The Agent Skills Directory

[SAFE]: A detailed analysis of the provided SKILL.md and reference files (animation-craft.md, examples.md, hooks.md, and model-specs.md) shows that the skill is focused on assisting users with creative prompting. There is no evidence of credential harvesting, unauthorized data exfiltration, or attempts to bypass agent safety protocols. The use of specialized tools in the 'higgsfield' namespace is consistent with the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture involves ingesting user-supplied concepts and styles to build complex prompts, which is a common surface for indirect prompt injection. Ingestion points: User intent regarding character design, animation style, and scene actions (SKILL.md). Boundary markers: The skill does not explicitly utilize delimiters or 'ignore embedded instructions' markers when interpolating user data into prompts. Capability inventory: The skill utilizes video and image generation tools as well as media upload and balance check capabilities. Sanitization: The skill incorporates a critical safety control by requiring the agent to present the final assembled prompt and parameters to the user for explicit review and confirmation ('Present for review' and 'Opt-in generation' sections in SKILL.md) before any generation tool is called.

higgsfield-cartoon