Search-cluster-skill has 7 tools with basic descriptions and minimal parameter documentation. All tools are READ_ONLY and follow a consistent naming pattern (verb_noun: search_*, ddg_search, wiki_search, reddit_search, news_search, stealth_fetch). However, parameter schemas are extremely sparse, most tools accept only a 'query' string with minimal constraint documentation. No output schemas are documented. Error handling is present (try/except blocks with stderr warnings) but does not guide LLM recovery. The server lacks pagination, result limits are hardcoded (8 for wiki, 5 for reddit), and no structured error messages are returned to the agent. Tool descriptions are adequate (50-150 chars) but lack WHEN/WHY guidance. No tool annotations (readOnlyHint, etc.) are present despite all tools being read-only.

• Document output schema for all tools. Example: { type: 'object', properties: { results: { type: 'array', items: { type: 'object', properties: { source: { type: 'string', enum: ['news', 'ddg', 'google', 'wiki', 'reddit'] }, title: { type: 'string' }, link: { type: 'string', format: 'uri' }, snippet: { type: 'string' } } } } }
• Add 'limit' parameter (default 10, range 1-50) and 'offset' parameter to all search tools. Return total_results count so agents can paginate. Update descriptions to explain when to use each provider (e.g., 'Use news_search for recent news articles; use wiki_search for encyclopedic content').
• Modify error handling to return structured error objects to the agent instead of logging to stderr. Example: { error: 'API_QUOTA_EXCEEDED', message: 'Google Custom Search quota exceeded. Falling back to DuckDuckGo.', retryable: true }
• Add readOnlyHint: true to all tool definitions to signal to agents that these tools are safe to retry and do not modify state.
• Add parameter descriptions explaining query format expectations. Example: 'Search query string (1-500 characters). Supports natural language or boolean operators (AND, OR, NOT). Queries with 5+ words may return fewer results.'