Three tools with clear verb-noun naming (search_, list_, download_) and reasonable descriptions (100-150 chars). All parameters have type definitions and descriptions. However, output schemas are not explicitly documented in the source code, only inferred from skill_success() calls. Parameter descriptions lack constraint details (e.g., valid asset_type enum values, resolution options). Error handling uses generic skill_exception() without recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.

• Add formal JSON Schema output definitions to each tool. Document the structure of 'assets' (array of {id, name, type, categories, tags, thumbnail_url}), 'files' (dict of file_type → resolutions), and 'asset_descriptor' (AssetDescriptor fields).
• Convert prose constraints to enums: asset_type ∈ {models, hdris, textures, all}; file_type ∈ {gltf, hdr, exr, ...}; resolution ∈ {1k, 2k, 4k, ...}. Query the Poly Haven API once to enumerate valid values.
• Add tool annotations: mark download_polyhaven_asset with destructiveHint=true; mark search_polyhaven_assets and list_polyhaven_files with readOnlyHint=true.
• Enhance error messages with recovery hints. E.g., 'Variant not found. Call list_polyhaven_files(asset_id="...") to see available resolutions and formats.'
• Add parameter constraints: limit 1 - 100 (default 10); output_dir must be a writable directory path; resolution and file_type should reference the enum values from list_polyhaven_files().