The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses sensitive credential files to authenticate with the HeyGen API.
Evidence: scripts/lib/heygen.mjs contains heygenCredential() which reads from ~/.heygen/credentials and process.env.HEYGEN_API_KEY.
Evidence: scripts/lib/heygen.mjs implements loadEnvFromDir() which recursively searches for and loads .env files up to five directories above the project root.
[EXTERNAL_DOWNLOADS]: The skill automatically installs missing Python dependencies at runtime using the system package manager.
Evidence: scripts/lib/bgm.mjs uses spawnSync('pip', ['install', ...]) to install transformers, torch, soundfile, numpy, and google-genai if they are not detected on the system.
[DYNAMIC_EXECUTION]: The skill generates and executes Python code at runtime to handle background music (BGM) generation.
Evidence: scripts/lib/bgm.mjs defines musicgenScript(), which returns a multi-line Python string that is executed via spawn('python3', ['-c', script], ...).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data which is used to influence the behavior of media generation engines.
Ingestion points: audio_request.json (specifically lines[].text for TTS and bgm.prompt / bgm.query for music generation).
Boundary markers: Absent. User prompts are interpolated directly into service requests or local scripts.
Capability inventory: The skill possesses capabilities for network operations (fetch to HeyGen/external URLs), file system writes (fs.writeFileSync for audio/meta files), and subprocess execution (spawn for Python, FFmpeg, and npx).
Sanitization: The skill uses JSON.stringify() when injecting user-provided prompts into its dynamically generated Python scripts, providing some protection against command injection, but does not sanitize against indirect prompt injection targeting the AI model logic.

hyperframes-media