Sentinel has 5 tools with basic schemas and minimal descriptions. All tools are READ_ONLY filesystem/testing operations. Descriptions are present but generic (10-50 chars), falling below the 50-200 char baseline for LLM-optimized tools. Input schemas exist but lack depth: parameters have types and minimal descriptions. No output schemas documented. No error handling guidance, no parameter constraints (enums, ranges), no composition hints. Tool names follow verb_noun convention (read_file, search_code, list_directory) but descriptions do not explain WHEN to use each tool vs alternatives (e.g., search_code vs search_files distinction is unclear). No tool annotations (readOnlyHint, idempotentHint). No pagination support despite tools that could return large results (list_directory, search_files). No audit trail or permission declarations visible.

• Expand each tool description to 50-200 chars. Example for search_code: 'Search repository code by query string. Supports regex patterns. Returns file paths, line numbers, and matched text snippets. Use this to find implementations; use search_files to locate files by name.'
• Document output schemas for all tools. Example for list_directory: 'Returns {items: [{name: string, type: "file"|"dir", size_bytes?: number}], total: number}'
• Add parameter constraints and examples. Example for search_code query: 'Search query string. Supports regex (e.g., "def test_.*"). Case-sensitive. Max 500 chars.'
• Add pagination to list_directory and search_* tools: accept limit (1-100, default 20) and offset/cursor parameters. Return total count and next_cursor.
• Add tool annotations to schema: set readOnlyHint=true and idempotentHint=true for all tools to signal safety to agents.