canvas-designer
Pass
Audited by Gen Agent Trust Hub on May 19, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions using keywords like 'CRITICAL' and 'MUST' to dictate agent behavior regarding skill activation based on specific triggers. Additionally, it processes user-provided design markers ([@design_marker:...]) which represent an indirect prompt injection surface. The skill mitigates this by requiring the agent to use visual_understanding results to describe the full scene and construct precise prompts, rather than blindly following instructions embedded in the marker.\n
- Ingestion points: SKILL.md and reference/image/design-marker.md process user markers.\n
- Boundary markers: No explicit delimiters are used for marker content.\n
- Capability inventory: create_canvas, generate_canvas_images, search_canvas_images, generate_canvas_videos, query_video_generation, search_image_prompts (in SKILL.md).\n
- Sanitization: The agent is instructed to use visual descriptions of the reference images to inform and constrain the generated prompt.\n- [COMMAND_EXECUTION]: The skill uses run_sdk_snippet to execute Python code snippets that interact with the platform's canvas and media generation tools. This is the primary method of operation described throughout the skill and reference documentation.\n- [EXTERNAL_DOWNLOADS]: The skill utilizes the search_canvas_images tool to find and download images from external web sources to the canvas, as detailed in reference/image/image-search.md.
Audit Metadata