Two tools with minimal documentation. process_cv_pdf has a basic description (67 chars) but no visible input schema in the provided code, only a parameter name 'file_path' with a brief description. get_official_website_and_generate_job_urls has a French description (92 chars) and a single string parameter. Neither tool shows structured output schemas, error handling guidance, or parameter validation rules. Tool names are verb-noun but descriptions lack WHEN/WHY context. No evidence of pagination, result limits, or recovery guidance for failures.

• Define explicit JSON Schema for both tools with type constraints. For process_cv_pdf: accept file_path (string, required, path pattern), optional language_hint (enum: 'fr', 'en', 'auto'). For get_official_website_and_generate_job_urls: company_name (string, required, 1-200 chars).
• Document output schemas. process_cv_pdf should return {corrected_text: string, detected_language: string, confidence: number, stats: {pages: int, words: int}}. get_official_website_and_generate_job_urls should return {website_url: string, job_urls: [string], source: string}.
• Split get_official_website_and_generate_job_urls into two tools: search_company_website(company_name) → {url, confidence} and generate_job_posting_urls(website_url) → {urls: [string]}. This enables independent reuse and clearer composition.
• Expand descriptions to 100-150 chars with WHEN/WHY context. E.g., 'Extract and correct text from a CV PDF using OCR and language-aware spell-checking. Use this when you need to parse unstructured CV files before analysis. Returns corrected text, detected language, and quality metrics.'
• Add error handling guidance. Document that process_cv_pdf may fail on encrypted PDFs or scanned images with poor quality, and suggest fallback to manual review. For get_official_website_and_generate_job_urls, document that DuckDuckGo may not find obscure companies and suggest alternative search strategies.