types.chat import (
+                    ChatCompletionMessageParam,
+                    ChatCompletionStreamOptionsParam,
+                )
+
                 stream = await client.chat.completions.create(
                     model=model,
                     messages=cast(list[ChatCompletionMessageParam], messages),
@@ -1817,36 +1900,17 @@ async def _generate_llm_continuation(
         # Build system prompt
         system_prompt, _ = await _build_system_prompt(user_id)
 
+        # Build messages in OpenAI format
         messages = session.to_openai_messages()
         if system_prompt:
+            from openai.types.chat import ChatCompletionSystemMessageParam
+
             system_message = ChatCompletionSystemMessageParam(
                 role="system",
                 content=system_prompt,
             )
             messages = [system_message] + messages
 
-        # Apply context window management to prevent oversized requests
-        context_result = await _manage_context_window(
-            messages=messages,
-            model=config.model,
-            api_key=config.api_key,
-            base_url=config.base_url,
-        )
-
-        if context_result.error and "System prompt dropped" not in context_result.error:
-            logger.error(
-                f"Context window management failed for session {session_id}: "
-                f"{context_result.error} (tokens={context_result.token_count})"
-            )
-            return
-
-        messages = context_result.messages
-        if context_result.was_compacted:
-            logger.info(
-                f"Context compacted for LLM continuation: "
-                f"{context_result.token_count} tokens"
-            )
-
         # Build extra_body for tracing
         extra_body: dict[str, Any] = {
             "posthogProperties": {
@@ -1859,54 +1923,19 @@ async def _generate_llm_continuation(
         if session_id:
             extra_body["session_id"] = session_id[:128]
 
-        retry_count = 0
+        # Make non-streaming LLM call (no tools - just text response)
-        last_error: Exception | None = None
+        from typing import cast
-        response = None
 
-        while retry_count <= MAX_RETRIES:
+        from openai.types.chat import ChatCompletionMessageParam
-            try:
-                logger.info(
-                    f"Generating LLM continuation for session {session_id}"
-                    f"{f' (retry {retry_count}/{MAX_RETRIES})' if retry_count > 0 else ''}"
-                )
 
-                response = await client.chat.completions.create(
+        # No tools parameter = text-only response (no tool calls)
-                    model=config.model,
+        response = await client.chat.completions.create(
-                    messages=cast(list[ChatCompletionMessageParam], messages),
+            model=config.model,
-                    extra_body=extra_body,
+            messages=cast(list[ChatCompletionMessageParam], messages),
-                )
+            extra_body=extra_body,
-                last_error = None  # Clear any previous error on success
+        )
-                break  # Success, exit retry loop
-            except Exception as e:
-                last_error = e
-                if _is_retryable_error(e) and retry_count < MAX_RETRIES:
-                    retry_count += 1
-                    delay = min(
-                        BASE_DELAY_SECONDS * (2 ** (retry_count - 1)),
-                        MAX_DELAY_SECONDS,
-                    )
-                    logger.warning(
-                        f"Retryable error in LLM continuation: {e!s}. "
-                        f"Retrying in {delay:.1f}s (attempt {retry_count}/{MAX_RETRIES})"
-                    )
-                    await asyncio.sleep(delay)
-                    continue
-                else:
-                    # Non-retryable error - log and exit gracefully
-                    logger.error(
-                        f"Non-retryable error in LLM continuation: {e!s}",
-                        exc_info=True,
-                    )
-                    return
 
-        if last_error:
+        if response.choices and response.choices[0].message.content:
-            logger.error(
-                f"Max retries ({MAX_RETRIES}) exceeded for LLM continuation. "
-                f"Last error: {last_error!s}"
-            )
-            return
-
-        if response and response.choices and response.choices[0].message.content:
             assistant_content = response.choices[0].message.content
 
             # Reload session from DB to avoid race condition with user messages

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

@@ -139,10 +139,11 @@ async def decompose_goal_external(
     """
     client = _get_client()
 
