Single tool with clear naming (prompt.get) and adequate schema. Description is present but generic (72 chars, below ideal 50-200 range). Input schema properly typed with required field and object properties. However, output schema is not documented, LLMs cannot predict what fields the response contains. No error handling guidance, no parameter constraints (e.g., promptId format), and no mention of variable substitution behavior or edge cases. Lacks composition guidance for multi-step workflows.

• Document the output schema explicitly. Example: 'Returns {id: string, content: string, version: number, createdAt: string, variables: string[]}'. This enables LLMs to chain calls and extract the right fields.
• Expand description to 100-150 chars explaining: 'Retrieves the production-promoted version of a prompt template. Use this to fetch the latest stable prompt for inference. Returns the template content with optional variable placeholders for substitution.'
• Add format constraint to promptId: 'Prompt ID or name (alphanumeric, 1-64 chars, case-insensitive). E.g., "system-prompt-v1" or "123".'
• Document variable substitution behavior: 'Variables object maps template placeholder names to values. Missing variables are left as-is in the output. Reserved names: __version__, __timestamp__.'
• Add error handling examples in description: 'Returns error if promptId not found (suggest available prompts) or if variable substitution fails (list required variables).'
• Consider adding a 'version' parameter to retrieve non-production versions for testing or rollback scenarios.