Two tools with clear verb-noun naming (search, fetch_content) and basic descriptions. Both have input schemas with type definitions and parameter descriptions. However, descriptions are minimal (under 100 chars), lack context on when to use each tool, and output schemas are not documented. No error handling guidance, no pagination support for search results, and no tool annotations. Schemas are present but sparse, parameters lack constraints (e.g., max_results has no bounds documented). Missing idempotency hints and recovery guidance.

• Expand tool descriptions to 80-150 chars. Example: 'search: Search DuckDuckGo for web results. Returns up to max_results ranked by relevance. Use this to find URLs, then fetch_content to extract text. Supports safe_search and region filters.' This clarifies intent, output format, and relationship to fetch_content.
• Document output schemas in code comments or a separate schema file. search() should return {results: [{title, link, snippet, position}], total_count, has_more}. fetch_content() should return {content: string, url, extracted_length, truncated}. LLMs need this to plan downstream calls.
• Add parameter constraints: max_results (1-50, default 10), start_index (0+, default 0), max_length (1-50000, default 10000). Document in descriptions: 'max_results: Maximum results to return (1-50, default 10). Capped to prevent context window exhaustion.'
• Implement pagination for search(). Add offset or next_cursor parameter. Return total_count and has_more flag so agents know when to fetch the next page. Document: 'Call with offset=10 to fetch results 11-20.'
• Add error handling with recovery hints. Catch HTTP errors, timeouts, and rate limits. Return structured errors: {error: 'rate_limited', retry_after_seconds: 60, message: 'DuckDuckGo rate limit hit. Retry after 60 seconds.'}. This guides LLM retry logic.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry