The Agent Skills Directory

[SAFE]: The skill consists of Markdown documentation that defines workflows for 3D asset creation and export using Blockbench. It does not contain any executable code or malicious instructions.
[COMMAND_EXECUTION]: The skill references a tool named risky_eval and explicitly instructs the agent to adhere to existing authorization and repository protocols when using it, which is a defensive instruction regarding a platform-provided capability.
[INDIRECT_PROMPT_INJECTION]: The skill instructions involve retrieving and processing data from Blockbench models, which constitutes a surface for indirect prompt injection if those models contain malicious metadata.
Ingestion points: The skill reads model structure, textures, and mesh information via list_outline, list_textures, and get_mesh_info in SKILL.md.
Boundary markers: The instructions do not specify delimiters or markers to separate ingested model data from the agent's internal instructions.
Capability inventory: The skill includes file export capabilities (export_model) and references a tool for command execution (risky_eval).
Sanitization: There are no instructions for sanitizing or validating model metadata or names before they are processed by the agent.

blockbench-use