The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the username parameter. This input is interpolated directly into instructions for image and video generation models (GPT-image-2 and Kling-omni) without sanitization or protective boundary markers. A malicious user could provide a username containing instructions intended to manipulate the output of these external models or the generated audio commentary. * Ingestion points: The username and reference_image_url are accepted as user input in SKILL.md. * Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when interpolating these values into generation prompts. * Capability inventory: The skill utilizes generate_image_edit, generate_reference_video, and analyze_media tools in SKILL.md. * Sanitization: No sanitization or validation of the username input is performed before interpolation.
[PROMPT_INJECTION]: The analyze_media tool call used for the 'Avatar-type probe' is vulnerable to content-based injection. Text or visual patterns within the user-supplied reference image could be designed to deceive the AI classifier, potentially allowing a user to bypass the skill's internal moderation checks for trademarked or copyrighted characters.

baseball-trend