-    if context:
+    # Build the request payload
-        description = f"{description}\n\nAdditional context from user:\n{context}"
-
     payload: dict[str, Any] = {"description": description}
+    if context:
+        # The external service uses user_instruction for additional context
+        payload["user_instruction"] = context
     if library_agents:
         payload["library_agents"] = library_agents
 

autogpt_platform/backend/backend/api/ws_api.py

@@ -66,24 +66,18 @@ async def event_broadcaster(manager: ConnectionManager):
     execution_bus = AsyncRedisExecutionEventBus()
     notification_bus = AsyncRedisNotificationEventBus()
 
-    try:
+    async def execution_worker():
+        async for event in execution_bus.listen("*"):
+            await manager.send_execution_update(event)
 
-        async def execution_worker():
+    async def notification_worker():
-            async for event in execution_bus.listen("*"):
+        async for notification in notification_bus.listen("*"):
-                await manager.send_execution_update(event)
+            await manager.send_notification(
+                user_id=notification.user_id,
+                payload=notification.payload,
+            )
 
-        async def notification_worker():
+    await asyncio.gather(execution_worker(), notification_worker())
-            async for notification in notification_bus.listen("*"):
-                await manager.send_notification(
-                    user_id=notification.user_id,
-                    payload=notification.payload,
-                )
-
-        await asyncio.gather(execution_worker(), notification_worker())
-    finally:
-        # Ensure PubSub connections are closed on any exit to prevent leaks
-        await execution_bus.close()
-        await notification_bus.close()
 
 
 async def authenticate_websocket(websocket: WebSocket) -> str:

autogpt_platform/backend/backend/blocks/encoder_block.py

@@ -1,77 +0,0 @@
-"""Text encoding block for converting special characters to escape sequences."""
-
-import codecs
-
-from backend.data.block import (
-    Block,
-    BlockCategory,
-    BlockOutput,
-    BlockSchemaInput,
-    BlockSchemaOutput,
-)
-from backend.data.model import SchemaField
-
-
-class TextEncoderBlock(Block):
-    """
-    Encodes a string by converting special characters into escape sequences.
-
-    This block is the inverse of TextDecoderBlock. It takes text containing
-    special characters (like newlines, tabs, etc.) and converts them into
-    their escape sequence representations (e.g., newline becomes \\n).
-    """
-
-    class Input(BlockSchemaInput):
-        """Input schema for TextEncoderBlock."""
-
-        text: str = SchemaField(
-            description="A string containing special characters to be encoded",
-            placeholder="Your text with newlines and quotes to encode",
-        )
-
-    class Output(BlockSchemaOutput):
-        """Output schema for TextEncoderBlock."""
-
-        encoded_text: str = SchemaField(
-            description="The encoded text with special characters converted to escape sequences"
-        )
-        error: str = SchemaField(description="Error message if encoding fails")
-
-    def __init__(self):
-        super().__init__(
-            id="5185f32e-4b65-4ecf-8fbb-873f003f09d6",
-            description="Encodes a string by converting special characters into escape sequences",
-            categories={BlockCategory.TEXT},
-            input_schema=TextEncoderBlock.Input,
-            output_schema=TextEncoderBlock.Output,
-            test_input={
-                "text": """Hello
-World!
-This is a "quoted" string."""
-            },
-            test_output=[
-                (
-                    "encoded_text",
-                    """Hello\\nWorld!\\nThis is a "quoted" string.""",
-                )
-            ],
-        )
-
-    async def run(self, input_data: Input, **kwargs) -> BlockOutput:
-        """
-        Encode the input text by converting special characters to escape sequences.
-
-        Args:
-            input_data: The input containing the text to encode.
-            **kwargs: Additional keyword arguments (unused).
-
-        Yields:
-            The encoded text with escape sequences, or an error message if encoding fails.
-        """
-        try:
-            encoded_text = codecs.encode(input_data.text, "unicode_escape").decode(
-                "utf-8"
-            )
-            yield "encoded_text", encoded_text
-        except Exception as e:
-            yield "error", f"Encoding error: {str(e)}"

