The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts gen_dub.py and sk_cli.py use subprocess.run to call ffmpeg, ffprobe, and the screenkite CLI. These calls use list-based arguments which mitigate shell injection risks.
[EXTERNAL_DOWNLOADS]: The skill connects to api.fish.audio and api.elevenlabs.io to perform text-to-speech synthesis. These are established services and the skill uses standard API patterns over HTTPS.
[PROMPT_INJECTION]: The skill processes transcription data which may contain untrusted content. Ingestion points: Untrusted data enters via transcript.json and cues.json. Boundary markers: None identified in the code logic. Capability inventory: Subprocess execution of media tools and network requests for TTS. Sanitization: The skill instructions mandate a human review step in Phase 4 and Phase 6, where the user must confirm translations and the mutation plan before any destructive actions are taken.
[CREDENTIALS_UNSAFE]: The skill retrieves API keys from .env files in the project path or ~/.config/env/. While this involves reading local configuration files, it is a standard method for managing secrets in local development environments and does not constitute an unsafe hardcoding of credentials.

screenkite-audio-dub