The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user-provided data, including text prompts, reference images, and audio clips, which are then passed to external video models. It implements safety guardrails by instructing the agent to explicitly verify consent for real-person references and to decline requests for deceptive content or unauthorized celebrity likenesses.
Ingestion points: User prompts, reference images (Soul ID training), and audio inputs.
Boundary markers: None explicitly defined for prompt wrapping, but ethical guidelines are provided.
Capability inventory: Communicates with the Higgsfield Cloud API and third-party model gateways.
Sanitization: Instructions require the agent to confirm rights/consent and enforce a deepfake prohibition policy.
[DATA_EXPOSURE]: The instructions document the standard authentication methods for the Higgsfield SDK, mentioning environment variables like HF_KEY and the Authorization: Bearer header. No hardcoded secrets, private keys, or sensitive local file paths are exposed.
[EXTERNAL_DOWNLOADS]: The skill references the official Higgsfield Python SDK (higgsfield-ai/higgsfield-client) and provides links to official documentation, pricing pages, and reputable third-party API resellers. All referenced resources are consistent with the skill's primary purpose and originate from relevant service domains.

higgsfield-video