scroll-world
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied strings for subjects, brand names, and tones, which are then interpolated into AI prompt templates and shell commands. Without robust sanitization or strict boundary markers, this input surface could be exploited to influence the behavior of the agent or the content generated by downstream AI models.
- [EXTERNAL_DOWNLOADS]: The skill retrieves generated images and videos from external URLs provided by AI generation platforms like Higgsfield and Monid. These downloads are required for the skill's primary purpose of creating media assets.
- [COMMAND_EXECUTION]: The workflow relies on the
Bashtool to execute CLI utilities such ashiggsfield,monid,codex,ffmpeg, andffprobe. These tools are used for generating assets, extracting frames, and final video encoding. - [DATA_EXFILTRATION]: The skill uses
curl -Tto upload locally generated image frames to Monid's file system during the video generation process. This is a functional requirement for the Monid-based rendering pipeline. - [DYNAMIC_EXECUTION]: The skill dynamically constructs and executes prompt files and script segments based on user-provided configuration. This is a standard mechanism used by the skill to generate customized landing page content.
Audit Metadata