autogpt_platform/backend/backend/blocks/test/test_text_encoder.py

@@ -1,77 +0,0 @@
-import pytest
-
-from backend.blocks.encoder_block import TextEncoderBlock
-
-
-@pytest.mark.asyncio
-async def test_text_encoder_basic():
-    """Test basic encoding of newlines and special characters."""
-    block = TextEncoderBlock()
-    result = []
-    async for output in block.run(TextEncoderBlock.Input(text="Hello\nWorld")):
-        result.append(output)
-
-    assert len(result) == 1
-    assert result[0][0] == "encoded_text"
-    assert result[0][1] == "Hello\\nWorld"
-
-
-@pytest.mark.asyncio
-async def test_text_encoder_multiple_escapes():
-    """Test encoding of multiple escape sequences."""
-    block = TextEncoderBlock()
-    result = []
-    async for output in block.run(
-        TextEncoderBlock.Input(text="Line1\nLine2\tTabbed\rCarriage")
-    ):
-        result.append(output)
-
-    assert len(result) == 1
-    assert result[0][0] == "encoded_text"
-    assert "\\n" in result[0][1]
-    assert "\\t" in result[0][1]
-    assert "\\r" in result[0][1]
-
-
-@pytest.mark.asyncio
-async def test_text_encoder_unicode():
-    """Test that unicode characters are handled correctly."""
-    block = TextEncoderBlock()
-    result = []
-    async for output in block.run(TextEncoderBlock.Input(text="Hello 世界\n")):
-        result.append(output)
-
-    assert len(result) == 1
-    assert result[0][0] == "encoded_text"
-    # Unicode characters should be escaped as \uXXXX sequences
-    assert "\\n" in result[0][1]
-
-
-@pytest.mark.asyncio
-async def test_text_encoder_empty_string():
-    """Test encoding of an empty string."""
-    block = TextEncoderBlock()
-    result = []
-    async for output in block.run(TextEncoderBlock.Input(text="")):
-        result.append(output)
-
-    assert len(result) == 1
-    assert result[0][0] == "encoded_text"
-    assert result[0][1] == ""
-
-
-@pytest.mark.asyncio
-async def test_text_encoder_error_handling():
-    """Test that encoding errors are handled gracefully."""
-    from unittest.mock import patch
-
-    block = TextEncoderBlock()
-    result = []
-
-    with patch("codecs.encode", side_effect=Exception("Mocked encoding error")):
-        async for output in block.run(TextEncoderBlock.Input(text="test")):
-            result.append(output)
-
-    assert len(result) == 1
-    assert result[0][0] == "error"
-    assert "Mocked encoding error" in result[0][1]

autogpt_platform/backend/backend/data/event_bus.py

@@ -133,23 +133,10 @@ class RedisEventBus(BaseRedisEventBus[M], ABC):
 
 
 class AsyncRedisEventBus(BaseRedisEventBus[M], ABC):
-    def __init__(self):
-        self._pubsub: AsyncPubSub | None = None
-
     @property
     async def connection(self) -> redis.AsyncRedis:
         return await redis.get_redis_async()
 
-    async def close(self) -> None:
-        """Close the PubSub connection if it exists."""
-        if self._pubsub is not None:
-            try:
-                await self._pubsub.close()
-            except Exception:
-                logger.warning("Failed to close PubSub connection", exc_info=True)
-            finally:
-                self._pubsub = None
-
     async def publish_event(self, event: M, channel_key: str):
         """
         Publish an event to Redis. Gracefully handles connection failures
@@ -170,7 +157,6 @@ class AsyncRedisEventBus(BaseRedisEventBus[M], ABC):
             await self.connection, channel_key
         )
         assert isinstance(pubsub, AsyncPubSub)
-        self._pubsub = pubsub
 
         if "*" in channel_key:
             await pubsub.psubscribe(full_channel_name)

autogpt_platform/backend/test/agent_generator/test_service.py

@@ -102,7 +102,7 @@ class TestDecomposeGoalExternal:
 
