* Log a supervisor action

* Check if a message is from the authorized chat

* Check rate limit for a chat. Returns true if allowed, false if throttled. * Sends a warning message on first throttle within the window.

* Find the context that contains #cascade (the chat element)

* Check if injects are disabled via config

* Capture the chat HTML + CSS from #cascade * Returns raw HTML with CSS to preserve exact IDE styling

* Escape Markdown v1 special characters for Telegram * Only _ * ` [ need escaping in parse_mode: 'Markdown'

* Build comprehensive app knowledge from live config

* Connect to CDP and track execution contexts

* Deep merge source into target (target values are overwritten by source)

* Auto-discover the active CDP port by scanning known ports. * Caches the result so subsequent calls are instant.

* Scan running processes to find Antigravity language server * and extract port + CSRF token from command line

* Find CDP workbench targets * If a preferred workspace is configured, filters to that workspace's window

* Record an action for rate limiting

* Make API request to the language server

* Build a text-based progress bar for Telegram * @param {number} percent - 0 to 100 * @returns {string}

* Build smart conversation history: summarize older messages, keep recent ones. * When history grows beyond SUMMARIZE_THRESHOLD, older messages are c

* Build the system prompt that gives the supervisor full knowledge of the app

* Check captured chat HTML for state changes, auto-accept commands, and send Telegram notifications. * Only triggers on transitions (e.g., first erro

* Check for full-page error dialogs (modal overlays outside #cascade). * These include "Agent terminated due to error", "Model quota reached", etc.

* Check rate limiting

* Execute a parsed action * In suggestMode, inject/click actions are queued for human approval

* Extract plain text and buttons from chat HTML

* Calculate remaining time string from reset timestamp