The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text (lyrics) and audio metadata, which are used to generate storyboard content and visual prompts for video generation models.
Ingestion points: Untrusted data enters the skill via the lyrics text input and audio file paths defined in SKILL.md and processed by scripts/lyrics-parser.py.
Boundary markers: The skill does not implement delimiters or safety instructions in the output JSON or Markdown files to prevent downstream AI models from following instructions potentially embedded within the processed lyrics.
Capability inventory: The skill utilizes shell command execution (running its internal Python scripts), file system read/write operations (e.g., in scripts/merge-output.py), and generates prompts intended to be consumed by other AI models (such as Kling, Sora, or Veo).
Sanitization: There is no evidence of sanitization or content filtering for the lyrics before they are integrated into the final script outputs.
[COMMAND_EXECUTION]: The skill's workflow relies on executing local Python scripts to perform audio analysis and data processing tasks.
Evidence: SKILL.md specifies the use of python3 scripts/audio-analyzer.py, scripts/lyrics-parser.py, and scripts/scene-assigner.py within the local environment to generate the final MV storyboard.

virtual-singer-mv-script