The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the belt CLI tool for video production tasks. The skill configuration proactively limits the execution environment by restricting the Bash tool to only allow commands starting with belt, which is a recognized security best practice for scoping agent capabilities.
[PROMPT_INJECTION]: The skill processes user-supplied text and media URLs, creating a surface for indirect prompt injection. This is inherent to the primary purpose of video generation skills and the risk is minimized by the restricted tool-call permissions.
Ingestion points: The voice_script text and the image and audio URL parameters used in the belt app run command examples in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are included in the prompt templates.
Capability inventory: The allowed-tools frontmatter in SKILL.md restricts the Bash tool to belt commands only.
Sanitization: There is no evidence of input validation or sanitization for the strings passed to the CLI tools.

talking-head-production