The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted textual data from video transcripts using tools like read_script and read_captions (identified in SKILL.md and references/captions.md). Since the skill has the authority to modify the project timeline and visual assets via apply_script, edit_item, and edit_asset, it is vulnerable to indirect prompt injection. Malicious instructions embedded in a video's transcript could potentially trick the agent into performing unauthorized edits or deletions. There are no mentioned boundary markers or sanitization routines to mitigate this risk.
[DYNAMIC_EXECUTION]: The skill involves the runtime generation and modification of JSX code to create motion graphics through the create_motion_graphic_from_code tool (described in references/motion-graphics.md). While this is a primary feature intended for advanced visual editing, the generation of executable UI code at runtime based on high-level agent planning or user input is a sensitive operation that could be exploited if the input content is not strictly controlled.

talking-head-guide