The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and act upon user-supplied creative briefs and external reference data, establishing an ingestion surface for untrusted instructions.
Ingestion points: Processes user-provided briefs, existing Blender scenes, and external images retrieved via search tools or browser actions as described in SKILL.md and references/astra-workflow.md.
Boundary markers: The workflow uses a 'working contract' and 'visual contract' to explicitly record and verify decisions, which provides a layer of human-in-the-loop oversight but does not include programmatic delimiters for all interpolated data.
Capability inventory: The skill can invoke specialized Blender modeling, rendering, and simulation workflows, as well as write local state and reference documentation files (reference_notes.md, work_state.md).
Sanitization: The guidance emphasizes visual and structural verification against references but does not detail methods for sanitizing data or metadata from external files or web search results.
[EXTERNAL_DOWNLOADS]: The skill references the retrieval of 3D assets from well-known industry repositories and points to official developer documentation for model adaptation.
Evidence: Mentions the use of Poly Haven for material downloads and links to official documentation at developers.openai.com in references/astra-workflow.md.

blender-art-direction-intake