Four tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. All tools have descriptions (10-60 chars), but most are terse and lack actionable context. Input schemas present but parameters lack descriptions in most cases. No error handling guidance, no output schemas documented, and no tool annotations (readOnlyHint/destructiveHint). Naming is verb-forward (load_, list_, set_, get_) but descriptions do not explain WHEN to use each tool or what downstream operations require. Variable interpolation and frontmatter stripping are implementation details not surfaced in descriptions.

• Add formal output schemas to all tools. E.g., load_prompt returns {type: 'string', description: 'Prompt content with frontmatter stripped and variables interpolated'}; list_prompts returns {type: 'array', items: {type: 'object', properties: {name: string, ...}}}; get_prompt_health returns {type: 'object', properties: {errors: array, warnings: array, ...}}.
• Expand parameter descriptions with constraints and examples. E.g., 'name: Prompt file name without .md extension (alphanumeric + underscore, case-sensitive). Returns error if file not found or unreadable.'
• Add tool annotations: set_variable with destructiveHint=true; load_prompt, list_prompts, get_prompt_health with readOnlyHint=true. This signals to agents which operations are safe to retry.
• Document error responses with recovery guidance. E.g., 'If prompt not found, call list_prompts() to see available names' or 'If variable interpolation fails, check that all {{key}} placeholders have corresponding set_variable() calls.'
• Expand tool descriptions to 100-150 chars explaining purpose, when to use, and what to do with results. E.g., 'Load a prompt by name, strip YAML frontmatter, and interpolate variables set via set_variable(). Use this to retrieve ready-to-use prompt text for sending to Claude. Returns error if prompt file not found.'