council-extract

/council-extract — Extract Model Council Response to Markdown

Extract a completed Perplexity Model Council response from the current browser tab, converting it to structured markdown with model attribution, and save to disk.

Input

$ARGUMENTS = Optional output filename or tabId. Defaults to timestamped name. If a number, treated as tabId.

Prerequisites

• Browser-bridge MCP server running with Chrome extension connected
• A completed Model Council response visible in an active Perplexity tab
• Read ~/.claude/perplexity-selectors.json for CSS selectors

Workflow

Step 1: Identify the target tab

• browser_get_tabs to find tabs with perplexity.ai URLs
• If $ARGUMENTS is a number, use that as the specific tabId
• Otherwise use the most recently active Perplexity tab in the current session group
• If no Perplexity tab found, error: "No Perplexity tab found. Navigate to a council response first."
• Store the target tabId for all subsequent tool calls

Step 2: DOM Recon (adaptive selector discovery)

Run browser_evaluate on the target tab with this script to discover the actual DOM structure:

(() => {
  const results = {
    proseBlocks: document.querySelectorAll('.prose').length,
    proseFallback: document.querySelectorAll("[class*='prose']").length,
    modelNames: [],
    modelRows: [],
    completedIndicators: 0,
    thinkingSteps: 0,
    hasCloseButton: false,
    hasThreeDotMenu: false,
    pageUrl: location.href,
    pageTitle: document.title,
    queryText: ''
  };

  // Find model name elements — look for short text matching known model patterns
  const MODEL_PATTERNS = [/Claude/i, /GPT/i, /Gemini/i, /Llama/i, /Grok/i, /Mistral/i, /Sonar/i, /DeepSeek/i];
  document.querySelectorAll('div, span').forEach(el => {
    const text = el.textContent?.trim();
    if (text && text.length < 40 && MODEL_PATTERNS.some(p => p.test(text))) {
      // Check this isn't a huge container — should be a leaf-ish element
      if (el.children.length <= 3 && !results.modelNames.includes(text)) {
        results.modelNames.push(text);
      }
    }
  });

  // Try configured selectors
  const selectors = {
    councilModelRow: "[class*='interactable'][class*='appearance-none']",
    councilModelRowFallback: "[class*='gap-x-xs'][class*='items-center']",
    councilModelName: "div[class*='font-sans'][class*='text-xs'][class*='font-medium']",
    councilCompletedIndicator: "[class*='Completed'], svg[class*='check']",
    councilThinkingSteps: "[class*='steps']",
    councilPanelClose: "button[aria-label='Close']",
    councilThreeDotMenu: "button[aria-label='More'], [class*='overflow']"
  };

  results.modelRows = document.querySelectorAll(selectors.councilModelRow).length ||
    document.querySelectorAll(selectors.councilModelRowFallback).length;
  results.completedIndicators = document.querySelectorAll(selectors.councilCompletedIndicator).length;
  results.thinkingSteps = document.querySelectorAll(selectors.councilThinkingSteps).length;
  results.hasCloseButton = !!document.querySelector(selectors.councilPanelClose);
  results.hasThreeDotMenu = !!document.querySelector(selectors.councilThreeDotMenu);

  // Try to extract query text from the page
  const askInput = document.querySelector('#ask-input');
  if (askInput) results.queryText = askInput.textContent?.trim() || '';
  // Also try the first user message block
  if (!results.queryText) {
    const userMsg = document.querySelector('[class*="query"], [class*="question"]');
    if (userMsg) results.queryText = userMsg.textContent?.trim().slice(0, 200) || '';
  }

  return results;
})()