sketch-to-cad
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using
ffmpegto extract frames from user-provided video files (Step 1 inSKILL.md). This involves spawning a subprocess with untrusted input file names. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (user-uploaded images and videos) to extract dimensions and features. This represents an attack surface where malicious content (e.g., text instructions in a sketch or metadata in a video) could attempt to influence the agent's behavior during the reconstruction process.
- Ingestion points: Video and image files processed in
SKILL.md(Workflow Step 1). - Boundary markers: Absent. The instructions do not define delimiters or provide warnings to ignore embedded instructions found within visual data.
- Capability inventory: Execution of
ffmpegsubprocesses, file system writes for frame extraction, and parameter script generation for CAD tools. - Sanitization: Absent. No validation or sanitization of filenames or visual content is specified.
Audit Metadata