The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script audio/scripts/lib/bgm.mjs uses python -m pip install to dynamically install Python packages from PyPI (including transformers, torch, soundfile, numpy, and google-genai) at runtime when audio generation features are triggered.
[DYNAMIC_EXECUTION]: The skill generates and executes inline Python scripts for background music generation and authentication tasks. It also shells out to various external binaries such as ffmpeg, ffprobe, npx, and bun using Node.js spawn and execFileSync functions.
[EXTERNAL_DOWNLOADS]: The skill fetches color lookup table (.cube) files from HeyGen's CDN (static.heygen.ai) and interacts with the Openverse API to search for and download sound effects. It includes logic in scripts/lib/media-fetch.mjs to validate redirect targets and prevent downloads from private network IP ranges (SSRF protection).
[DATA_EXFILTRATION]: The skill accesses sensitive credential information in the user's home directory (~/.heygen/credentials) to authenticate with the HeyGen API. It also transmits anonymous usage telemetry (capturing media types and providers used) to PostHog (us.i.posthog.com), which is documented and can be opted out via environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data sources, including transcription files (SRT, VTT, JSON) and user-provided "intent" descriptions. These inputs are used to drive media generation and manipulation commands, creating an ingestion surface where malicious instructions could potentially influence agent behavior.

media-use