The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Provides instructions to download and execute the HeyGen CLI installation script from https://static.heygen.ai/cli/install.sh via a shell pipe to bash. HeyGen is a well-known AI media service, and this pattern is its official installation method.
[DATA_EXFILTRATION]: Sends anonymous usage telemetry to PostHog (posthog.com), which is not in the standard whitelist. The telemetry records coarse event data such as media type, resolution source, and provider. The skill includes a clear opt-out mechanism via the HYPERFRAMES_NO_TELEMETRY or DO_NOT_TRACK environment variables.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of several external Python packages for local processing, including mflux, parakeet-mlx, kokoro-onnx, and google-genai.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted natural language data (e.g., the --intent flag and script text) and interpolates it into prompts for external AI generation tools.
Ingestion points: scripts/resolve.mjs and audio/scripts/audio.mjs.
Boundary markers: None identified in the provided documentation for the generation steps.
Capability inventory: Executes shell commands (ffmpeg, heygen, mflux) and performs network requests to third-party AI APIs.
Sanitization: No explicit sanitization or filtering of intent text is mentioned before model interpolation.
[CREDENTIALS_UNSAFE]: Accesses sensitive configuration paths such as ~/.heygen/credentials and reads environment variables like $ELEVENLABS_API_KEY and $GEMINI_API_KEY. This access is necessary for the skill's core functionality of integrating with these media services.

media-use