The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as an adapter that processes project-specific instructions and user briefs to automate Digital Content Creation (DCC) workflows.
Ingestion points: External project instructions, user-provided briefs, and the referenced application-specific markdown files (e.g., references/blender.md).
Boundary markers: Instructions explicitly direct the agent to respect existing user authorizations and project boundaries, though they do not define specific text delimiters for untrusted inputs.
Capability inventory: Capabilities include running Python and MEL scripts within Houdini, Maya, and Nuke, as well as executing local CLI tools like comfy-cli and mayatools.
Sanitization: The skill relies on its library of "audited lessons" and "provenance" checks to ensure that generated scripts follow safe, verified patterns.
[DYNAMIC_EXECUTION]: The skill contains technical snippets and guidelines for generating and executing code at runtime within host environments.
Evidence: references/houdini.md provides a Python template for programmatically unlocking and updating Houdini Digital Asset (HDA) definitions.
Evidence: references/nuke.md references the creation and compilation of Blink kernels and inference examples.
Context: This behavior is categorized as low-risk dynamic execution as it uses known templates to perform the skill's primary function of professional software automation.

vfx-workflow