bluera-knowledge 0.33.1 → 0.34.0

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "bluera-knowledge",
-  "version": "0.33.1",
+  "version": "0.34.0",
   "description": "Clone repos, crawl docs, search locally. Fast, authoritative answers for AI coding agents.",
   "author": {
     "name": "Bluera Inc",

package/.claude-plugin/settings.json

@@ -1,6 +1,4 @@
 {
-  "skill_activation": false,
   "websearch_suggestions": false,
-  "webfetch_suggestions": false,
-  "bk_read_suggestions": false
+  "webfetch_suggestions": false
 }

package/CHANGELOG.md

@@ -2,6 +2,33 @@
 
 All notable changes to this project will be documented in this file. See [commit-and-tag-version](https://github.com/absolute-version/commit-and-tag-version) for commit guidelines.
 
+## [0.34.0](https://github.com/blueraai/bluera-knowledge/compare/v0.28.0...v0.34.0) (2026-03-08)
+
+
+### Features
+
+* **activation:** rewrite MCP tool descriptions and simplify hooks for proactive BK usage ([0880aa7](https://github.com/blueraai/bluera-knowledge/commit/0880aa78aa8189957555c58d6e67a562cbae13cb))
+* add statusline module with store count display ([4446501](https://github.com/blueraai/bluera-knowledge/commit/44465015a4734cd7c253bda51c98ce149561fda8))
+* **eval:** 3-agent comparison with BK Grep agent and token tracking ([b3045a8](https://github.com/blueraai/bluera-knowledge/commit/b3045a8ae7df7a2e13ecdf992ff4d1055521460b))
+* **eval:** add agent quality eval comparing with-BK vs without-BK answers ([d8c62d8](https://github.com/blueraai/bluera-knowledge/commit/d8c62d804bd10d572049f4f2a01deb30cb00ccf5))
+* **hooks:** add PostToolUse hook for WebSearch BK suggestions ([d0420b4](https://github.com/blueraai/bluera-knowledge/commit/d0420b473c73092b5e6a28be2699f5fc40e090c3))
+* **mcp:** add file count estimation and ETA to store creation responses ([3d71a30](https://github.com/blueraai/bluera-knowledge/commit/3d71a307275fa06b9810ed606583f4ec2acc5841))
+* **mcp:** add stores:pull command for git pull + re-index ([7ca809c](https://github.com/blueraai/bluera-knowledge/commit/7ca809c3675be19d9d0e29e71c5f75d4f40fce35))
+* **mcp:** optimize search-to-read workflow with store paths, relatedFiles, find-files intent, and file-based get_full_context ([4eb9be9](https://github.com/blueraai/bluera-knowledge/commit/4eb9be97598e475576603b2b9a565946197986ac))
+* search infrastructure, benchmark framework, and model registry ([285ff2f](https://github.com/blueraai/bluera-knowledge/commit/285ff2f5574c4d53b61b31e84fdec43553364e98))
+* **search:** add strong-signal FTS bypass, position-aware reranking, and query expansion ([c9b6eac](https://github.com/blueraai/bluera-knowledge/commit/c9b6eac7de9a6f17100b37879304df5a634b5559))
+* training pipeline, evaluation gate, and experiment docs ([d90b395](https://github.com/blueraai/bluera-knowledge/commit/d90b395a330f4ab09ac6d71db77d659c415aa87d))
+
+
+### Bug Fixes
+
+* **ci:** remove simple mode from validation and lower coverage threshold ([7aea1ac](https://github.com/blueraai/bluera-knowledge/commit/7aea1ac466de8ee13c6dc85c00932838d41ff829))
+* **config:** wire BK_MODEL env var override in ConfigService.load() ([be56426](https://github.com/blueraai/bluera-knowledge/commit/be564263176410c3eb1441bf85c4fdce48ddc7c4))
+* **crawl:** fix nested session error and remove simple mode ([79447b4](https://github.com/blueraai/bluera-knowledge/commit/79447b4604e54760af433deb21e177f9a71cbe8e))
+* **hooks:** drain stdin in shell hooks, register web-research hook, add settings.json ([1cff9b8](https://github.com/blueraai/bluera-knowledge/commit/1cff9b8ab38e732ffa6d9fb2de8b398fddc98302))
+* **mcp:** add runtime files to package.json files array and improve bootstrap error on corrupt cache ([e195f93](https://github.com/blueraai/bluera-knowledge/commit/e195f93a4600c65bcd0fcbb45380aaa1d32bb356))
+* **mcp:** prevent browser postinstall from crashing bootstrap ([e27fe02](https://github.com/blueraai/bluera-knowledge/commit/e27fe021bdb34c8ca5b8bf93720a176435e7973c))
+
 ## [0.33.1](https://github.com/blueraai/bluera-knowledge/compare/v0.28.0...v0.33.1) (2026-03-03)
 
 

package/dist/{chunk-KDZDLJUY.js → chunk-4S6LWHKI.js}

@@ -2,7 +2,7 @@ import {
   createLogger,
   summarizePayload,
   truncateForLog
-} from "./chunk-3TB7TDVF.js";
+} from "./chunk-FYHKBCIH.js";
 
 // src/crawl/intelligent-crawler.ts
 import { EventEmitter } from "events";
@@ -375,10 +375,13 @@ ${this.truncateMarkdown(markdown, 1e5)}`;
         args.push("--json-schema", JSON.stringify(jsonSchema));
         args.push("--output-format", "json");
       }
+      const cleanEnv = Object.fromEntries(
+        Object.entries(process.env).filter(([key]) => !key.startsWith("CLAUDE"))
+      );
       const proc = spawn(claudePath, args, {
         stdio: ["pipe", "pipe", "pipe"],
         cwd: process.cwd(),
-        env: { ...process.env }
+        env: cleanEnv
       });
       let stdout = "";
       let stderr = "";
@@ -457,26 +460,6 @@ ${this.truncateMarkdown(markdown, 1e5)}`;
   }
 };
 
-// src/crawl/link-extractor.ts
-import * as cheerio2 from "cheerio";
-function extractLinks(html, baseUrl) {
-  const $ = cheerio2.load(html);
-  const links = /* @__PURE__ */ new Set();
-  $("a[href]").each((_, el) => {
-    const href = $(el).attr("href");
-    if (href !== void 0 && href !== "") {
-      try {
-        const absoluteUrl = new URL(href, baseUrl).href;
-        if (absoluteUrl.startsWith("http://") || absoluteUrl.startsWith("https://")) {
-          links.add(absoluteUrl);
-        }
-      } catch {
-      }
-    }
-  });
-  return [...links];
-}
-
 // src/crawl/playwright-crawler.ts
 import { chromium } from "playwright";
 var logger2 = createLogger("playwright-crawler");
@@ -583,17 +566,21 @@ var IntelligentCrawler = class extends EventEmitter {
     this.stopped = false;
   }
   /**
-   * Crawl a website with intelligent or simple mode
+   * Crawl a website using Claude-driven intelligent mode
    */
   async *crawl(seedUrl, options = {}) {
-    const { crawlInstruction, extractInstruction, maxPages = 50, simple = false } = options;
+    const {
+      crawlInstruction = "crawl all pages linked from this URL",
+      extractInstruction,
+      maxPages = 50
+    } = options;
     this.visited.clear();
     this.stopped = false;
     logger3.info(
       {
         seedUrl,
         maxPages,
-        mode: simple ? "simple" : crawlInstruction !== void 0 && crawlInstruction !== "" ? "intelligent" : "simple",
+        mode: "intelligent",
         hasExtractInstruction: extractInstruction !== void 0
       },
       "Starting crawl"
@@ -604,19 +591,14 @@ var IntelligentCrawler = class extends EventEmitter {
       totalPages: maxPages
     };
     this.emit("progress", startProgress);
-    const useIntelligentMode = !simple && crawlInstruction !== void 0 && crawlInstruction !== "";
-    if (useIntelligentMode) {
-      yield* this.crawlIntelligent(
-        seedUrl,
-        crawlInstruction,
-        extractInstruction,
-        maxPages,
-        options.useHeadless ?? true,
-        options.preComputedStrategy
-      );
-    } else {
-      yield* this.crawlSimple(seedUrl, extractInstruction, maxPages, options.useHeadless ?? true);
-    }
+    yield* this.crawlIntelligent(
+      seedUrl,
+      crawlInstruction,
+      extractInstruction,
+      maxPages,
+      options.useHeadless ?? true,
+      options.preComputedStrategy
+    );
     logger3.info(
       {
         seedUrl,
@@ -706,72 +688,6 @@ var IntelligentCrawler = class extends EventEmitter {
       }
     }
   }
-  /**
-   * Simple mode: BFS crawling with depth limit
-   */
-  async *crawlSimple(seedUrl, extractInstruction, maxPages, useHeadless = true) {
-    const queue = [{ url: seedUrl, depth: 0 }];
-    const maxDepth = 2;
-    let pagesVisited = 0;
-    while (queue.length > 0 && pagesVisited < maxPages && !this.stopped) {
-      const current = queue.shift();
-      if (!current || this.visited.has(current.url) || current.depth > maxDepth) {
-        continue;
-      }
-      try {
-        const result = await this.crawlSinglePage(
-          current.url,
-          extractInstruction,
-          pagesVisited,
-          useHeadless
-        );
-        result.depth = current.depth;
-        pagesVisited++;
-        yield result;
-        if (current.depth < maxDepth) {
-          try {
-            const links = await this.extractLinks(current.url, useHeadless);
-            if (links.length === 0) {
-              logger3.debug({ url: current.url }, "No links found - page may be a leaf node");
-            } else {
-              logger3.debug(
-                { url: current.url, linkCount: links.length },
-                "Links extracted from page"
-              );
-            }
-            for (const link of links) {
-              if (!this.visited.has(link) && this.isSameDomain(seedUrl, link)) {
-                queue.push({ url: link, depth: current.depth + 1 });
-              }
-            }
-          } catch (error) {
-            const errorProgress = {
-              type: "error",
-              pagesVisited,
-              totalPages: maxPages,
-              currentUrl: current.url,
-              message: `Failed to extract links from ${current.url}`,
-              error: error instanceof Error ? error : new Error(String(error))
-            };
-            this.emit("progress", errorProgress);
-          }
-        }
-      } catch (error) {
-        const errorObj = error instanceof Error ? error : new Error(String(error));
-        if (errorObj.message.includes("Extraction failed") || errorObj.message.includes("Claude CLI not available") || errorObj.message.includes("Headless fetch failed")) {
-          throw errorObj;
-        }
-        const simpleErrorProgress = {
-          type: "error",
-          pagesVisited,
-          totalPages: maxPages,
-          currentUrl: current.url,
-          error: errorObj
-        };
-        this.emit("progress", simpleErrorProgress);
-      }
-    }
-  }
   /**
    * Crawl a single page: fetch, convert to markdown, optionally extract
    */
@@ -869,40 +785,6 @@ var IntelligentCrawler = class extends EventEmitter {
       );
     }
   }
-  /**
-   * Extract links from a page using Playwright or cheerio
-   */
-  async extractLinks(url, useHeadless = true) {
-    try {
-      if (useHeadless) {
-        const result = await fetchWithPlaywright(url);
-        return result.links;
-      }
-      const response = await axios.get(url, {
-        timeout: this.config.timeout ?? DEFAULT_TIMEOUT2,
-        headers: {
-          "User-Agent": this.config.userAgent ?? DEFAULT_USER_AGENT
-        }
-      });
-      return extractLinks(response.data, url);
-    } catch (error) {
-      const errorMessage = error instanceof Error ? error.message : String(error);
-      logger3.error({ url, error: errorMessage }, "Failed to extract links");
-      throw new Error(`Link extraction failed for ${url}: ${errorMessage}`);
-    }
-  }
-  /**
-   * Check if two URLs are from the same domain
-   */
-  isSameDomain(url1, url2) {
-    try {
-      const domain1 = new URL(url1).hostname.toLowerCase();
-      const domain2 = new URL(url2).hostname.toLowerCase();
-      return domain1 === domain2 || domain1.endsWith(`.${domain2}`) || domain2.endsWith(`.${domain1}`);
-    } catch {
-      return false;
-    }
-  }
   /**
    * Stop the crawler
    */
@@ -916,4 +798,4 @@ export {
   getCrawlStrategy,
   IntelligentCrawler
 };
-//# sourceMappingURL=chunk-KDZDLJUY.js.map
+//# sourceMappingURL=chunk-4S6LWHKI.js.map

package/dist/chunk-4S6LWHKI.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/crawl/intelligent-crawler.ts","../src/crawl/article-converter.ts","../src/crawl/markdown-utils.ts","../src/crawl/claude-client.ts","../src/crawl/playwright-crawler.ts"],"sourcesContent":["/**\n * Intelligent web crawler with natural language control\n */\n\nimport { EventEmitter } from 'node:events';\nimport axios from 'axios';\nimport { convertHtmlToMarkdown } from './article-converter.js';\nimport { ClaudeClient, type CrawlStrategy } from './claude-client.js';\nimport { fetchWithPlaywright, closeBrowser } from './playwright-crawler.js';\nimport { createLogger, summarizePayload } from '../logging/index.js';\n\nconst logger = createLogger('crawler');\n\n/**\n * Get crawl strategy from Claude CLI without initializing lancedb.\n * Call this BEFORE createServices() to avoid fork-safety issues with lancedb.\n *\n * LanceDB's native Rust code is not fork-safe. If we spawn Claude CLI\n * after lancedb is loaded, the fork corrupts lancedb's mutex state.\n */\nexport async function getCrawlStrategy(\n  seedUrl: string,\n  crawlInstruction: string,\n  useHeadless: boolean = true\n): Promise<CrawlStrategy> {\n  if (!ClaudeClient.isAvailable()) {\n    throw new Error('Claude CLI not available: install Claude Code for intelligent crawling');\n  }\n\n  const client = new ClaudeClient();\n\n  // Fetch seed HTML\n  let seedHtml: string;\n  if (useHeadless) {\n    const headlessResult = await fetchWithPlaywright(seedUrl);\n    seedHtml = headlessResult.html;\n  } else {\n    const response = await axios.get(seedUrl, {\n      timeout: 30000,\n      headers: {\n        'User-Agent':\n          'Mozilla/5.0 (compatible; BlueraKnowledge/1.0; +https://github.com/blueraai/bluera-knowledge)',\n      },\n    });\n    seedHtml = typeof response.data === 'string' ? response.data : JSON.stringify(response.data);\n  }\n\n  // Get strategy from Claude\n  return client.determineCrawlUrls(seedUrl, seedHtml, crawlInstruction);\n}\n\nexport interface CrawlConfig {\n  userAgent?: string; // User-Agent header for requests\n  timeout?: number; // Per-page timeout in ms (default: 30000)\n  maxConcurrency?: number; // Max concurrent requests (for future use)\n}\n\nexport interface CrawlOptions {\n  crawlInstruction?: string; // Natural language: what to crawl\n  extractInstruction?: string; // Natural language: what to extract\n  maxPages?: number; // Max pages to crawl (default: 50)\n  timeout?: number; // Per-page timeout in ms (default: 30000)\n  useHeadless?: boolean; // Enable headless browser for JavaScript-rendered sites\n  preComputedStrategy?: CrawlStrategy; // Pre-computed strategy (to avoid fork issues with lancedb)\n}\n\nexport interface CrawlResult {\n  url: string;\n  title?: string;\n  markdown: string;\n  extracted?: string;\n  depth?: number;\n}\n\nexport interface CrawlProgress {\n  type: 'start' | 'strategy' | 'page' | 'extraction' | 'complete' | 'error';\n  pagesVisited: number;\n  totalPages: number;\n  currentUrl?: string;\n  message?: string;\n  error?: Error;\n}\n\n// Default values for crawl config\nconst DEFAULT_USER_AGENT =\n  'Mozilla/5.0 (compatible; BlueraKnowledge/1.0; +https://github.com/blueraai/bluera-knowledge)';\nconst DEFAULT_TIMEOUT = 30000;\n\n/**\n * Intelligent crawler that uses Claude CLI for strategy and extraction\n */\nexport class IntelligentCrawler extends EventEmitter {\n  private readonly claudeClient: ClaudeClient;\n  private readonly visited: Set<string>;\n  private readonly config: CrawlConfig;\n  private stopped: boolean;\n\n  constructor(config?: CrawlConfig) {\n    super();\n    this.claudeClient = new ClaudeClient();\n    this.visited = new Set();\n    this.config = config ?? {};\n    this.stopped = false;\n  }\n\n  /**\n   * Crawl a website using Claude-driven intelligent mode\n   */\n  async *crawl(seedUrl: string, options: CrawlOptions = {}): AsyncIterable<CrawlResult> {\n    const {\n      crawlInstruction = 'crawl all pages linked from this URL',\n      extractInstruction,\n      maxPages = 50,\n    } = options;\n\n    this.visited.clear();\n    this.stopped = false;\n\n    logger.info(\n      {\n        seedUrl,\n        maxPages,\n        mode: 'intelligent',\n        hasExtractInstruction: extractInstruction !== undefined,\n      },\n      'Starting crawl'\n    );\n\n    const startProgress: CrawlProgress = {\n      type: 'start',\n      pagesVisited: 0,\n      totalPages: maxPages,\n    };\n    this.emit('progress', startProgress);\n\n    yield* this.crawlIntelligent(\n      seedUrl,\n      crawlInstruction,\n      extractInstruction,\n      maxPages,\n      options.useHeadless ?? true,\n      options.preComputedStrategy\n    );\n\n    logger.info(\n      {\n        seedUrl,\n        pagesVisited: this.visited.size,\n      },\n      'Crawl complete'\n    );\n\n    // Warn if crawl discovered far fewer pages than requested\n    if (this.visited.size === 1 && maxPages > 1) {\n      const warningProgress: CrawlProgress = {\n        type: 'error',\n        pagesVisited: this.visited.size,\n        totalPages: maxPages,\n        message: `Warning: Only crawled 1 page despite maxPages=${String(maxPages)}. Link discovery may have failed. If using --fast mode, try without it for JavaScript-heavy sites.`,\n        error: new Error('Low page discovery'),\n      };\n      this.emit('progress', warningProgress);\n    }\n\n    const completeProgress: CrawlProgress = {\n      type: 'complete',\n      pagesVisited: this.visited.size,\n      totalPages: this.visited.size,\n    };\n    this.emit('progress', completeProgress);\n  }\n\n  /**\n   * Intelligent mode: Use Claude to determine which URLs to crawl\n   */\n  private async *crawlIntelligent(\n    seedUrl: string,\n    crawlInstruction: string,\n    extractInstruction: string | undefined,\n    maxPages: number,\n    useHeadless: boolean = true,\n    preComputedStrategy?: CrawlStrategy\n  ): AsyncIterable<CrawlResult> {\n    let strategy: CrawlStrategy;\n\n    // Use pre-computed strategy if provided (avoids fork issues with lancedb)\n    if (preComputedStrategy !== undefined) {\n      strategy = preComputedStrategy;\n      const strategyProgress: CrawlProgress = {\n        type: 'strategy',\n        pagesVisited: 0,\n        totalPages: maxPages,\n        message: `Using pre-computed strategy: ${String(strategy.urls.length)} URLs to crawl`,\n      };\n      this.emit('progress', strategyProgress);\n    } else {\n      // Check if Claude CLI is available before attempting intelligent mode\n      if (!ClaudeClient.isAvailable()) {\n        throw new Error('Claude CLI not available: install Claude Code for intelligent crawling');\n      }\n\n      try {\n        // Step 1: Fetch seed page HTML\n        const strategyStartProgress: CrawlProgress = {\n          type: 'strategy',\n          pagesVisited: 0,\n          totalPages: maxPages,\n          currentUrl: seedUrl,\n          message: 'Analyzing page structure with Claude...',\n        };\n        this.emit('progress', strategyStartProgress);\n\n        const seedHtml = await this.fetchHtml(seedUrl, useHeadless);\n\n        // Step 2: Ask Claude which URLs to crawl (pass seedUrl for relative URL resolution)\n        strategy = await this.claudeClient.determineCrawlUrls(seedUrl, seedHtml, crawlInstruction);\n\n        const strategyCompleteProgress: CrawlProgress = {\n          type: 'strategy',\n          pagesVisited: 0,\n          totalPages: maxPages,\n          message: `Claude identified ${String(strategy.urls.length)} URLs to crawl: ${strategy.reasoning}`,\n        };\n        this.emit('progress', strategyCompleteProgress);\n      } catch (error) {\n        // Re-throw strategy errors - do not fall back silently\n        throw error instanceof Error ? error : new Error(String(error));\n      }\n    }\n\n    // Step 3: Crawl each URL from Claude's strategy\n    let pagesVisited = 0;\n\n    for (const url of strategy.urls) {\n      if (this.stopped || pagesVisited >= maxPages) break;\n      if (this.visited.has(url)) continue;\n\n      try {\n        const result = await this.crawlSinglePage(\n          url,\n          extractInstruction,\n          pagesVisited,\n          useHeadless\n        );\n        pagesVisited++;\n        yield result;\n      } catch (error) {\n        const pageErrorProgress: CrawlProgress = {\n          type: 'error',\n          pagesVisited,\n          totalPages: maxPages,\n          currentUrl: url,\n          error: error instanceof Error ? error : new Error(String(error)),\n        };\n        this.emit('progress', pageErrorProgress);\n      }\n    }\n  }\n\n  /**\n   * Crawl a single page: fetch, convert to markdown, optionally extract\n   */\n  private async crawlSinglePage(\n    url: string,\n    extractInstruction: string | undefined,\n    pagesVisited: number,\n    useHeadless: boolean = true\n  ): Promise<CrawlResult> {\n    const pageProgress: CrawlProgress = {\n      type: 'page',\n      pagesVisited,\n      totalPages: 0,\n      currentUrl: url,\n    };\n    this.emit('progress', pageProgress);\n\n    // Mark as visited\n    this.visited.add(url);\n\n    // Fetch HTML\n    const html = await this.fetchHtml(url, useHeadless);\n\n    // Convert to clean markdown using slurp-ai techniques\n    // Note: convertHtmlToMarkdown throws on errors, no need to check success\n    const conversion = await convertHtmlToMarkdown(html, url);\n\n    logger.debug(\n      {\n        url,\n        title: conversion.title,\n        markdownLength: conversion.markdown.length,\n      },\n      'Article converted to markdown'\n    );\n\n    let extracted: string | undefined;\n\n    // Optional: Extract specific information using Claude\n    if (extractInstruction !== undefined && extractInstruction !== '') {\n      // Throw if extraction requested but Claude CLI isn't available\n      if (!ClaudeClient.isAvailable()) {\n        throw new Error('Claude CLI not available: install Claude Code for extraction');\n      }\n\n      const extractionProgress: CrawlProgress = {\n        type: 'extraction',\n        pagesVisited,\n        totalPages: 0,\n        currentUrl: url,\n      };\n      this.emit('progress', extractionProgress);\n\n      extracted = await this.claudeClient.extractContent(conversion.markdown, extractInstruction);\n    }\n\n    return {\n      url,\n      ...(conversion.title !== undefined && { title: conversion.title }),\n      markdown: conversion.markdown,\n      ...(extracted !== undefined && { extracted }),\n    };\n  }\n\n  /**\n   * Fetch HTML content from a URL\n   */\n  private async fetchHtml(url: string, useHeadless: boolean = true): Promise<string> {\n    const startTime = Date.now();\n    logger.debug({ url, useHeadless }, 'Fetching HTML');\n\n    if (useHeadless) {\n      try {\n        const result = await fetchWithPlaywright(url);\n        const durationMs = Date.now() - startTime;\n        logger.info(\n          {\n            url,\n            useHeadless: true,\n            durationMs,\n            ...summarizePayload(result.html, 'raw-html', url),\n          },\n          'Raw HTML fetched'\n        );\n        return result.html;\n      } catch (error) {\n        // Wrap with distinctive message for error categorization\n        throw new Error(\n          `Headless fetch failed: ${error instanceof Error ? error.message : String(error)}`\n        );\n      }\n    }\n\n    // Original axios implementation for static sites\n    try {\n      const response = await axios.get<string>(url, {\n        timeout: this.config.timeout ?? DEFAULT_TIMEOUT,\n        headers: {\n          'User-Agent': this.config.userAgent ?? DEFAULT_USER_AGENT,\n        },\n      });\n\n      const durationMs = Date.now() - startTime;\n      logger.info(\n        {\n          url,\n          useHeadless: false,\n          durationMs,\n          ...summarizePayload(response.data, 'raw-html', url),\n        },\n        'Raw HTML fetched'\n      );\n\n      return response.data;\n    } catch (error) {\n      logger.error(\n        { url, error: error instanceof Error ? error.message : String(error) },\n        'Failed to fetch HTML'\n      );\n      throw new Error(\n        `Failed to fetch ${url}: ${error instanceof Error ? error.message : String(error)}`\n      );\n    }\n  }\n\n  /**\n   * Stop the crawler\n   */\n  async stop(): Promise<void> {\n    this.stopped = true;\n    await closeBrowser();\n  }\n}\n","/**\n * Article converter using @extractus/article-extractor and Turndown\n * Produces clean markdown from HTML using slurp-ai techniques\n */\n\nimport { extractFromHtml } from '@extractus/article-extractor';\nimport TurndownService from 'turndown';\nimport { gfm } from 'turndown-plugin-gfm';\nimport { preprocessHtmlForCodeBlocks, cleanupMarkdown } from './markdown-utils.js';\nimport { createLogger, truncateForLog } from '../logging/index.js';\n\nconst logger = createLogger('article-converter');\n\nexport interface ConversionResult {\n  markdown: string;\n  title?: string;\n}\n\n/**\n * Convert HTML to clean markdown using best practices from slurp-ai\n *\n * Pipeline:\n * 1. Extract main article content (strips navigation, ads, boilerplate)\n * 2. Preprocess HTML (handle MkDocs code blocks)\n * 3. Convert to markdown with Turndown + GFM\n * 4. Cleanup markdown (regex patterns)\n */\nexport async function convertHtmlToMarkdown(html: string, url: string): Promise<ConversionResult> {\n  logger.debug({ url, htmlLength: html.length }, 'Starting HTML conversion');\n\n  try {\n    // Step 1: Extract main article content\n    let articleHtml: string;\n    let title: string | undefined;\n\n    try {\n      const article = await extractFromHtml(html, url);\n      if (article?.content !== undefined && article.content !== '') {\n        articleHtml = article.content;\n        title = article.title !== undefined && article.title !== '' ? article.title : undefined;\n        logger.debug(\n          {\n            url,\n            title,\n            extractedLength: articleHtml.length,\n            usedFullHtml: false,\n          },\n          'Article content extracted'\n        );\n      } else {\n        // Fallback to full HTML if extraction fails\n        articleHtml = html;\n        logger.debug(\n          { url, usedFullHtml: true },\n          'Article extraction returned empty, using full HTML'\n        );\n      }\n    } catch (extractError) {\n      // Fallback to full HTML if extraction fails\n      articleHtml = html;\n      logger.debug(\n        {\n          url,\n          usedFullHtml: true,\n          error: extractError instanceof Error ? extractError.message : String(extractError),\n        },\n        'Article extraction failed, using full HTML'\n      );\n    }\n\n    // Step 2: Preprocess HTML for code blocks\n    const preprocessed = preprocessHtmlForCodeBlocks(articleHtml);\n\n    // Step 3: Configure Turndown with custom rules\n    const turndownService = new TurndownService({\n      headingStyle: 'atx', // Use # style headings\n      codeBlockStyle: 'fenced', // Use ``` style code blocks\n      fence: '```',\n      emDelimiter: '*',\n      strongDelimiter: '**',\n      linkStyle: 'inlined',\n    });\n\n    // Add GitHub Flavored Markdown support (tables, strikethrough, task lists)\n    turndownService.use(gfm);\n\n    // Custom rule for headings with anchors (from slurp-ai)\n    turndownService.addRule('headingsWithAnchors', {\n      filter: ['h1', 'h2', 'h3', 'h4', 'h5', 'h6'],\n      replacement(content: string, node: HTMLElement): string {\n        const level = Number(node.nodeName.charAt(1));\n        const hashes = '#'.repeat(level);\n        const cleanContent = content\n          .replace(/\\[\\]\\([^)]*\\)/g, '') // Remove empty links\n          .replace(/\\s+/g, ' ') // Normalize whitespace\n          .trim();\n        return cleanContent !== '' ? `\\n\\n${hashes} ${cleanContent}\\n\\n` : '';\n      },\n    });\n\n    // Convert to markdown\n    const rawMarkdown = turndownService.turndown(preprocessed);\n\n    // Step 4: Cleanup markdown with comprehensive regex patterns\n    const markdown = cleanupMarkdown(rawMarkdown);\n\n    logger.debug(\n      {\n        url,\n        title,\n        rawMarkdownLength: rawMarkdown.length,\n        finalMarkdownLength: markdown.length,\n      },\n      'HTML to markdown conversion complete'\n    );\n\n    // Log markdown preview at trace level\n    logger.trace(\n      {\n        url,\n        markdownPreview: truncateForLog(markdown, 1000),\n      },\n      'Markdown content preview'\n    );\n\n    return {\n      markdown,\n      ...(title !== undefined && { title }),\n    };\n  } catch (error) {\n    logger.error(\n      {\n        url,\n        error: error instanceof Error ? error.message : String(error),\n      },\n      'HTML to markdown conversion failed'\n    );\n\n    // Re-throw errors - do not return graceful degradation\n    throw error instanceof Error ? error : new Error(String(error));\n  }\n}\n","/**\n * Markdown conversion utilities ported from slurp-ai\n * Source: https://github.com/ratacat/slurp-ai\n *\n * These utilities handle complex documentation site patterns (MkDocs, Sphinx, etc.)\n * and produce clean, well-formatted markdown.\n */\n\nimport * as cheerio from 'cheerio';\n\n/**\n * Detect language from code element class names.\n * Handles various class naming patterns from different highlighters.\n */\nfunction detectLanguageFromClass(className: string | undefined): string {\n  if (className === undefined || className === '') return '';\n\n  // Common patterns: \"language-python\", \"lang-js\", \"highlight-python\", \"python\", \"hljs language-python\"\n  const patterns = [\n    /language-(\\w+)/i,\n    /lang-(\\w+)/i,\n    /highlight-(\\w+)/i,\n    /hljs\\s+(\\w+)/i,\n    /^(\\w+)$/i,\n  ];\n\n  for (const pattern of patterns) {\n    const match = className.match(pattern);\n    if (match?.[1] !== undefined) {\n      const lang = match[1].toLowerCase();\n      // Filter out common non-language classes\n      if (!['hljs', 'highlight', 'code', 'pre', 'block', 'inline'].includes(lang)) {\n        return lang;\n      }\n    }\n  }\n\n  return '';\n}\n\n/**\n * Escape HTML special characters for safe embedding in HTML.\n */\nfunction escapeHtml(text: string): string {\n  return text\n    .replace(/&/g, '&amp;')\n    .replace(/</g, '&lt;')\n    .replace(/>/g, '&gt;')\n    .replace(/\"/g, '&quot;')\n    .replace(/'/g, '&#039;');\n}\n\n/**\n * Preprocess HTML to handle MkDocs/Material theme code blocks.\n *\n * MkDocs wraps code in tables for line numbers:\n * <table><tbody><tr><td>line numbers</td><td><pre><code>code</code></pre></td></tr></tbody></table>\n *\n * This function converts them to standard <pre><code> blocks that Turndown handles correctly.\n * Also strips syntax highlighting spans and empty anchors from code.\n */\nexport function preprocessHtmlForCodeBlocks(html: string): string {\n  if (!html || typeof html !== 'string') return html;\n\n  const $ = cheerio.load(html);\n\n  // Handle MkDocs/Material table-wrapped code blocks\n  $('table').each((_i, table) => {\n    const $table = $(table);\n\n    // Check if this table contains a code block\n    const $codeCell = $table.find('td pre code, td div pre code');\n\n    if ($codeCell.length > 0) {\n      // This is a code block table - extract the code\n      const $pre = $codeCell.closest('pre');\n      const $code = $codeCell.first();\n\n      // Get language from class\n      let language = detectLanguageFromClass($code.attr('class'));\n      if (!language) {\n        language = detectLanguageFromClass($pre.attr('class'));\n      }\n\n      // Get the text content, stripping all inner HTML tags\n      const codeText = $code.text();\n\n      // Create a clean pre > code block\n      const cleanPre = `<pre><code class=\"language-${language}\">${escapeHtml(codeText)}</code></pre>`;\n\n      // Replace the entire table with the clean code block\n      $table.replaceWith(cleanPre);\n    }\n  });\n\n  // Strip empty anchor tags used for line numbers\n  $('pre a, code a').each((_i, anchor) => {\n    const $anchor = $(anchor);\n    if (!$anchor.text().trim()) {\n      $anchor.remove();\n    }\n  });\n\n  // Strip syntax highlighting spans inside code blocks, keeping only text\n  $('pre span, code span').each((_i, span) => {\n    const $span = $(span);\n    $span.replaceWith($span.text());\n  });\n\n  // Handle standalone pre blocks that might have spans/anchors\n  $('pre').each((_i, pre) => {\n    const $pre = $(pre);\n    // If this pre has a code child, it was already processed\n    if ($pre.find('code').length === 0) {\n      // Direct pre without code - get text content\n      const text = $pre.text();\n      const lang = detectLanguageFromClass($pre.attr('class'));\n      $pre.html(`<code class=\"language-${lang}\">${escapeHtml(text)}</code>`);\n    }\n  });\n\n  return $.html();\n}\n\n/**\n * Apply comprehensive cleanup rules to markdown content.\n *\n * Formatting rules:\n * - Double newlines between paragraphs and headings\n * - Double newlines before lists when preceded by normal text\n * - Single newlines between list items\n * - No blank lines inside code blocks\n */\nexport function cleanupMarkdown(markdown: string): string {\n  if (!markdown) return '';\n\n  const trimmed = markdown.trim();\n  if (trimmed === '') return '';\n\n  let result = trimmed;\n\n  // 0. Fix broken headings where ## is on its own line followed by the text\n  // Pattern: \"## \\n\\nSome text\" → \"## Some text\"\n  result = result.replace(/^(#{1,6})\\s*\\n\\n+(\\S[^\\n]*)/gm, '$1 $2');\n\n  // 0.5. Normalize multiple spaces after heading markers to single space\n  // Pattern: \"##  Subtitle\" → \"## Subtitle\"\n  result = result.replace(/(#{1,6})\\s{2,}/g, '$1 ');\n\n  // 1. Fix navigation links with excessive whitespace\n  result = result.replace(/\\*\\s+\\[\\s*([^\\n]+?)\\s*\\]\\(([^)]+)\\)/g, '* [$1]($2)');\n\n  // 2. Handle headings with specific newline requirements\n\n  // Text followed by heading should have a single newline between them (no blank line)\n  result = result.replace(/([^\\n])\\n\\n+(#\\s)/g, '$1\\n$2');\n\n  // Add double newlines between text and next heading\n  result = result.replace(/(Some text\\.)\\n(##\\s)/g, '$1\\n\\n$2');\n\n  // Double newlines after a heading when followed by text\n  result = result.replace(/(#{1,6}\\s[^\\n]+)\\n([^#\\n])/g, '$1\\n\\n$2');\n\n  // Double newlines between headings\n  result = result.replace(/(#{1,6}\\s[^\\n]+)\\n(#{1,6}\\s)/g, '$1\\n\\n$2');\n\n  // 3. Lists - ensure all list items have single newlines only\n  result = result.replace(/(\\* Item 1)\\n\\n+(\\* Item 2)\\n\\n+(\\* Item 3)/g, '$1\\n$2\\n$3');\n\n  // 3.5. General list item spacing - ensure single newlines between list items\n  result = result.replace(/(^\\*\\s[^\\n]+)\\n{2,}(^\\*\\s)/gm, '$1\\n$2');\n\n  // 4. Clean up excessive blank lines (3+ newlines → 2 newlines)\n  result = result.replace(/\\n{3,}/g, '\\n\\n');\n\n  // 5. Code blocks - no blank lines after opening or before closing backticks\n  result = result.replace(/(```[^\\n]*)\\n\\n+/g, '$1\\n');\n  result = result.replace(/\\n\\n+```/g, '\\n```');\n\n  // 6. Remove empty list items\n  result = result.replace(/\\*\\s*\\n\\s*\\*/g, '*');\n\n  // 7. Strip any remaining HTML tags that leaked through (common in MkDocs/Material)\n  // Remove table structure tags\n  result = result.replace(/<\\/?table[^>]*>/gi, '');\n  result = result.replace(/<\\/?tbody[^>]*>/gi, '');\n  result = result.replace(/<\\/?thead[^>]*>/gi, '');\n  result = result.replace(/<\\/?tr[^>]*>/gi, '');\n  result = result.replace(/<\\/?td[^>]*>/gi, '');\n  result = result.replace(/<\\/?th[^>]*>/gi, '');\n\n  // Remove empty anchor tags: <a></a> or <a id=\"...\"></a>\n  result = result.replace(/<a[^>]*><\\/a>/gi, '');\n\n  // Remove span tags (syntax highlighting remnants)\n  result = result.replace(/<\\/?span[^>]*>/gi, '');\n\n  // Remove div tags\n  result = result.replace(/<\\/?div[^>]*>/gi, '');\n\n  // Remove pre/code tags that leaked\n  result = result.replace(/<\\/?pre[^>]*>/gi, '');\n  result = result.replace(/<\\/?code[^>]*>/gi, '');\n\n  // 8. Remove empty markdown links: [](url) and []()\n  result = result.replace(/\\[\\]\\([^)]*\\)/g, '');\n\n  // 9. Remove codelineno references that leaked into content\n  // Pattern: [](_file.md#__codelineno-N-M)\n  result = result.replace(/\\[\\]\\([^)]*#__codelineno-[^)]+\\)/g, '');\n\n  // Also clean inline codelineno patterns\n  result = result.replace(/\\[?\\]?\\([^)]*#__codelineno-[^)]*\\)/g, '');\n\n  // 10. Clean up any double-escaped HTML entities that might result\n  result = result.replace(/&amp;lt;/g, '&lt;');\n  result = result.replace(/&amp;gt;/g, '&gt;');\n  result = result.replace(/&amp;amp;/g, '&amp;');\n\n  // 11. Final cleanup - normalize excessive whitespace from removed tags\n  result = result.replace(/\\n{3,}/g, '\\n\\n');\n  result = result.replace(/[ \\t]+\\n/g, '\\n');\n\n  return result;\n}\n","/**\n * Claude CLI client for intelligent crawling and extraction\n * Uses `claude -p` programmatically to analyze page structure and extract content\n */\n\nimport { spawn, execSync } from 'node:child_process';\nimport { existsSync } from 'node:fs';\nimport { homedir } from 'node:os';\nimport { join } from 'node:path';\n\n/**\n * Schema for crawl strategy response from Claude\n */\nexport interface CrawlStrategy {\n  urls: string[];\n  reasoning: string;\n}\n\nconst CRAWL_STRATEGY_SCHEMA = {\n  type: 'object',\n  properties: {\n    urls: {\n      type: 'array',\n      items: { type: 'string' },\n      description: 'List of URLs to crawl based on the instruction',\n    },\n    reasoning: {\n      type: 'string',\n      description: 'Brief explanation of why these URLs were selected',\n    },\n  },\n  required: ['urls', 'reasoning'],\n};\n\n/**\n * Client for interacting with Claude Code CLI\n */\nexport class ClaudeClient {\n  private readonly timeout: number;\n  private static availabilityChecked = false;\n  private static available = false;\n  private static claudePath: string | null = null;\n\n  /**\n   * Get the path to the Claude CLI binary\n   * Checks in order:\n   * 1. CLAUDE_BIN environment variable (explicit override)\n   * 2. ~/.claude/local/claude (newer installation location)\n   * 3. ~/.local/bin/claude (standard installation location)\n   * 4. 'claude' in PATH (custom installations)\n   */\n  static getClaudePath(): string | null {\n    // Check environment variable override\n    const envPath = process.env['CLAUDE_BIN'];\n    if (envPath !== undefined && envPath !== '' && existsSync(envPath)) {\n      return envPath;\n    }\n\n    // Check ~/.claude/local/claude (newer location)\n    const claudeLocalPath = join(homedir(), '.claude', 'local', 'claude');\n    if (existsSync(claudeLocalPath)) {\n      return claudeLocalPath;\n    }\n\n    // Check ~/.local/bin/claude (standard location)\n    const localBinPath = join(homedir(), '.local', 'bin', 'claude');\n    if (existsSync(localBinPath)) {\n      return localBinPath;\n    }\n\n    // Check if 'claude' is in PATH (custom installations, uses 'command -v' which handles aliases)\n    try {\n      const result = execSync('command -v claude', { stdio: ['pipe', 'pipe', 'ignore'] });\n      const path = result.toString().trim();\n      if (path) {\n        return path;\n      }\n    } catch {\n      // Not in PATH\n    }\n\n    return null;\n  }\n\n  /**\n   * Check if Claude CLI is available\n   * Result is cached after first check for performance\n   */\n  static isAvailable(): boolean {\n    if (!ClaudeClient.availabilityChecked) {\n      ClaudeClient.claudePath = ClaudeClient.getClaudePath();\n      ClaudeClient.available = ClaudeClient.claudePath !== null;\n      ClaudeClient.availabilityChecked = true;\n    }\n    return ClaudeClient.available;\n  }\n\n  /**\n   * Get the cached Claude path (call isAvailable first)\n   */\n  static getCachedPath(): string | null {\n    return ClaudeClient.claudePath;\n  }\n\n  /**\n   * Reset availability cache (for testing)\n   */\n  static resetAvailabilityCache(): void {\n    ClaudeClient.availabilityChecked = false;\n    ClaudeClient.available = false;\n  }\n\n  constructor(options: { timeout?: number } = {}) {\n    this.timeout = options.timeout ?? 30000; // 30s default\n  }\n\n  /**\n   * Determine which URLs to crawl based on natural language instruction\n   *\n   * @param seedUrl - The URL of the seed page (for resolving relative URLs)\n   * @param seedHtml - HTML content of the seed page\n   * @param instruction - Natural language crawl instruction (e.g., \"scrape all Getting Started pages\")\n   * @returns List of URLs to crawl with reasoning\n   */\n  async determineCrawlUrls(\n    seedUrl: string,\n    seedHtml: string,\n    instruction: string\n  ): Promise<CrawlStrategy> {\n    const prompt = `You are analyzing a webpage to determine which pages to crawl based on the user's instruction.\n\nBase URL: ${seedUrl}\n\nInstruction: ${instruction}\n\nWebpage HTML (analyze the navigation structure, links, and content):\n${this.truncateHtml(seedHtml, 50000)}\n\nBased on the instruction, extract and return a list of absolute URLs that should be crawled. When you encounter relative URLs (starting with \"/\" or without a protocol), resolve them against the Base URL. For example, if Base URL is \"https://example.com/docs\" and you see href=\"/docs/hooks\", return \"https://example.com/docs/hooks\".\n\nLook for navigation menus, sidebars, headers, and link structures that match the instruction.\n\nReturn only URLs that are relevant to the instruction. If the instruction mentions specific sections (e.g., \"Getting Started\"), find links in those sections.`;\n\n    try {\n      const result = await this.callClaude(prompt, CRAWL_STRATEGY_SCHEMA);\n      const rawParsed: unknown = JSON.parse(result);\n\n      // Claude CLI with --json-schema returns wrapper: {type, result, structured_output: {...}}\n      // Extract structured_output if present, otherwise use raw response\n      const parsed = this.extractStructuredOutput(rawParsed);\n\n      // Validate and narrow type\n      if (\n        typeof parsed !== 'object' ||\n        parsed === null ||\n        !('urls' in parsed) ||\n        !('reasoning' in parsed) ||\n        !Array.isArray(parsed.urls) ||\n        parsed.urls.length === 0 ||\n        typeof parsed.reasoning !== 'string' ||\n        !parsed.urls.every((url) => typeof url === 'string')\n      ) {\n        throw new Error('Claude returned invalid crawl strategy');\n      }\n\n      // Type is now properly narrowed - urls is string[] after validation\n      return { urls: parsed.urls, reasoning: parsed.reasoning };\n    } catch (error) {\n      throw new Error(\n        `Failed to determine crawl strategy: ${error instanceof Error ? error.message : String(error)}`\n      );\n    }\n  }\n\n  /**\n   * Extract specific information from markdown content using natural language\n   *\n   * @param markdown - Page content in markdown format\n   * @param instruction - Natural language extraction instruction (e.g., \"extract pricing info\")\n   * @returns Extracted information as text\n   */\n  async extractContent(markdown: string, instruction: string): Promise<string> {\n    const prompt = `${instruction}\n\nContent to analyze:\n${this.truncateMarkdown(markdown, 100000)}`;\n\n    try {\n      const result = await this.callClaude(prompt);\n      return result.trim();\n    } catch (error) {\n      throw new Error(\n        `Failed to extract content: ${error instanceof Error ? error.message : String(error)}`\n      );\n    }\n  }\n\n  /**\n   * Call Claude CLI with a prompt\n   *\n   * @param prompt - The prompt to send to Claude\n   * @param jsonSchema - Optional JSON schema for structured output\n   * @returns Claude's response as a string\n   */\n  private async callClaude(prompt: string, jsonSchema?: Record<string, unknown>): Promise<string> {\n    return new Promise<string>((resolve, reject) => {\n      // Ensure we have Claude path\n      const claudePath = ClaudeClient.getCachedPath();\n      if (claudePath === null) {\n        reject(new Error('Claude CLI not available'));\n        return;\n      }\n\n      const args = ['-p'];\n\n      // Add JSON schema if provided\n      if (jsonSchema) {\n        args.push('--json-schema', JSON.stringify(jsonSchema));\n        args.push('--output-format', 'json');\n      }\n\n      // Strip CLAUDE-prefixed env vars to avoid nested session detection\n      const cleanEnv = Object.fromEntries(\n        Object.entries(process.env).filter(([key]) => !key.startsWith('CLAUDE'))\n      );\n\n      const proc = spawn(claudePath, args, {\n        stdio: ['pipe', 'pipe', 'pipe'],\n        cwd: process.cwd(),\n        env: cleanEnv,\n      });\n\n      let stdout = '';\n      let stderr = '';\n      let timeoutId: NodeJS.Timeout | undefined;\n\n      // Set timeout\n      if (this.timeout > 0) {\n        timeoutId = setTimeout(() => {\n          proc.kill('SIGTERM');\n          reject(new Error(`Claude CLI timed out after ${String(this.timeout)}ms`));\n        }, this.timeout);\n      }\n\n      proc.stdout.on('data', (chunk: Buffer) => {\n        stdout += chunk.toString();\n      });\n\n      proc.stderr.on('data', (chunk: Buffer) => {\n        stderr += chunk.toString();\n      });\n\n      proc.on('close', (code: number | null) => {\n        if (timeoutId !== undefined) {\n          clearTimeout(timeoutId);\n        }\n\n        if (code === 0) {\n          resolve(stdout.trim());\n        } else {\n          reject(\n            new Error(`Claude CLI exited with code ${String(code)}${stderr ? `: ${stderr}` : ''}`)\n          );\n        }\n      });\n\n      proc.on('error', (err) => {\n        if (timeoutId !== undefined) {\n          clearTimeout(timeoutId);\n        }\n        reject(new Error(`Failed to spawn Claude CLI: ${err.message}`));\n      });\n\n      // Write prompt to stdin\n      proc.stdin.write(prompt);\n      proc.stdin.end();\n    });\n  }\n\n  /**\n   * Truncate HTML to a maximum length (keep important parts)\n   */\n  private truncateHtml(html: string, maxLength: number): string {\n    if (html.length <= maxLength) return html;\n\n    // Try to keep the beginning (usually has navigation)\n    return `${html.substring(0, maxLength)}\\n\\n[... HTML truncated ...]`;\n  }\n\n  /**\n   * Truncate markdown to a maximum length\n   */\n  private truncateMarkdown(markdown: string, maxLength: number): string {\n    if (markdown.length <= maxLength) return markdown;\n\n    return `${markdown.substring(0, maxLength)}\\n\\n[... content truncated ...]`;\n  }\n\n  /**\n   * Type guard to check if value is a record (plain object)\n   */\n  private isRecord(value: unknown): value is Record<string, unknown> {\n    return typeof value === 'object' && value !== null && !Array.isArray(value);\n  }\n\n  /**\n   * Extract structured_output from Claude CLI wrapper format if present.\n   * Claude CLI with --json-schema returns: {type, result, structured_output: {...}}\n   * This method extracts the inner structured_output, or returns the raw value if not wrapped.\n   */\n  private extractStructuredOutput(rawParsed: unknown): unknown {\n    if (this.isRecord(rawParsed) && 'structured_output' in rawParsed) {\n      const structuredOutput = rawParsed['structured_output'];\n      if (typeof structuredOutput === 'object') {\n        return structuredOutput;\n      }\n    }\n    return rawParsed;\n  }\n}\n","/**\n * Node.js Playwright-based web crawler\n * Replaces crawl4ai Python dependency for headless browser crawling\n */\n\nimport { chromium, type Browser, type Page } from 'playwright';\nimport { createLogger, summarizePayload } from '../logging/index.js';\n\nconst logger = createLogger('playwright-crawler');\n\n// Default timeout for page navigation\nconst DEFAULT_TIMEOUT = 30000;\n\n// User agent to identify our crawler\nconst USER_AGENT =\n  'Mozilla/5.0 (compatible; BlueraKnowledge/1.0; +https://github.com/blueraai/bluera-knowledge)';\n\nexport interface PlaywrightFetchResult {\n  html: string;\n  links: string[];\n}\n\n/**\n * Singleton browser instance for connection reuse\n */\nlet browserInstance: Browser | null = null;\n\n/**\n * Get or create the browser instance\n */\nasync function getBrowser(): Promise<Browser> {\n  if (browserInstance?.isConnected() !== true) {\n    logger.debug('Launching new Chromium browser instance');\n    browserInstance = await chromium.launch({\n      headless: true,\n    });\n  }\n  return browserInstance;\n}\n\n/**\n * Close the browser instance (call on cleanup)\n */\nexport async function closeBrowser(): Promise<void> {\n  if (browserInstance?.isConnected() === true) {\n    await browserInstance.close();\n    browserInstance = null;\n  }\n}\n\n/**\n * Fetch a URL using Playwright headless browser\n * Returns the rendered HTML and discovered links\n */\nexport async function fetchWithPlaywright(\n  url: string,\n  timeoutMs: number = DEFAULT_TIMEOUT\n): Promise<PlaywrightFetchResult> {\n  const startTime = Date.now();\n  logger.debug({ url, timeoutMs }, 'Fetching page with Playwright');\n\n  const browser = await getBrowser();\n  const context = await browser.newContext({\n    userAgent: USER_AGENT,\n  });\n\n  let page: Page | null = null;\n\n  try {\n    page = await context.newPage();\n\n    // Navigate to the URL and wait for network idle\n    await page.goto(url, {\n      waitUntil: 'networkidle',\n      timeout: timeoutMs,\n    });\n\n    // Get the full rendered HTML\n    const html = await page.content();\n\n    // Extract all links from the page\n    const links = await page.evaluate(() => {\n      const anchors = document.querySelectorAll('a[href]');\n      const hrefs: string[] = [];\n\n      anchors.forEach((anchor) => {\n        const href = anchor.getAttribute('href');\n        if (href !== null && href !== '') {\n          try {\n            // Resolve relative URLs to absolute\n            const absoluteUrl = new URL(href, document.baseURI).href;\n            // Only include http/https links\n            if (absoluteUrl.startsWith('http://') || absoluteUrl.startsWith('https://')) {\n              hrefs.push(absoluteUrl);\n            }\n          } catch {\n            // Skip invalid URLs\n          }\n        }\n      });\n\n      // Deduplicate\n      return [...new Set(hrefs)];\n    });\n\n    const durationMs = Date.now() - startTime;\n    logger.info(\n      {\n        url,\n        durationMs,\n        linkCount: links.length,\n        ...summarizePayload(html, 'raw-html', url),\n      },\n      'Page fetched with Playwright'\n    );\n\n    return { html, links };\n  } finally {\n    // Clean up the context (closes the page too)\n    await context.close();\n  }\n}\n\n/**\n * Check if Playwright browsers are installed\n * Fast check without launching a browser\n */\nexport function isPlaywrightInstalled(): boolean {\n  try {\n    // Try to get the chromium executable path\n    // This will throw if browsers aren't installed\n    const execPath = chromium.executablePath();\n    logger.debug({ execPath }, 'Playwright Chromium executable found');\n    return true;\n  } catch {\n    return false;\n  }\n}\n"],"mappings":";;;;;;;AAIA,SAAS,oBAAoB;AAC7B,OAAO,WAAW;;;ACAlB,SAAS,uBAAuB;AAChC,OAAO,qBAAqB;AAC5B,SAAS,WAAW;;;ACCpB,YAAY,aAAa;AAMzB,SAAS,wBAAwB,WAAuC;AACtE,MAAI,cAAc,UAAa,cAAc,GAAI,QAAO;AAGxD,QAAM,WAAW;AAAA,IACf;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF;AAEA,aAAW,WAAW,UAAU;AAC9B,UAAM,QAAQ,UAAU,MAAM,OAAO;AACrC,QAAI,QAAQ,CAAC,MAAM,QAAW;AAC5B,YAAM,OAAO,MAAM,CAAC,EAAE,YAAY;AAElC,UAAI,CAAC,CAAC,QAAQ,aAAa,QAAQ,OAAO,SAAS,QAAQ,EAAE,SAAS,IAAI,GAAG;AAC3E,eAAO;AAAA,MACT;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AACT;AAKA,SAAS,WAAW,MAAsB;AACxC,SAAO,KACJ,QAAQ,MAAM,OAAO,EACrB,QAAQ,MAAM,MAAM,EACpB,QAAQ,MAAM,MAAM,EACpB,QAAQ,MAAM,QAAQ,EACtB,QAAQ,MAAM,QAAQ;AAC3B;AAWO,SAAS,4BAA4B,MAAsB;AAChE,MAAI,CAAC,QAAQ,OAAO,SAAS,SAAU,QAAO;AAE9C,QAAM,IAAY,aAAK,IAAI;AAG3B,IAAE,OAAO,EAAE,KAAK,CAAC,IAAI,UAAU;AAC7B,UAAM,SAAS,EAAE,KAAK;AAGtB,UAAM,YAAY,OAAO,KAAK,8BAA8B;AAE5D,QAAI,UAAU,SAAS,GAAG;AAExB,YAAM,OAAO,UAAU,QAAQ,KAAK;AACpC,YAAM,QAAQ,UAAU,MAAM;AAG9B,UAAI,WAAW,wBAAwB,MAAM,KAAK,OAAO,CAAC;AAC1D,UAAI,CAAC,UAAU;AACb,mBAAW,wBAAwB,KAAK,KAAK,OAAO,CAAC;AAAA,MACvD;AAGA,YAAM,WAAW,MAAM,KAAK;AAG5B,YAAM,WAAW,8BAA8B,QAAQ,KAAK,WAAW,QAAQ,CAAC;AAGhF,aAAO,YAAY,QAAQ;AAAA,IAC7B;AAAA,EACF,CAAC;AAGD,IAAE,eAAe,EAAE,KAAK,CAAC,IAAI,WAAW;AACtC,UAAM,UAAU,EAAE,MAAM;AACxB,QAAI,CAAC,QAAQ,KAAK,EAAE,KAAK,GAAG;AAC1B,cAAQ,OAAO;AAAA,IACjB;AAAA,EACF,CAAC;AAGD,IAAE,qBAAqB,EAAE,KAAK,CAAC,IAAI,SAAS;AAC1C,UAAM,QAAQ,EAAE,IAAI;AACpB,UAAM,YAAY,MAAM,KAAK,CAAC;AAAA,EAChC,CAAC;AAGD,IAAE,KAAK,EAAE,KAAK,CAAC,IAAI,QAAQ;AACzB,UAAM,OAAO,EAAE,GAAG;AAElB,QAAI,KAAK,KAAK,MAAM,EAAE,WAAW,GAAG;AAElC,YAAM,OAAO,KAAK,KAAK;AACvB,YAAM,OAAO,wBAAwB,KAAK,KAAK,OAAO,CAAC;AACvD,WAAK,KAAK,yBAAyB,IAAI,KAAK,WAAW,IAAI,CAAC,SAAS;AAAA,IACvE;AAAA,EACF,CAAC;AAED,SAAO,EAAE,KAAK;AAChB;AAWO,SAAS,gBAAgB,UAA0B;AACxD,MAAI,CAAC,SAAU,QAAO;AAEtB,QAAM,UAAU,SAAS,KAAK;AAC9B,MAAI,YAAY,GAAI,QAAO;AAE3B,MAAI,SAAS;AAIb,WAAS,OAAO,QAAQ,iCAAiC,OAAO;AAIhE,WAAS,OAAO,QAAQ,mBAAmB,KAAK;AAGhD,WAAS,OAAO,QAAQ,wCAAwC,YAAY;AAK5E,WAAS,OAAO,QAAQ,sBAAsB,QAAQ;AAGtD,WAAS,OAAO,QAAQ,0BAA0B,UAAU;AAG5D,WAAS,OAAO,QAAQ,+BAA+B,UAAU;AAGjE,WAAS,OAAO,QAAQ,iCAAiC,UAAU;AAGnE,WAAS,OAAO,QAAQ,gDAAgD,YAAY;AAGpF,WAAS,OAAO,QAAQ,gCAAgC,QAAQ;AAGhE,WAAS,OAAO,QAAQ,WAAW,MAAM;AAGzC,WAAS,OAAO,QAAQ,qBAAqB,MAAM;AACnD,WAAS,OAAO,QAAQ,aAAa,OAAO;AAG5C,WAAS,OAAO,QAAQ,iBAAiB,GAAG;AAI5C,WAAS,OAAO,QAAQ,qBAAqB,EAAE;AAC/C,WAAS,OAAO,QAAQ,qBAAqB,EAAE;AAC/C,WAAS,OAAO,QAAQ,qBAAqB,EAAE;AAC/C,WAAS,OAAO,QAAQ,kBAAkB,EAAE;AAC5C,WAAS,OAAO,QAAQ,kBAAkB,EAAE;AAC5C,WAAS,OAAO,QAAQ,kBAAkB,EAAE;AAG5C,WAAS,OAAO,QAAQ,mBAAmB,EAAE;AAG7C,WAAS,OAAO,QAAQ,oBAAoB,EAAE;AAG9C,WAAS,OAAO,QAAQ,mBAAmB,EAAE;AAG7C,WAAS,OAAO,QAAQ,mBAAmB,EAAE;AAC7C,WAAS,OAAO,QAAQ,oBAAoB,EAAE;AAG9C,WAAS,OAAO,QAAQ,kBAAkB,EAAE;AAI5C,WAAS,OAAO,QAAQ,qCAAqC,EAAE;AAG/D,WAAS,OAAO,QAAQ,uCAAuC,EAAE;AAGjE,WAAS,OAAO,QAAQ,aAAa,MAAM;AAC3C,WAAS,OAAO,QAAQ,aAAa,MAAM;AAC3C,WAAS,OAAO,QAAQ,cAAc,OAAO;AAG7C,WAAS,OAAO,QAAQ,WAAW,MAAM;AACzC,WAAS,OAAO,QAAQ,aAAa,IAAI;AAEzC,SAAO;AACT;;;ADrNA,IAAM,SAAS,aAAa,mBAAmB;AAgB/C,eAAsB,sBAAsB,MAAc,KAAwC;AAChG,SAAO,MAAM,EAAE,KAAK,YAAY,KAAK,OAAO,GAAG,0BAA0B;AAEzE,MAAI;AAEF,QAAI;AACJ,QAAI;AAEJ,QAAI;AACF,YAAM,UAAU,MAAM,gBAAgB,MAAM,GAAG;AAC/C,UAAI,SAAS,YAAY,UAAa,QAAQ,YAAY,IAAI;AAC5D,sBAAc,QAAQ;AACtB,gBAAQ,QAAQ,UAAU,UAAa,QAAQ,UAAU,KAAK,QAAQ,QAAQ;AAC9E,eAAO;AAAA,UACL;AAAA,YACE;AAAA,YACA;AAAA,YACA,iBAAiB,YAAY;AAAA,YAC7B,cAAc;AAAA,UAChB;AAAA,UACA;AAAA,QACF;AAAA,MACF,OAAO;AAEL,sBAAc;AACd,eAAO;AAAA,UACL,EAAE,KAAK,cAAc,KAAK;AAAA,UAC1B;AAAA,QACF;AAAA,MACF;AAAA,IACF,SAAS,cAAc;AAErB,oBAAc;AACd,aAAO;AAAA,QACL;AAAA,UACE;AAAA,UACA,cAAc;AAAA,UACd,OAAO,wBAAwB,QAAQ,aAAa,UAAU,OAAO,YAAY;AAAA,QACnF;AAAA,QACA;AAAA,MACF;AAAA,IACF;AAGA,UAAM,eAAe,4BAA4B,WAAW;AAG5D,UAAM,kBAAkB,IAAI,gBAAgB;AAAA,MAC1C,cAAc;AAAA;AAAA,MACd,gBAAgB;AAAA;AAAA,MAChB,OAAO;AAAA,MACP,aAAa;AAAA,MACb,iBAAiB;AAAA,MACjB,WAAW;AAAA,IACb,CAAC;AAGD,oBAAgB,IAAI,GAAG;AAGvB,oBAAgB,QAAQ,uBAAuB;AAAA,MAC7C,QAAQ,CAAC,MAAM,MAAM,MAAM,MAAM,MAAM,IAAI;AAAA,MAC3C,YAAY,SAAiB,MAA2B;AACtD,cAAM,QAAQ,OAAO,KAAK,SAAS,OAAO,CAAC,CAAC;AAC5C,cAAM,SAAS,IAAI,OAAO,KAAK;AAC/B,cAAM,eAAe,QAClB,QAAQ,kBAAkB,EAAE,EAC5B,QAAQ,QAAQ,GAAG,EACnB,KAAK;AACR,eAAO,iBAAiB,KAAK;AAAA;AAAA,EAAO,MAAM,IAAI,YAAY;AAAA;AAAA,IAAS;AAAA,MACrE;AAAA,IACF,CAAC;AAGD,UAAM,cAAc,gBAAgB,SAAS,YAAY;AAGzD,UAAM,WAAW,gBAAgB,WAAW;AAE5C,WAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA;AAAA,QACA,mBAAmB,YAAY;AAAA,QAC/B,qBAAqB,SAAS;AAAA,MAChC;AAAA,MACA;AAAA,IACF;AAGA,WAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA,iBAAiB,eAAe,UAAU,GAAI;AAAA,MAChD;AAAA,MACA;AAAA,IACF;AAEA,WAAO;AAAA,MACL;AAAA,MACA,GAAI,UAAU,UAAa,EAAE,MAAM;AAAA,IACrC;AAAA,EACF,SAAS,OAAO;AACd,WAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK;AAAA,MAC9D;AAAA,MACA;AAAA,IACF;AAGA,UAAM,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;AAAA,EAChE;AACF;;;AExIA,SAAS,OAAO,gBAAgB;AAChC,SAAS,kBAAkB;AAC3B,SAAS,eAAe;AACxB,SAAS,YAAY;AAUrB,IAAM,wBAAwB;AAAA,EAC5B,MAAM;AAAA,EACN,YAAY;AAAA,IACV,MAAM;AAAA,MACJ,MAAM;AAAA,MACN,OAAO,EAAE,MAAM,SAAS;AAAA,MACxB,aAAa;AAAA,IACf;AAAA,IACA,WAAW;AAAA,MACT,MAAM;AAAA,MACN,aAAa;AAAA,IACf;AAAA,EACF;AAAA,EACA,UAAU,CAAC,QAAQ,WAAW;AAChC;AAKO,IAAM,eAAN,MAAM,cAAa;AAAA,EACP;AAAA,EACjB,OAAe,sBAAsB;AAAA,EACrC,OAAe,YAAY;AAAA,EAC3B,OAAe,aAA4B;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAU3C,OAAO,gBAA+B;AAEpC,UAAM,UAAU,QAAQ,IAAI,YAAY;AACxC,QAAI,YAAY,UAAa,YAAY,MAAM,WAAW,OAAO,GAAG;AAClE,aAAO;AAAA,IACT;AAGA,UAAM,kBAAkB,KAAK,QAAQ,GAAG,WAAW,SAAS,QAAQ;AACpE,QAAI,WAAW,eAAe,GAAG;AAC/B,aAAO;AAAA,IACT;AAGA,UAAM,eAAe,KAAK,QAAQ,GAAG,UAAU,OAAO,QAAQ;AAC9D,QAAI,WAAW,YAAY,GAAG;AAC5B,aAAO;AAAA,IACT;AAGA,QAAI;AACF,YAAM,SAAS,SAAS,qBAAqB,EAAE,OAAO,CAAC,QAAQ,QAAQ,QAAQ,EAAE,CAAC;AAClF,YAAM,OAAO,OAAO,SAAS,EAAE,KAAK;AACpC,UAAI,MAAM;AACR,eAAO;AAAA,MACT;AAAA,IACF,QAAQ;AAAA,IAER;AAEA,WAAO;AAAA,EACT;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,OAAO,cAAuB;AAC5B,QAAI,CAAC,cAAa,qBAAqB;AACrC,oBAAa,aAAa,cAAa,cAAc;AACrD,oBAAa,YAAY,cAAa,eAAe;AACrD,oBAAa,sBAAsB;AAAA,IACrC;AACA,WAAO,cAAa;AAAA,EACtB;AAAA;AAAA;AAAA;AAAA,EAKA,OAAO,gBAA+B;AACpC,WAAO,cAAa;AAAA,EACtB;AAAA;AAAA;AAAA;AAAA,EAKA,OAAO,yBAA+B;AACpC,kBAAa,sBAAsB;AACnC,kBAAa,YAAY;AAAA,EAC3B;AAAA,EAEA,YAAY,UAAgC,CAAC,GAAG;AAC9C,SAAK,UAAU,QAAQ,WAAW;AAAA,EACpC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUA,MAAM,mBACJ,SACA,UACA,aACwB;AACxB,UAAM,SAAS;AAAA;AAAA,YAEP,OAAO;AAAA;AAAA,eAEJ,WAAW;AAAA;AAAA;AAAA,EAGxB,KAAK,aAAa,UAAU,GAAK,CAAC;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAQhC,QAAI;AACF,YAAM,SAAS,MAAM,KAAK,WAAW,QAAQ,qBAAqB;AAClE,YAAM,YAAqB,KAAK,MAAM,MAAM;AAI5C,YAAM,SAAS,KAAK,wBAAwB,SAAS;AAGrD,UACE,OAAO,WAAW,YAClB,WAAW,QACX,EAAE,UAAU,WACZ,EAAE,eAAe,WACjB,CAAC,MAAM,QAAQ,OAAO,IAAI,KAC1B,OAAO,KAAK,WAAW,KACvB,OAAO,OAAO,cAAc,YAC5B,CAAC,OAAO,KAAK,MAAM,CAAC,QAAQ,OAAO,QAAQ,QAAQ,GACnD;AACA,cAAM,IAAI,MAAM,wCAAwC;AAAA,MAC1D;AAGA,aAAO,EAAE,MAAM,OAAO,MAAM,WAAW,OAAO,UAAU;AAAA,IAC1D,SAAS,OAAO;AACd,YAAM,IAAI;AAAA,QACR,uCAAuC,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,CAAC;AAAA,MAC/F;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASA,MAAM,eAAe,UAAkB,aAAsC;AAC3E,UAAM,SAAS,GAAG,WAAW;AAAA;AAAA;AAAA,EAG/B,KAAK,iBAAiB,UAAU,GAAM,CAAC;AAErC,QAAI;AACF,YAAM,SAAS,MAAM,KAAK,WAAW,MAAM;AAC3C,aAAO,OAAO,KAAK;AAAA,IACrB,SAAS,OAAO;AACd,YAAM,IAAI;AAAA,QACR,8BAA8B,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,CAAC;AAAA,MACtF;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EASA,MAAc,WAAW,QAAgB,YAAuD;AAC9F,WAAO,IAAI,QAAgB,CAAC,SAAS,WAAW;AAE9C,YAAM,aAAa,cAAa,cAAc;AAC9C,UAAI,eAAe,MAAM;AACvB,eAAO,IAAI,MAAM,0BAA0B,CAAC;AAC5C;AAAA,MACF;AAEA,YAAM,OAAO,CAAC,IAAI;AAGlB,UAAI,YAAY;AACd,aAAK,KAAK,iBAAiB,KAAK,UAAU,UAAU,CAAC;AACrD,aAAK,KAAK,mBAAmB,MAAM;AAAA,MACrC;AAGA,YAAM,WAAW,OAAO;AAAA,QACtB,OAAO,QAAQ,QAAQ,GAAG,EAAE,OAAO,CAAC,CAAC,GAAG,MAAM,CAAC,IAAI,WAAW,QAAQ,CAAC;AAAA,MACzE;AAEA,YAAM,OAAO,MAAM,YAAY,MAAM;AAAA,QACnC,OAAO,CAAC,QAAQ,QAAQ,MAAM;AAAA,QAC9B,KAAK,QAAQ,IAAI;AAAA,QACjB,KAAK;AAAA,MACP,CAAC;AAED,UAAI,SAAS;AACb,UAAI,SAAS;AACb,UAAI;AAGJ,UAAI,KAAK,UAAU,GAAG;AACpB,oBAAY,WAAW,MAAM;AAC3B,eAAK,KAAK,SAAS;AACnB,iBAAO,IAAI,MAAM,8BAA8B,OAAO,KAAK,OAAO,CAAC,IAAI,CAAC;AAAA,QAC1E,GAAG,KAAK,OAAO;AAAA,MACjB;AAEA,WAAK,OAAO,GAAG,QAAQ,CAAC,UAAkB;AACxC,kBAAU,MAAM,SAAS;AAAA,MAC3B,CAAC;AAED,WAAK,OAAO,GAAG,QAAQ,CAAC,UAAkB;AACxC,kBAAU,MAAM,SAAS;AAAA,MAC3B,CAAC;AAED,WAAK,GAAG,SAAS,CAAC,SAAwB;AACxC,YAAI,cAAc,QAAW;AAC3B,uBAAa,SAAS;AAAA,QACxB;AAEA,YAAI,SAAS,GAAG;AACd,kBAAQ,OAAO,KAAK,CAAC;AAAA,QACvB,OAAO;AACL;AAAA,YACE,IAAI,MAAM,+BAA+B,OAAO,IAAI,CAAC,GAAG,SAAS,KAAK,MAAM,KAAK,EAAE,EAAE;AAAA,UACvF;AAAA,QACF;AAAA,MACF,CAAC;AAED,WAAK,GAAG,SAAS,CAAC,QAAQ;AACxB,YAAI,cAAc,QAAW;AAC3B,uBAAa,SAAS;AAAA,QACxB;AACA,eAAO,IAAI,MAAM,+BAA+B,IAAI,OAAO,EAAE,CAAC;AAAA,MAChE,CAAC;AAGD,WAAK,MAAM,MAAM,MAAM;AACvB,WAAK,MAAM,IAAI;AAAA,IACjB,CAAC;AAAA,EACH;AAAA;AAAA;AAAA;AAAA,EAKQ,aAAa,MAAc,WAA2B;AAC5D,QAAI,KAAK,UAAU,UAAW,QAAO;AAGrC,WAAO,GAAG,KAAK,UAAU,GAAG,SAAS,CAAC;AAAA;AAAA;AAAA,EACxC;AAAA;AAAA;AAAA;AAAA,EAKQ,iBAAiB,UAAkB,WAA2B;AACpE,QAAI,SAAS,UAAU,UAAW,QAAO;AAEzC,WAAO,GAAG,SAAS,UAAU,GAAG,SAAS,CAAC;AAAA;AAAA;AAAA,EAC5C;AAAA;AAAA;AAAA;AAAA,EAKQ,SAAS,OAAkD;AACjE,WAAO,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAAA,EAC5E;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOQ,wBAAwB,WAA6B;AAC3D,QAAI,KAAK,SAAS,SAAS,KAAK,uBAAuB,WAAW;AAChE,YAAM,mBAAmB,UAAU,mBAAmB;AACtD,UAAI,OAAO,qBAAqB,UAAU;AACxC,eAAO;AAAA,MACT;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACF;;;AC3TA,SAAS,gBAAyC;AAGlD,IAAMA,UAAS,aAAa,oBAAoB;AAGhD,IAAM,kBAAkB;AAGxB,IAAM,aACJ;AAUF,IAAI,kBAAkC;AAKtC,eAAe,aAA+B;AAC5C,MAAI,iBAAiB,YAAY,MAAM,MAAM;AAC3C,IAAAA,QAAO,MAAM,yCAAyC;AACtD,sBAAkB,MAAM,SAAS,OAAO;AAAA,MACtC,UAAU;AAAA,IACZ,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAKA,eAAsB,eAA8B;AAClD,MAAI,iBAAiB,YAAY,MAAM,MAAM;AAC3C,UAAM,gBAAgB,MAAM;AAC5B,sBAAkB;AAAA,EACpB;AACF;AAMA,eAAsB,oBACpB,KACA,YAAoB,iBACY;AAChC,QAAM,YAAY,KAAK,IAAI;AAC3B,EAAAA,QAAO,MAAM,EAAE,KAAK,UAAU,GAAG,+BAA+B;AAEhE,QAAM,UAAU,MAAM,WAAW;AACjC,QAAM,UAAU,MAAM,QAAQ,WAAW;AAAA,IACvC,WAAW;AAAA,EACb,CAAC;AAED,MAAI,OAAoB;AAExB,MAAI;AACF,WAAO,MAAM,QAAQ,QAAQ;AAG7B,UAAM,KAAK,KAAK,KAAK;AAAA,MACnB,WAAW;AAAA,MACX,SAAS;AAAA,IACX,CAAC;AAGD,UAAM,OAAO,MAAM,KAAK,QAAQ;AAGhC,UAAM,QAAQ,MAAM,KAAK,SAAS,MAAM;AACtC,YAAM,UAAU,SAAS,iBAAiB,SAAS;AACnD,YAAM,QAAkB,CAAC;AAEzB,cAAQ,QAAQ,CAAC,WAAW;AAC1B,cAAM,OAAO,OAAO,aAAa,MAAM;AACvC,YAAI,SAAS,QAAQ,SAAS,IAAI;AAChC,cAAI;AAEF,kBAAM,cAAc,IAAI,IAAI,MAAM,SAAS,OAAO,EAAE;AAEpD,gBAAI,YAAY,WAAW,SAAS,KAAK,YAAY,WAAW,UAAU,GAAG;AAC3E,oBAAM,KAAK,WAAW;AAAA,YACxB;AAAA,UACF,QAAQ;AAAA,UAER;AAAA,QACF;AAAA,MACF,CAAC;AAGD,aAAO,CAAC,GAAG,IAAI,IAAI,KAAK,CAAC;AAAA,IAC3B,CAAC;AAED,UAAM,aAAa,KAAK,IAAI,IAAI;AAChC,IAAAA,QAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA;AAAA,QACA,WAAW,MAAM;AAAA,QACjB,GAAG,iBAAiB,MAAM,YAAY,GAAG;AAAA,MAC3C;AAAA,MACA;AAAA,IACF;AAEA,WAAO,EAAE,MAAM,MAAM;AAAA,EACvB,UAAE;AAEA,UAAM,QAAQ,MAAM;AAAA,EACtB;AACF;;;AJ9GA,IAAMC,UAAS,aAAa,SAAS;AASrC,eAAsB,iBACpB,SACA,kBACA,cAAuB,MACC;AACxB,MAAI,CAAC,aAAa,YAAY,GAAG;AAC/B,UAAM,IAAI,MAAM,wEAAwE;AAAA,EAC1F;AAEA,QAAM,SAAS,IAAI,aAAa;AAGhC,MAAI;AACJ,MAAI,aAAa;AACf,UAAM,iBAAiB,MAAM,oBAAoB,OAAO;AACxD,eAAW,eAAe;AAAA,EAC5B,OAAO;AACL,UAAM,WAAW,MAAM,MAAM,IAAI,SAAS;AAAA,MACxC,SAAS;AAAA,MACT,SAAS;AAAA,QACP,cACE;AAAA,MACJ;AAAA,IACF,CAAC;AACD,eAAW,OAAO,SAAS,SAAS,WAAW,SAAS,OAAO,KAAK,UAAU,SAAS,IAAI;AAAA,EAC7F;AAGA,SAAO,OAAO,mBAAmB,SAAS,UAAU,gBAAgB;AACtE;AAmCA,IAAM,qBACJ;AACF,IAAMC,mBAAkB;AAKjB,IAAM,qBAAN,cAAiC,aAAa;AAAA,EAClC;AAAA,EACA;AAAA,EACA;AAAA,EACT;AAAA,EAER,YAAY,QAAsB;AAChC,UAAM;AACN,SAAK,eAAe,IAAI,aAAa;AACrC,SAAK,UAAU,oBAAI,IAAI;AACvB,SAAK,SAAS,UAAU,CAAC;AACzB,SAAK,UAAU;AAAA,EACjB;AAAA;AAAA;AAAA;AAAA,EAKA,OAAO,MAAM,SAAiB,UAAwB,CAAC,GAA+B;AACpF,UAAM;AAAA,MACJ,mBAAmB;AAAA,MACnB;AAAA,MACA,WAAW;AAAA,IACb,IAAI;AAEJ,SAAK,QAAQ,MAAM;AACnB,SAAK,UAAU;AAEf,IAAAD,QAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA;AAAA,QACA,MAAM;AAAA,QACN,uBAAuB,uBAAuB;AAAA,MAChD;AAAA,MACA;AAAA,IACF;AAEA,UAAM,gBAA+B;AAAA,MACnC,MAAM;AAAA,MACN,cAAc;AAAA,MACd,YAAY;AAAA,IACd;AACA,SAAK,KAAK,YAAY,aAAa;AAEnC,WAAO,KAAK;AAAA,MACV;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,QAAQ,eAAe;AAAA,MACvB,QAAQ;AAAA,IACV;AAEA,IAAAA,QAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA,cAAc,KAAK,QAAQ;AAAA,MAC7B;AAAA,MACA;AAAA,IACF;AAGA,QAAI,KAAK,QAAQ,SAAS,KAAK,WAAW,GAAG;AAC3C,YAAM,kBAAiC;AAAA,QACrC,MAAM;AAAA,QACN,cAAc,KAAK,QAAQ;AAAA,QAC3B,YAAY;AAAA,QACZ,SAAS,iDAAiD,OAAO,QAAQ,CAAC;AAAA,QAC1E,OAAO,IAAI,MAAM,oBAAoB;AAAA,MACvC;AACA,WAAK,KAAK,YAAY,eAAe;AAAA,IACvC;AAEA,UAAM,mBAAkC;AAAA,MACtC,MAAM;AAAA,MACN,cAAc,KAAK,QAAQ;AAAA,MAC3B,YAAY,KAAK,QAAQ;AAAA,IAC3B;AACA,SAAK,KAAK,YAAY,gBAAgB;AAAA,EACxC;AAAA;AAAA;AAAA;AAAA,EAKA,OAAe,iBACb,SACA,kBACA,oBACA,UACA,cAAuB,MACvB,qBAC4B;AAC5B,QAAI;AAGJ,QAAI,wBAAwB,QAAW;AACrC,iBAAW;AACX,YAAM,mBAAkC;AAAA,QACtC,MAAM;AAAA,QACN,cAAc;AAAA,QACd,YAAY;AAAA,QACZ,SAAS,gCAAgC,OAAO,SAAS,KAAK,MAAM,CAAC;AAAA,MACvE;AACA,WAAK,KAAK,YAAY,gBAAgB;AAAA,IACxC,OAAO;AAEL,UAAI,CAAC,aAAa,YAAY,GAAG;AAC/B,cAAM,IAAI,MAAM,wEAAwE;AAAA,MAC1F;AAEA,UAAI;AAEF,cAAM,wBAAuC;AAAA,UAC3C,MAAM;AAAA,UACN,cAAc;AAAA,UACd,YAAY;AAAA,UACZ,YAAY;AAAA,UACZ,SAAS;AAAA,QACX;AACA,aAAK,KAAK,YAAY,qBAAqB;AAE3C,cAAM,WAAW,MAAM,KAAK,UAAU,SAAS,WAAW;AAG1D,mBAAW,MAAM,KAAK,aAAa,mBAAmB,SAAS,UAAU,gBAAgB;AAEzF,cAAM,2BAA0C;AAAA,UAC9C,MAAM;AAAA,UACN,cAAc;AAAA,UACd,YAAY;AAAA,UACZ,SAAS,qBAAqB,OAAO,SAAS,KAAK,MAAM,CAAC,mBAAmB,SAAS,SAAS;AAAA,QACjG;AACA,aAAK,KAAK,YAAY,wBAAwB;AAAA,MAChD,SAAS,OAAO;AAEd,cAAM,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;AAAA,MAChE;AAAA,IACF;AAGA,QAAI,eAAe;AAEnB,eAAW,OAAO,SAAS,MAAM;AAC/B,UAAI,KAAK,WAAW,gBAAgB,SAAU;AAC9C,UAAI,KAAK,QAAQ,IAAI,GAAG,EAAG;AAE3B,UAAI;AACF,cAAM,SAAS,MAAM,KAAK;AAAA,UACxB;AAAA,UACA;AAAA,UACA;AAAA,UACA;AAAA,QACF;AACA;AACA,cAAM;AAAA,MACR,SAAS,OAAO;AACd,cAAM,oBAAmC;AAAA,UACvC,MAAM;AAAA,UACN;AAAA,UACA,YAAY;AAAA,UACZ,YAAY;AAAA,UACZ,OAAO,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;AAAA,QACjE;AACA,aAAK,KAAK,YAAY,iBAAiB;AAAA,MACzC;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAc,gBACZ,KACA,oBACA,cACA,cAAuB,MACD;AACtB,UAAM,eAA8B;AAAA,MAClC,MAAM;AAAA,MACN;AAAA,MACA,YAAY;AAAA,MACZ,YAAY;AAAA,IACd;AACA,SAAK,KAAK,YAAY,YAAY;AAGlC,SAAK,QAAQ,IAAI,GAAG;AAGpB,UAAM,OAAO,MAAM,KAAK,UAAU,KAAK,WAAW;AAIlD,UAAM,aAAa,MAAM,sBAAsB,MAAM,GAAG;AAExD,IAAAA,QAAO;AAAA,MACL;AAAA,QACE;AAAA,QACA,OAAO,WAAW;AAAA,QAClB,gBAAgB,WAAW,SAAS;AAAA,MACtC;AAAA,MACA;AAAA,IACF;AAEA,QAAI;AAGJ,QAAI,uBAAuB,UAAa,uBAAuB,IAAI;AAEjE,UAAI,CAAC,aAAa,YAAY,GAAG;AAC/B,cAAM,IAAI,MAAM,8DAA8D;AAAA,MAChF;AAEA,YAAM,qBAAoC;AAAA,QACxC,MAAM;AAAA,QACN;AAAA,QACA,YAAY;AAAA,QACZ,YAAY;AAAA,MACd;AACA,WAAK,KAAK,YAAY,kBAAkB;AAExC,kBAAY,MAAM,KAAK,aAAa,eAAe,WAAW,UAAU,kBAAkB;AAAA,IAC5F;AAEA,WAAO;AAAA,MACL;AAAA,MACA,GAAI,WAAW,UAAU,UAAa,EAAE,OAAO,WAAW,MAAM;AAAA,MAChE,UAAU,WAAW;AAAA,MACrB,GAAI,cAAc,UAAa,EAAE,UAAU;AAAA,IAC7C;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAc,UAAU,KAAa,cAAuB,MAAuB;AACjF,UAAM,YAAY,KAAK,IAAI;AAC3B,IAAAA,QAAO,MAAM,EAAE,KAAK,YAAY,GAAG,eAAe;AAElD,QAAI,aAAa;AACf,UAAI;AACF,cAAM,SAAS,MAAM,oBAAoB,GAAG;AAC5C,cAAM,aAAa,KAAK,IAAI,IAAI;AAChC,QAAAA,QAAO;AAAA,UACL;AAAA,YACE;AAAA,YACA,aAAa;AAAA,YACb;AAAA,YACA,GAAG,iBAAiB,OAAO,MAAM,YAAY,GAAG;AAAA,UAClD;AAAA,UACA;AAAA,QACF;AACA,eAAO,OAAO;AAAA,MAChB,SAAS,OAAO;AAEd,cAAM,IAAI;AAAA,UACR,0BAA0B,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,CAAC;AAAA,QAClF;AAAA,MACF;AAAA,IACF;AAGA,QAAI;AACF,YAAM,WAAW,MAAM,MAAM,IAAY,KAAK;AAAA,QAC5C,SAAS,KAAK,OAAO,WAAWC;AAAA,QAChC,SAAS;AAAA,UACP,cAAc,KAAK,OAAO,aAAa;AAAA,QACzC;AAAA,MACF,CAAC;AAED,YAAM,aAAa,KAAK,IAAI,IAAI;AAChC,MAAAD,QAAO;AAAA,QACL;AAAA,UACE;AAAA,UACA,aAAa;AAAA,UACb;AAAA,UACA,GAAG,iBAAiB,SAAS,MAAM,YAAY,GAAG;AAAA,QACpD;AAAA,QACA;AAAA,MACF;AAEA,aAAO,SAAS;AAAA,IAClB,SAAS,OAAO;AACd,MAAAA,QAAO;AAAA,QACL,EAAE,KAAK,OAAO,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,EAAE;AAAA,QACrE;AAAA,MACF;AACA,YAAM,IAAI;AAAA,QACR,mBAAmB,GAAG,KAAK,iBAAiB,QAAQ,MAAM,UAAU,OAAO,KAAK,CAAC;AAAA,MACnF;AAAA,IACF;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,MAAM,OAAsB;AAC1B,SAAK,UAAU;AACf,UAAM,aAAa;AAAA,EACrB;AACF;","names":["logger","logger","DEFAULT_TIMEOUT"]}

