The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads multiple machine learning models from external sources upon first use, including Kokoro-82M, various Whisper models, and u2net weights, caching them in ~/.cache/hyperframes/. Evidence: Weights ranging from ~168 MB to 3.1 GB are downloaded and cached (SKILL.md).- [REMOTE_CODE_EXECUTION]: The skill utilizes npx hyperframes to fetch and execute the CLI tool directly from the npm registry. Evidence: Multiple commands in SKILL.md invoke functionality via npx hyperframes.- [COMMAND_EXECUTION]: The instructions recommend system-level software installations that may require administrative privileges. Evidence: Usage of apt-get install espeak-ng and brew install espeak-ng is mentioned in SKILL.md.- [PROMPT_INJECTION]: The skill exposes an indirect prompt injection and command injection surface by interpolating user-controlled data into shell-executed commands.
Ingestion points: The skill accepts user-provided text strings for narration and processes external files such as script.txt and audio.mp3 (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are used to isolate untrusted content from the command arguments.
Capability inventory: The agent executes shell commands via npx and writes to local cache directories (SKILL.md).
Sanitization: There is no documentation of sanitization or validation logic to prevent malicious input from altering command execution.

hyperframes-media