Convert Gemini finish reason to g3 stop reason

Helper to count messages by role

Count commits between two refs

Count total warnings and errors.

Create an adapter for the given model type, if one exists

Helper to create an agent with mock provider and custom README content

Create an LLM provider for the planner based on config

Create research tools for web-based research. These tools enable asynchronous web research via a background scout agent.

Create a summary request prompt for the current conversation.

Helper to create a test agent

Create WebDriver browser automation tools. These tools enable browser automation via Safari WebDriver.

Get the current stage definition

Perform ACD dehydration - save current conversation state to a fragment. Called at the end of each turn when ACD is enabled. This saves all non-syste

Auto-detect Chrome for Testing binary at ~/.chrome-for-testing/. When no `chrome_binary` is configured, this checks for a Chrome for Testing installa

Auto-detect ChromeDriver for Testing binary at ~/.chrome-for-testing/. When no `chromedriver_binary` is configured, this checks for a Chrome for Test

Dispatch a tool call to the appropriate handler. This function routes tool calls to their implementations in the `tools/` module, providing a single

Display context window progress bar.

Display a compact single-line status for the current stage

ANSI color code for duration display based on elapsed time. Returns empty string for fast operations, yellow/orange/red for slower ones.

Format elapsed time for display

Ensure context window has capacity before streaming. Tries thinning first (cheap), then compaction (requires LLM call). Returns Ok(true) if request.me

Estimate token count from byte size

Equals: exactly one value must match the expected string.

Matches: check if any claim value matches a regex pattern.

Execute the `background_process` tool.

Execute code in the specified language

Extract code blocks from LLM response and execute them with UI options

Execute JavaScript code (requires Node.js)

Execute the load_toolset tool. This tool loads a named toolset and returns the tool definitions so the agent learns how to use the newly available to

Execute the `plan_approve` tool.

Execute the `plan_read` tool.

Execute the `plan_write` tool.

Execute Python code

Execute the `read_file` tool.

Execute the `read_image` tool.

Execute the remember tool. Merges new notes with existing memory and saves to file.

Execute the research tool - spawns scout agent in background and returns immediately. This is the **async** version of research. It: 1. Registers a n

Execute the research_status tool - check status of pending research tasks.

Execute the `shell` tool.

Execute the `str_replace` tool.

Execute a tool call with an optional working directory (for discovery commands)

Execute the `webdriver_back` tool.

Execute the `webdriver_click` tool.

Execute the `webdriver_execute_script` tool.

Execute the `webdriver_find_element` tool.

Execute the `webdriver_find_elements` tool.

Execute the `webdriver_forward` tool.

Execute the `webdriver_get_page_source` tool.

Execute the `webdriver_get_title` tool.

Execute the `webdriver_get_url` tool.

Execute the `webdriver_navigate` tool.

Execute the `webdriver_quit` tool.

Execute the `webdriver_refresh` tool.

Execute the `webdriver_screenshot` tool.

Execute the `webdriver_send_keys` tool.

Execute the `webdriver_start` tool.

Execute the `write_envelope` tool. Accepts YAML facts, writes the action envelope, runs verification, and displays a compact pipeline summary via the

Execute the `write_file` tool.

Explore C/C++ codebase

Explore Go codebase

Explore Java codebase

Explore JavaScript codebase

Explore Kotlin codebase

Explore Markdown documentation

Explore Python codebase

Explore Ruby codebase

Explore Rust codebase

Explore SQL files

Explore Swift codebase

Explore TypeScript codebase

Explore YAML configuration files

Extract text content from an assistant message.

Extract coach feedback using multiple fallback methods Tries extraction in this order: 1. Session log file (most reliable for final_output tool calls

Extract code blocks from markdown-formatted text

Extract final_output from message array

Extract coach feedback by reading from the coach agent's specific log file. Uses the coach agent's session ID to find the exact log file.

Extract the last block from a given string

Extract path and content from various argument formats.

Extract messages that should be preserved across compaction.

Extract summary from session logs

Extract a summary from the response (for coach feedback in autonomous mode) This looks for the last substantial text block in the response. Kept for b

Extract tool calls from Gemini response parts

Recursively extract values from a YAML value into facts.