Mixed quality across 6 tools. ask_user and tool_suggest have solid schemas and descriptions (80-90 chars), but tool_search, read, echo_session, and rg lack depth. ask_user is well-structured with timeout handling and custom answer support. tool_search and tool_suggest descriptions are generic (under 100 chars). read lacks parameter descriptions for offset/limit. echo_session is trivial (single string param). rg has good description but no input validation guidance. No tool annotations (readOnlyHint, destructiveHint). Error handling present but not recovery-focused. Schemas visible but descriptions inconsistent across tools.

• Add tool annotations to schema: mark ask_user as idempotent (safe to retry), read/tool_search/tool_suggest/rg as readOnly. This enables agents to reason about retry safety.
• Expand tool_search and tool_suggest descriptions to 150-200 chars. Explain ranking algorithm, when to use each, and what metadata is returned. Include example: 'Returns tools ranked by keyword match. Use tool_search for exact name matches, tool_suggest for task-based discovery.'
• Document read parameter semantics: 'offset: 1-indexed line number (1 = first line). limit: max lines to return. Returns error if offset > file length.'
• Remove or repurpose echo_session. If it's a demo, move to examples/. If it's production, add validation, error handling, and clear use case documentation.
• For rg truncation, return structured output: {truncated: true, tempFile: '/path/to/full-results.txt', lineCount: 2001, byteCount: 51201}. Document how to read the temp file (via read tool).