The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses node:child_process (spawn and execFile) to run media processing tools like ffprobe and local Python-based AI models. These operations are essential for auditing HyperFrame project statuses and generating audio/video content. Arguments are passed as arrays, which prevents shell-based command injection.
[EXTERNAL_DOWNLOADS]: The skill is configured to download AI models from HuggingFace (specifically from the mlx-community organization) and interacts with AI generation APIs. These downloads and references are from well-known services and are required for the skill's primary functionality.
[DATA_EXFILTRATION]: To fulfill media generation requests, the skill sends prompts and metadata to external AI provider endpoints such as Moonrouter and BytePlus. These network operations are triggered by "requested" status flags on records in the Busabase backend.
[REMOTE_CODE_EXECUTION]: Local Python scripts included in the repository (gen_voice.py) or stored in project-relative paths are executed at runtime. These scripts handle the integration with local inference engines.
[INDIRECT_PROMPT_INJECTION]: The skill ingests script beats, character profiles, and shot descriptions from a Busabase workspace and interpolates them into prompts for external AI generation models. While no explicit boundary markers are used in the prompt templates, the data flows are consistent with the skill's intended creative workflow.

kelly-drama