The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local shell, Python, and Node.js scripts to manage the video production pipeline. Evidence includes calls to bash scripts/ep-dir.sh, python3 scripts/voice.py, ffmpeg, node audio/episodes/<期>.mjs, and various npm/npx commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (keywords or themes) to generate scripts and video scenes. This creates a vulnerability where malicious input could influence downstream automated actions.
Ingestion points: User-provided "keyword", "theme", or selections from "选题候选池.md" are used as the primary input for the workflow.
Boundary markers: The instructions do not specify any delimiters or safety constraints for the AI when interpolating user data into scripts or scene configurations.
Capability inventory: The workflow has extensive capabilities including file system write access, execution of Python and shell scripts, and rendering executable React components via Remotion.
Sanitization: There is no mention of sanitizing or validating user-supplied keywords before they are used in script generation or command arguments.
[DYNAMIC_EXECUTION]: The skill generates and subsequently executes source code as part of its normal operation. Specifically, it creates .tsx React components for video scenes and .mjs scripts for audio mapping, which are then processed by the Node.js runtime and Remotion compiler.

weilanx-creator-pixel-video