The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Multiple reference files (ADVANCED_NODE_TREES.md, NODE_TREE_OPTIMIZATION.md, and PROCEDURAL_MODELING.md) provide Python snippets that use the requests library to send dynamically generated code to a remote URL for execution.\n
Evidence: Snippets use requests.post(url, json={"code": code}) where url is an undefined variable, suggesting a staged or configurable remote execution endpoint.\n- [DATA_EXFILTRATION]: The instruction to send locally generated code to a remote URL via raw HTTP POST requests constitutes a significant data exfiltration risk, as workspace context or generated scripts are transmitted to an external server.\n- [METADATA_POISONING]: The version history in SKILL.md (v2.0.0) deceptively claims that legacy HTTP bridge patterns and requests wrappers were removed, whereas they remain central to the provided code in the advanced reference documentation.\n- [COMMAND_EXECUTION]: The Multi-Lane Scatter System in SKILL.md uses raise SystemExit, which can terminate the agent's Python interpreter and cause a denial of service if the environment does not match expectations.\n- [DYNAMIC_EXECUTION]: The skill assembles executable Python strings at runtime and instructs the agent to execute them via a remote network bridge instead of using the claimed Blender MCP toolset.\n- [OBFUSCATION]: The skill uses cryptic "obs 64" and "observation 64" markers in its version history and troubleshooting sections, and relies on an undefined global variable url for its primary network operations, which may be part of a staged configuration or activation mechanism.

blender-geometry-nodes