The Agent Skills Directory

[SAFE]: The skill demonstrates high security awareness by explicitly instructing the agent and user never to store API keys, tokens, or credentials within the persistence files (e.g., marca/avatar.json).
[SAFE]: Comprehensive ethical guardrails are enforced through clear instructions: it requires explicit consent for face and voice cloning, prohibits the impersonation of public figures or minors, and strictly forbids the generation of sexually explicit content.
[INDIRECT_PROMPT_INJECTION]: The skill maintains an attack surface by ingesting business data (marca.json) and external reference documents to define character traits for image and audio generation.
Ingestion points: The agent reads and processes data from marca.json, references/marca.md, and the output of the proveedores tool.
Capability inventory: The skill performs file system writes (marca/avatar.json), image generation/editing, media uploads, and voice synthesis via connected provider tools.
Boundary markers: Data integrity is protected by a strict JSON schema (references/avatar.schema.json) and defined structured fields, which mitigate risks of schema confusion or instruction leakage during data processing.
Sanitization: No explicit sanitization or filtering of external text is implemented beyond standard structured data validation.

avatar