The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses ffprobe in SKILL.md to programmatically inspect the technical properties (duration, bitrate, codec) of generated MP3 files. This is a controlled execution for file validation purposes.
[EXTERNAL_DOWNLOADS]: The skill references multiple official resources and documentation from Google and Google DeepMind for API usage and prompting guidelines, which are recognized as trusted organizations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data in the form of music generation prompts, custom lyrics, and image inputs which could potentially contain malicious instructions.
Ingestion points: User-provided text prompts, lyric blocks, and image inputs described in SKILL.md and references/api-and-prompting.md.
Boundary markers: The skill suggests using a Lyrics: block to separate lyrics from production instructions in references/api-and-prompting.md, though no technical escaping is enforced.
Capability inventory: The skill performs network requests to the Gemini Interactions API and executes local shell commands via ffprobe to analyze output files as noted in SKILL.md.
Sanitization: No explicit sanitization, filtering, or validation logic for external content is specified before interpolation into the generation prompt.

lyria