Three tools with explicit schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. lark_invoke has a complex _confirm parameter that lacks proper validation guidance. lark_discover and lark_get_skill descriptions are adequate (80-120 chars) but lack actionable context for LLM selection. No output schemas documented for any tool. Parameter descriptions exist but are sparse (avg ~40 chars vs baseline 72). Tool naming follows verb_noun pattern (good), but lark_invoke conflates two concerns (shortcut commands vs raw API calls). Error handling and recovery guidance absent.

• Split lark_invoke into two tools: lark_invoke_shortcut (service + command + flags) and lark_invoke_api (service + resource + method + flags). Each tool has one clear responsibility.
• Document output schemas for all three tools. Example for lark_discover: {tools: [{name: string, description: string, schema: object, risk: string}], total: number}. Example for lark_get_skill: {domain: string, section: string, content: string, examples: string[]}.
• Replace flags object with explicit parameter definitions for the most common flags (e.g., limit, offset, format, output). Use enums for constrained values (format: [json, csv, table]). Document that additional flags are passed as key-value pairs.
• Add error handling guidance to each tool description. Example: 'If no tools match the query, try a broader search term or list all tools by omitting the query parameter. If the service is not found, call lark_discover with category filter to see available services.'
• Add _confirm parameter documentation: 'Set to true to confirm destructive operations (delete, update, publish). If false, the tool returns a dry-run preview without executing. Required for any operation that modifies or removes data.'