The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The bridge script (scripts/blender-bridge.py) implements a local HTTP server that accepts Python code via POST requests and executes it using the exec() and compile() functions. This provides the agent with full scriptable control over the Blender environment. The bridge includes security controls by binding strictly to 127.0.0.1 and utilizing _reject_cross_origin logic to prevent unauthorized cross-origin requests from web browsers.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it generates and executes scripts based on potentially untrusted user-supplied data.
Ingestion points: Data provided by the user or external sources that the agent uses to build 3D models or scenes via SKILL.md workflows.
Boundary markers: None identified in the prompt templates to distinguish between instructions and data.
Capability inventory: Arbitrary Python execution via the bridge script (scripts/blender-bridge.py), including the ability to perform file operations and network requests.
Sanitization: No input sanitization or validation mechanisms are present for user-provided content before it is incorporated into executable scripts.
[COMMAND_EXECUTION]: The utility script scripts/blender-send.sh uses shell commands and curl to interact with the local bridge, facilitating the transfer and execution of Python payloads.

blender