Single tool with solid naming (ask_antigravity) and comprehensive description (287 chars, within baseline 34-392 range). Input schema is well-structured with three parameters, all typed and described. However, output schema is completely undocumented, the tool returns {success, text} but LLMs have no formal specification of this structure. No error handling guidance (e.g., what to do if agy binary is missing or times out). No parameter constraints (e.g., timeout format validation, path existence checks). Tool description is strong but lacks explicit mention of state-modifying behavior or retry safety.

• Document output schema explicitly in tool definition: {type: 'object', properties: {success: {type: 'boolean', description: '...'}, text: {type: 'string', description: '...'}}}. This enables LLMs to understand response structure and plan chaining.
• Add structured error responses with recovery guidance. Instead of returning {success: false, text: '...'}, return {success: false, error_code: 'AGY_NOT_FOUND', message: '...', recovery_hint: 'Install with: curl -fsSL https://antigravity.google/cli/install.sh | bash'}.
• Add enum constraint to 'model' parameter. Either hardcode valid options in schema or expose a discovery tool (e.g., 'list_antigravity_models') that returns available models. This prevents hallucinated model names.
• Add format and length constraints to 'paths' parameter. Specify: 'Absolute file or directory paths (must exist and be readable). Max 10 paths per call. Paths with spaces must be quoted.'
• Expose timeout as optional parameter (timeout_seconds: integer, min 30, max 600, default 300) so agents can control response latency for time-sensitive queries.