Record execution step

Send a streaming update in SSE protocol format. Args: event_type: The type of event (e.g., 'start_of_workflow', 'tool_ca

Helper method to log using task_log if available, otherwise skip logging.

Run the complete analysis and return summary statistics

Get UTC+8 timezone current time string

Send tool_call event. Args: tool_name: Name of the tool being called payload: Tool call arguments or results

Save as a single JSON file

Call a tool while maintaining the session.

Connect to the MCP server and initialize the session.

Create a visual progress bar for percentage display

Unified LLM call and logging processing. Args: system_prompt: System prompt for the LLM message_history: Con

Analyze browser session conversation flow

Close client connection. Note: For async clients (AsyncOpenAI, AsyncAnthropic), the connection will be closed when the client object

Send end_of_agent event. Args: agent_name: Internal name of the agent agent_id: The agent ID to end

Send end_of_llm event. Args: agent_name: Name of the agent that finished LLM call

Find the earliest start time from all completed files

Find the latest end time from all completed files

Fix incorrect server_name and tool_name in MCP XML tool calls. Uses the cached tool_server_mapping (parsed from system prompt) to determine

Format task plan as markdown checklist. Args: todos: List of task items message: Optional message to display at the top

Generate the final summarization prompt for an agent. Creates prompts that instruct agents to summarize their work and provide final ans

Connect to all configured servers and get their tool definitions. Returns a list suitable for passing to the Prompt generator.

Load task plan from the JSON file.

Normalize a string by: - Removing all white spaces - Optionally removing punctuation (if remove_punct is True) - Conv

Send start_of_agent event. Args: agent_name: Internal name of the agent display_name: Optional display name

Send start_of_llm event. Args: agent_name: Name of the agent making the LLM call display_name: Optional disp

Calculate Pass@n: number of tasks with at least one correct answer across all runs Returns: Tuple[int, float]: (pass_at_n_count,

Check for duplicate queries and handle rollback if needed. Args: tool_name: Name of the tool being called ar

r""" Extract the content of the last \boxed{...} occurrence in the given text. Supports: - Arbitrary levels of nested brace

Extract task ID from filename

Record a successful query execution.

Calculate overall accuracy (mean) and standard deviation across individual runs

Placeholder for HTML to Markdown conversion function In the original class, this would call self._convert()

Creates and initializes the core components of the agent pipeline. Args: cfg: The Hydra configuration object. Returns:

Traverse all values in the data structure: - If it's a string starting with http, apply urllib.parse.unquote - If it's a list, recursivel

Estimate completion time based on overall progress rate from all completed tasks

Executes the full pipeline for a single task. Args: cfg: The Hydra configuration object. task_id: A unique identifier for th

Execute a single tool call. :param server_name: Server name :param tool_name: Tool name :param arguments: Tool argume

Filter out keys with None values from arguments dictionary. Args: arguments: A dictionary to filter, or any other value Returns

Format tool execution results to be fed back to LLM as user messages. Only includes necessary information (results or errors). Long

Get all browser agent sessions

Get the log directory for the current benchmark and model.

Process tool call results. Only in demo mode: truncate scrape results to 20,000 chars to support more conversation turns.

Put data from other threads - use direct queue put for lower latency

Safely extract text content, handling different content formats

Save task plan to the JSON file.

Truncate result to MAX_RESULT_LEN. Args: result: The full result string to potentially truncate Returns: Truncated resu

Convert a DOCX file to Markdown format. Uses mammoth library to first convert DOCX to HTML, then converts the HTML to Markdown. Arg

Convert an HTML file to Markdown format. Args: local_path: Path to the HTML file to convert. Returns: DocumentConverter

Converts PPTX files to Markdown. Supports headings, tables and images with alt text. Args: local_path: Path to the PPTX file Re

Converts Excel files to Markdown using openpyxl. Preserves color formatting and other cell styling information. Args: local_path

Calculates and adds average values to a summary block.

Calculate number of turns from task data (excluding system prompt)

Encode audio file to base64 and determine format.

Use tiktoken to estimate the number of tokens in text

Use tiktoken to estimate the number of tokens in text

Generate a caption for an audio file using OpenAI's audio transcription. Args: audio_path: Path to the audio file Returns:

Generate a caption for an image using OpenAI's GPT-4o vision model. Args: image_path: Path to the image file Returns: C

Generate a caption for a video using OpenAI's GPT-4o vision model. Args: video_path: Path to the video file Returns: Ca

Get audio duration in seconds. Tries to use wave (for .wav), then falls back to mutagen (for mp3, etc). Returns 0.0 if duration cannot b

Determine the appropriate audio file extension from URL or content type. Args: url: The URL of the audio file content_type:

Get the latest task file for each task ID in a run directory

Handle MCP tag format errors and refusal keywords. Args: assistant_response_text: The LLM response text mess

Determine if LLM judge result indicates correct answer

Check if a task is completed based on its data

Parse new format tool name Args: tool_name: New format tool name, for example: - "tool-server_name

Remove tool results from messages Args: messages: List of message dictionaries keep_tool_result: Number of tool resul

Resolve the actual tool name by combining server_name and tool_name if server_name is not 'default'.

Check if we should block scraping of Hugging Face datasets/spaces. :param tool_name: The name of the tool being called :param

Updates a summary block with data from a single result.

Analyze conversation flow, including tool calls

Configure the miroflow_agent logger with consistent formatting

Convert message list to OpenAI JSON format ChatML Filter out messages with role 'tool', convert content None to empty string

Create MCP server configurations based on agent configuration. Dynamically generates StdioServerParameters for each tool specified in the

Check if current message_history + summary_prompt will exceed context If it will exceed, remove the last assistant-user pair and retu

Convert sub-agent configurations into tool definitions for the main agent. This allows the main agent to invoke sub-agents (like the browsin

Extract message history from log data and save as ChatML format Args: log_data: Log data dictionary output_dir: Output direc

Extract text from LLM response, excluding <use_mcp_tool> tags. Stops immediately when <use_mcp_tool> tag is encountered, returning only

Extract message history from log data and convert to OpenAI ChatML format Args: log_data: Log data dictionary Returns: