Method _parseEntry

src/knowledge/loader.js:107–159

* Parse a single knowledge file's metadata without loading its full content. * We still read the file once to extract heading + front-matter — these are * tiny so it's cheap. Full content is loaded later only for selected files.

(filePath, rootDir)

   * tiny so it's cheap. Full content is loaded later only for selected files.
   */
  _parseEntry(filePath, rootDir) {
    let stat;
    try { stat = fs.statSync(filePath); } catch { return null; }
    // Hard cap on file size — we don't want a rogue 50MB file blowing up memory
    if (stat.size > 100 * 1024) return null;

    let head;
    try { head = fs.readFileSync(filePath, 'utf-8'); }
    catch { return null; }

    const rel = path.relative(rootDir, filePath);
    const name = path.basename(filePath, path.extname(filePath));
    // Slug components from path become keywords (binary-search.md → ["binary","search"])
    const pathTokens = rel.toLowerCase().split(/[\\/_\-.\s]+/).filter(Boolean);

    // Optional YAML-ish front-matter: --- ... ---
    let keywords = [];
    let heading = '';
    let body = head;
    const fmMatch = head.match(/^---\n([\s\S]*?)\n---\n([\s\S]*)$/);
    if (fmMatch) {
      const fm = fmMatch[1];
      body = fmMatch[2];
      const kwMatch = fm.match(/keywords?:\s*(.+)/i);
      if (kwMatch) {
        keywords = kwMatch[1]
          .replace(/^\[|\]$/g, '')
          .split(/[,\s]+/)
          .map(s => s.trim().toLowerCase().replace(/^['"]|['"]$/g, ''))
          .filter(Boolean);
      }
    }
    const hMatch = body.match(/^#\s+(.+)/m);
    if (hMatch) heading = hMatch[1].trim();

    // Combine all keyword sources
    const allKeywords = new Set([
      ...pathTokens,
      ...keywords,
      ...heading.toLowerCase().split(/\W+/).filter(t => t.length > 2),
      name.toLowerCase(),
    ]);

    return {
      path: filePath,
      relPath: rel,
      name,
      heading,
      keywords: [...allKeywords],
      size: stat.size,
      _bodyCache: body, // already-read; keep for tiny files
    };
  }

  /**
   * Tokenize a query into normalized lowercase words ≥ 3 chars.