The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes code from the NPM registry using npx @getvhs/vhscli@latest to ensure the latest version of the vendor's platform tools is used.
[COMMAND_EXECUTION]: Utilizes shell commands including npx, file, sips, and ffmpeg to detect file types and process media assets.
[EXTERNAL_DOWNLOADS]: Fetches and runs the @getvhs/vhscli package from the public NPM registry at runtime.
[CREDENTIALS_UNSAFE]: Manages authentication tokens stored in the ~/.vhs/session.json file to maintain a session with the vendor's cloud services.
[DATA_EXFILTRATION]: Transfers user-provided media files (images, PDFs, and videos) to the vendor's server for analysis via the vhscli chat and generate commands.
[PROMPT_INJECTION]: The skill handles untrusted external data (media files) which presents a potential indirect prompt injection surface.
Ingestion points: Local files are ingested via -i, -f, and -v flags in the vhscli chat command as described in SKILL.md.
Boundary markers: No specific delimiters or boundary instructions are included in the prompt construction examples.
Capability inventory: The skill possesses capabilities for command execution and network communication via the vendor's CLI tool.
Sanitization: The instructions do not specify any sanitization or filtering of media content before it is uploaded and processed by cloud models.

vhscli