     @pytest.mark.asyncio
     async def test_decompose_goal_with_context(self):
-        """Test decomposition with additional context enriched into description."""
+        """Test decomposition with additional context."""
         mock_response = MagicMock()
         mock_response.json.return_value = {
             "success": True,
@@ -119,12 +119,9 @@ class TestDecomposeGoalExternal:
                 "Build a chatbot", context="Use Python"
             )
 
-        expected_description = (
-            "Build a chatbot\n\nAdditional context from user:\nUse Python"
-        )
         mock_client.post.assert_called_once_with(
             "/api/decompose-description",
-            json={"description": expected_description},
+            json={"description": "Build a chatbot", "user_instruction": "Use Python"},
         )
 
     @pytest.mark.asyncio

autogpt_platform/frontend/src/tests/signup.spec.ts

@@ -124,3 +124,4 @@ test("user can signup with existing email handling", async ({
     console.error("❌ Duplicate email handling test failed:", error);
   }
 });
+

docs/integrations/README.md

@@ -193,7 +193,6 @@ Below is a comprehensive list of all available blocks, categorized by their prim
 | [Get Current Time](block-integrations/text.md#get-current-time) | This block outputs the current time |
 | [Match Text Pattern](block-integrations/text.md#match-text-pattern) | Matches text against a regex pattern and forwards data to positive or negative output based on the match |
 | [Text Decoder](block-integrations/text.md#text-decoder) | Decodes a string containing escape sequences into actual text |
-| [Text Encoder](block-integrations/text.md#text-encoder) | Encodes a string by converting special characters into escape sequences |
 | [Text Replace](block-integrations/text.md#text-replace) | This block is used to replace a text with a new text |
 | [Text Split](block-integrations/text.md#text-split) | This block is used to split a text into a list of strings |
 | [Word Character Count](block-integrations/text.md#word-character-count) | Counts the number of words and characters in a given text |

docs/integrations/block-integrations/text.md

@@ -380,42 +380,6 @@ This is useful when working with data from APIs or files where escape sequences
 
 ---
 
-## Text Encoder
-
-### What it is
-Encodes a string by converting special characters into escape sequences
-
-### How it works
-<!-- MANUAL: how_it_works -->
-The Text Encoder takes the input string and applies Python's `unicode_escape` encoding (equivalent to `codecs.encode(text, "unicode_escape").decode("utf-8")`) to transform special characters like newlines, tabs, and backslashes into their escaped forms.
-
-The block relies on the input schema to ensure the value is a string; non-string inputs are rejected by validation, and any encoding failures surface as block errors. Non-ASCII characters are emitted as `\uXXXX` sequences, which is useful for ASCII-only payloads.
-<!-- END MANUAL -->
-
-### Inputs
-
-| Input | Description | Type | Required |
-|-------|-------------|------|----------|
-| text | A string containing special characters to be encoded | str | Yes |
-
-### Outputs
-
-| Output | Description | Type |
-|--------|-------------|------|
-| error | Error message if encoding fails | str |
-| encoded_text | The encoded text with special characters converted to escape sequences | str |
-
-### Possible use case
-<!-- MANUAL: use_case -->
-**JSON Payload Preparation**: Encode multiline or quoted text before embedding it in JSON string fields to ensure proper escaping.
-
-**Config/ENV Generation**: Convert template text into escaped strings for `.env` or YAML values that require special character handling.
-
-**Snapshot Fixtures**: Produce stable escaped strings for golden files or API tests where consistent text representation is needed.
-<!-- END MANUAL -->
-
----
-
 ## Text Replace
 
 ### What it is