Single tool 'gen_viz' has a reasonable description (89 chars) but critical schema and parameter documentation gaps. Input schema is visible with two string parameters (scene_name, code), both with descriptions. However, the output schema is NOT formally documented in the tool definition, only mentioned in the docstring as 'returns a dictionary containing the path to the rendered MP4 clip and its duration.' The tool name 'gen_viz' lacks a clear action verb (should be 'generate_visualization' or 'render_animation'). No error handling guidance, no input validation constraints, no output schema declaration. The tool performs a destructive write operation (generates files, runs Docker) but has no confirmation pattern or dry-run capability.

• Rename 'gen_viz' to 'generate_animation' or 'render_manim_scene' to start with a clear action verb matching LLM intent inference patterns.
• Add formal output schema to tool definition: {"type": "object", "properties": {"path": {"type": "string", "description": "Absolute file path to rendered MP4"}, "duration": {"type": "number", "description": "Video duration in seconds"}}, "required": ["path", "duration"]}
• Document input constraints: scene_name must be a valid Python class name (alphanumeric + underscore, no spaces); code must be valid Python with a Manim Scene subclass. Add examples in descriptions.
• Implement error classification: catch Docker errors separately from Manim syntax errors, ffprobe failures, and file I/O errors. Return structured error with error_type, message, and recovery_hint fields.
• Add a dry_run parameter (boolean, default false) to preview the command without executing Docker. Return the Docker command that would run, allowing LLM to validate before committing.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry