The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a utility script _postplus_shared/00-core/shared-runtime/scripts/download_videos_from_manifest_with_ytdlp.mjs that can download media from external URLs using the yt-dlp library based on a manifest.
[COMMAND_EXECUTION]: Subprocesses are used to run python3 (for media processing) and the vendor's CLI tool (postplus) for administrative tasks such as billing confirmation and capability execution.
[DATA_EXFILTRATION]: Shared scripts access local configuration files to retrieve session tokens, which are sent to the vendor's API endpoints for authentication. This is standard behavior for vendor-integrated skills.
[PROMPT_INJECTION]: The skill processes raw user material (fragments, drafts), which represents an indirect prompt injection attack surface.
Ingestion points: User-provided drafts and scattered thoughts processed according to SKILL.md.
Boundary markers: Absent; instructions do not specify the use of delimiters to isolate untrusted user input.
Capability inventory: Includes network operations, file system access, and subprocess execution via shared libraries.
Sanitization: No explicit sanitization or filtering of the source material is implemented.

xiaohongshu-notes