The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a dedicated bash script (scripts/mix.sh) to automate media mixing tasks via ffmpeg and ffprobe. It employs standard shell practices such as variable quoting and argument parsing.
Evidence: The skill invokes bash scripts/mix.sh with multiple arguments to process audio and video files.
[EXTERNAL_DOWNLOADS]: The skill provides guidelines and commands for downloading media assets from external repositories, including Pixabay and YouTube, using curl and yt-dlp.
Evidence: Download patterns like curl -L -o ... and yt-dlp -x ... are documented in SKILL.md and references/music-sources.md for acquiring background music.
[REMOTE_CODE_EXECUTION]: The skill leverages the uvx tool-runner to dynamically execute the edge-tts package. This involves fetching and running code from public registries at runtime.
Evidence: uvx edge-tts --text "$(cat ...)" in SKILL.md.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted input data from external files and network sources.
Ingestion points: Content read from 02_脚本/口播.txt and media downloaded from user-supplied URLs.
Boundary markers: The skill does not implement explicit delimiters or guardrails to separate external data from instructional content during processing.
Capability inventory: The execution environment provides access to shell commands, network requests, and file system operations.
Sanitization: The skill relies on the target tools to safely handle input data and does not perform independent sanitization or validation of the ingested text.

space-video-audio