The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a parameter disable_safety_filter which defaults to true. This represents an explicit instruction to override and bypass standard content safety protocols during the video generation process.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external dependencies, including belt-sh/cli and various skills from the inference-sh GitHub organization via npx. It also references remote content from raw.githubusercontent.com/inference-sh/skills/ for installation instructions.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute belt CLI commands, such as belt login and belt app run, which interact with external APIs and services.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface. It ingests untrusted data from external image and audio URLs and interpolates user-provided text (voice_script, video_prompt, voice_prompt) directly into commands without using boundary markers, sanitization, or instructions to ignore embedded commands.
Ingestion points: image (URL), audio (URL), voice_script (String), video_prompt (String), voice_prompt (String) defined in the inputs for belt app run.
Boundary markers: Absent. The user-provided content is passed directly as JSON input to the CLI tool.
Capability inventory: The skill uses Bash(belt *) to execute shell commands.
Sanitization: Absent. There is no evidence of input validation or escaping for the strings processed by the CLI tool.

p-video-avatar