The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates shell commands for ffmpeg, ffprobe, and curl by interpolating variables such as <slug>, <lang>, and <source>. The absence of strict sanitization or escaping for these variables creates a risk of command injection if they are manipulated to include shell metacharacters.
[DYNAMIC_EXECUTION]: The skill uses node -e to execute a JavaScript snippet that calls require() on a file path constructed using the <slug> variable. This dynamic loading from a computed path could be exploited to access arbitrary JSON files or interfere with the script's execution flow.
[INDIRECT_PROMPT_INJECTION]: Data ingested from a local edit.json file is used to construct ffmpeg filters. Since this external data is interpolated into a shell command without validation, an attacker capable of modifying the file could potentially inject malicious parameters into the video processing pipeline.
[COMMAND_EXECUTION]: The skill instructs the agent to source a .env file using the shell . command to load environment variables. If the agent is induced to write malicious content into this file (e.g., from a user's chat message containing shell commands), those commands would be executed when the file is sourced.
[EXTERNAL_DOWNLOADS]: The skill interacts with the ElevenLabs API (api.elevenlabs.io) to upload video files and download dubbed audio. This is a core part of the skill's functionality targeting a well-known service, but involves external transmission of media and credentials.

short-translate