The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs automated video processing using ffmpeg and ffprobe based on user-provided files. It includes instructions to "Install missing dependencies" and "Reuse... the project's documented startup command" (e.g., from a README or package.json), which involves executing arbitrary shell commands derived from potentially untrusted local project files or configurations.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external websites and media file metadata.
Ingestion points: Discovery of video usages through website routes, component source code, sitemaps, and rendered HTML content in audit.md and scan-and-optimize.md; video stream metadata and file properties retrieved via ffprobe in compression.md.
Boundary markers: The instructions do not define specific delimiters or warnings to treat ingested external data as non-instructional.
Capability inventory: The skill possesses capabilities for filesystem read/write access (to modify source code), browser-based network access, and shell command execution.
Sanitization: No explicit sanitization or validation of the data ingested from website content or file metadata is described before the data is used in command construction or reporting.
[DYNAMIC_EXECUTION]: The skill dynamically constructs complex shell commands by concatenating flags and parameters (e.g., INPUT, FILTER_FLAGS, AUDIO_FLAGS) based on properties and preferences discovered at runtime from the source files.

compress-video