tiguclaw exposes 9 tools with complete input schemas and descriptions in Korean. Naming follows verb_noun patterns (register_command, list_commands, delete_command, etc.). Descriptions are detailed but verbose (200-500+ chars, exceeding the 10-1024 baseline). All parameters have type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict return structures for chaining. Error handling guidance is minimal. Tool descriptions mix implementation details with user intent, reducing clarity. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels (WRITE, DESTRUCTIVE, READ_ONLY). Parameter descriptions are thorough but could be more concise for LLM parsing.

• Document output schemas for all tools. Example: list_commands should return {commands: [{name: string, description: string, prompt: string}], total: number}. This enables LLMs to chain tools and extract the right fields.
• Add tool annotations to schema. Mark register_command, delete_command, register_endpoint, delete_endpoint with destructiveHint: true. Mark list_commands, list_endpoints, list_sessions with readOnlyHint: true. This helps LLMs reason about side effects.
• Condense descriptions to 50-200 chars. Current register_endpoint description is 400+ chars. Shorten to: 'Create a custom HTTP endpoint that runs a prompt template. Supports GET/POST with query/body placeholders. Requires explicit user consent for full mode (external sends, file writes).' Move detailed rules to separate documentation.
• Add error recovery hints. For delete_command, add: 'If deletion fails, call list_commands() to verify the command exists and is not built-in.' For register_endpoint, add: 'If registration fails due to path conflict, call list_endpoints() to see existing routes.'
• Provide English descriptions alongside Korean. LLMs trained primarily on English may misparse Korean parameter descriptions. Add 'description_en' fields or translate all descriptions to English for broader LLM compatibility.