demo-video
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill is instructed to generate a
build.shshell script that invokes tools likeplaywright,edge-tts, andffmpeg. Because user-provided data such as narration text and scene titles are interpolated into this script, there is a risk of command injection if the agent fails to escape shell-sensitive characters.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user to define video scenes and narration.\n - Ingestion points: User scene descriptions, product details, and narration scripts serve as the primary input for generating output files (SKILL.md).\n
- Boundary markers: There are no explicit instructions for the agent to use delimiters or ignore instructions potentially embedded within user data.\n
- Capability inventory: The skill can create a directory (
demo-output/) containing multiple HTML files and an executable shell script.\n - Sanitization: The instructions provide no guidance for the agent to validate or sanitize user input before it is embedded in the generated HTML or shell scripts.\n- [DYNAMIC_EXECUTION]: The skill dynamically assembles a script (
build.sh) at runtime based on the specific parameters and scene manifest generated from user input. This runtime generation of executable code represents a significant attack surface.\n- [EXTERNAL_DOWNLOADS]: The skill references a third-party repository (github.com/vaddisrinivas/framecraft) in its documentation. As this is not a verified or trusted vendor source, it poses a potential risk if the agent or user is encouraged to download and execute code from this external link.
Audit Metadata