Crawlora plugin exposes 17 well-scoped tools with consistent naming (family.action convention) and complete input schemas using TypeBox. All tools have descriptions (avg ~100 chars) and typed parameters. However, descriptions are terse and lack LLM-optimized guidance on WHEN to use each tool, dependencies, or error recovery. Output schemas are not documented, LLMs cannot predict response structure for downstream chaining. No error handling guidance, no parameter constraints beyond enums on 2 tools, and no mention of rate limits or pagination for large result sets. Security: API key is properly injected via config/env, not exposed as parameter. Composition is clean (each tool does one thing), but missing pagination parameters on search tools that likely return many results.

• Document output schemas for all 17 tools. Include field names, types, and examples. E.g., 'Returns array of {url: string, title: string, snippet: string, position: number}' for google.search.
• Expand descriptions to 100 - 200 chars with LLM-optimized guidance: WHAT (action), WHEN (use case vs similar tools), WHAT IF (error recovery). E.g., 'Search Google for organic results. Use when you need current web info; prefer bing.search if Google is blocked. Returns up to 10 results; use pagination for more.'
• Add pagination to search tools: accept 'limit' (1 - 50, default 20) and 'offset' or 'next_cursor'. Document result truncation in descriptions. Cap results at 50 to prevent context explosion.
• Add error handling: wrap API calls in try-catch, return structured errors with recovery hints. E.g., {error: 'ticker_not_found', message: 'INVALID ticker. Try sec.company_search(q="company name") first.', retryable: false}.
• Add parameter constraints: sec.financials limit 1 - 20 (default 5), sec.filings limit 1 - 100 (default 50), jobs.hiring_signals token required for Greenhouse provider. Validate and reject invalid inputs with clear messages.