package/dist/{chunk-3TB7TDVF.js → chunk-FYHKBCIH.js}

@@ -357,7 +357,6 @@ var JobDetailsSchema = z.object({
   crawlInstruction: z.string().optional(),
   extractInstruction: z.string().optional(),
   maxPages: z.number().optional(),
-  simple: z.boolean().optional(),
   useHeadless: z.boolean().optional(),
   pagesCrawled: z.number().optional(),
   // Phase tracking
@@ -4952,6 +4951,32 @@ function mapSearchIntentToQueryIntent(intent) {
       return "how-to";
   }
 }
+var INTENT_EXPANSION_TERMS = {
+  "find-implementation": "source code implementation function class",
+  "find-documentation": "documentation guide tutorial example",
+  "find-usage": "usage example how to use",
+  "find-pattern": "pattern matching code structure",
+  "find-definition": "definition type interface declaration",
+  "find-files": "file module path"
+};
+function expandQueryWithIntent(query, intent) {
+  if (intent === void 0) return query;
+  const expansion = INTENT_EXPANSION_TERMS[intent];
+  return `${query} ${expansion}`;
+}
+function isStrongFtsSignal(query, ftsResults) {
+  if (ftsResults.length < 2) return false;
+  const top = ftsResults[0];
+  const second = ftsResults[1];
+  if (top === void 0 || second === void 0) return false;
+  if (second.score > 0 && top.score / second.score <= 2) return false;
+  const queryLower = query.toLowerCase();
+  const rawFile = top.metadata["file"] ?? top.metadata["path"];
+  const rawName = top.metadata["name"];
+  const filePath = typeof rawFile === "string" ? rawFile : "";
+  const name = typeof rawName === "string" ? rawName : "";
+  return filePath.toLowerCase().includes(queryLower) || name.toLowerCase().includes(queryLower);
+}
 var RRF_PRESETS = {
   code: { k: 25, vectorWeight: 0.75, ftsWeight: 0.25 },
   web: { k: 30, vectorWeight: 0.7, ftsWeight: 0.3 }
@@ -5040,7 +5065,8 @@ var SearchService = class {
     let rerankTimeMs;
     const fetchLimit = limit * 3;
     if (mode === "vector") {
-      const rawResults = await this.vectorSearchRaw(query.query, stores, fetchLimit);
+      const expandedQuery = expandQueryWithIntent(query.query, query.intent);
+      const rawResults = await this.vectorSearchRaw(expandedQuery, stores, fetchLimit);
       maxRawScore = rawResults.length > 0 ? rawResults[0]?.score ?? 0 : 0;
       allResults = this.normalizeAndFilterScores(rawResults, query.threshold).slice(0, fetchLimit);
     } else if (mode === "fts") {
@@ -5050,7 +5076,8 @@ var SearchService = class {
         query.query,
         stores,
         fetchLimit,
-        query.threshold
+        query.threshold,
+        query.intent
       );
       allResults = hybridResult.results;
       maxRawScore = hybridResult.maxRawScore;
@@ -5231,18 +5258,37 @@ var SearchService = class {
   /**
    * Internal hybrid search result with additional metadata for confidence calculation.
    */
-  async hybridSearchWithMetadata(query, stores, limit, threshold) {
+  async hybridSearchWithMetadata(query, stores, limit, threshold, searchIntent) {
     const intents = classifyQueryIntents(query);
     const envOverrides = parseSearchEnvOverrides(false);
     const candidateMultiplier = envOverrides.candidateMultiplier ?? DEFAULT_CANDIDATE_MULTIPLIER;
-    const rawVectorResults = await this.vectorSearchRaw(query, stores, limit * candidateMultiplier);
+    const ftsResults = await this.ftsSearch(query, stores, limit * candidateMultiplier);
+    if (isStrongFtsSignal(query, ftsResults)) {
+      logger4.debug(
+        { query, topScore: ftsResults[0]?.score },
+        "Strong FTS signal \u2014 skipping vector search"
+      );
+      const sorted2 = ftsResults.slice(0, limit).map((r, i) => ({
+        ...r,
+        score: Math.round((1 - i / Math.max(ftsResults.length, 1)) * 1e6) / 1e6
+      }));
+      if (threshold !== void 0) {
+        return { results: sorted2.filter((r) => r.score >= threshold), maxRawScore: 0 };
+      }
+      return { results: sorted2, maxRawScore: 0 };
+    }
+    const expandedQuery = expandQueryWithIntent(query, searchIntent);
+    const rawVectorResults = await this.vectorSearchRaw(
+      expandedQuery,
+      stores,
+      limit * candidateMultiplier
+    );
     const rawVectorScores = /* @__PURE__ */ new Map();
     rawVectorResults.forEach((r) => {
       rawVectorScores.set(r.id, r.score);
     });
     const maxRawScore = rawVectorResults.length > 0 ? rawVectorResults[0]?.score ?? 0 : 0;
     const vectorResults = this.normalizeAndFilterScores(rawVectorResults);
-    const ftsResults = await this.ftsSearch(query, stores, limit * candidateMultiplier);
     const vectorRanks = /* @__PURE__ */ new Map();
     const ftsRanks = /* @__PURE__ */ new Map();
     const allDocs = /* @__PURE__ */ new Map();
@@ -5320,10 +5366,20 @@ var SearchService = class {
       reranked.results.forEach((r) => {
         rerankedScores.set(r.id, r.rerankerScore);
       });
-      sorted = sortedAll.map((r) => ({
-        ...r,
-        rerankerScore: rerankedScores.get(r.id)
-      })).sort((a, b) => (b.rerankerScore ?? -Infinity) - (a.rerankerScore ?? -Infinity)).slice(0, limit);
+      const maxRrfScore = sortedAll[0]?.score ?? 1;
+      sorted = sortedAll.map((r, rrfRank) => {
+        const rerankerScore = rerankedScores.get(r.id);
+        if (rerankerScore === void 0) {
+          return { ...r, blendedScore: -Infinity };
+        }
+        const normalizedRrf = maxRrfScore > 0 ? r.score / maxRrfScore : 0;
+        const rrfWeight = rrfRank < 3 ? 0.7 : rrfRank < 10 ? 0.5 : 0.3;
+        const rerankerWeight = 1 - rrfWeight;
+        return {
+          ...r,
+          blendedScore: normalizedRrf * rrfWeight + rerankerScore * rerankerWeight
+        };
+      }).sort((a, b) => b.blendedScore - a.blendedScore).slice(0, limit);
     } else {
       sorted = sortedAll.slice(0, limit);
     }
@@ -7576,4 +7632,4 @@ export {
   createServices,
   destroyServices
 };
-//# sourceMappingURL=chunk-3TB7TDVF.js.map
+//# sourceMappingURL=chunk-FYHKBCIH.js.map