Single tool with critical definition gaps. Tool name 'convert_map_tool_walls_only' is verbose and unclear (contains 'tool' redundantly). Input schema present but minimal: only base64_image with generic description lacking format/size constraints. Output is a plain string URL with no schema documentation. No parameter validation guidance, no error recovery hints, no mention of side effects (file writes to static/). Tool description (67 chars) is borderline adequate but lacks WHEN to use it, prerequisites, or output structure. No pagination, no structured response, no idempotent guarantees documented. Follows none of the 54 Agentic Tool Patterns for composition, error handling, or output shaping.

• Rename tool to 'generate_3d_walls_from_map' or 'convert_2d_map_to_3d', verb_noun pattern, no redundancy.
• Add parameter constraints: 'base64_image: Base64-encoded PNG/JPG image (max 10MB, 256x256 - 4096x4096 px, grayscale or color). Must be a top-down floor plan or map with clear wall edges.'
• Document output schema: return structured JSON {url: string, format: 'zip', file_count: int, expires_in_seconds: int, wall_height_cm: float} instead of plain string.
• Expand tool description to 150 - 200 chars: 'Converts a 2D floor plan or map image (PNG/JPG) to a 3D OBJ model with wall geometry. Returns a ZIP containing individual wall OBJ files. Use this to generate 3D models from architectural drawings or hand-drawn maps. Caches results by image hash.'
• Add error classification: return {error: string, type: 'invalid_input'|'processing_failed'|'timeout', recovery: string} so LLM knows whether to retry, ask user for different input, or escalate.