The Agent Skills Directory

[DYNAMIC_EXECUTION]: The bridge script (scripts/blender-bridge.py) utilizes exec() and compile() functions to run Python code received via HTTP POST requests within the live Blender process.
[PERSISTENCE]: The documentation in SKILL.md recommends installing the bridge in the Blender startup directory (~/Library/Application Support/Blender/5.2/scripts/startup/), which causes the code-execution server to start automatically whenever the application is launched.
[COMMAND_EXECUTION]: The skill provides scripts/blender-send.sh, a bash script that uses curl to send arbitrary Python commands or script files to the local bridge server, effectively controlling the host application.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a high-capability execution environment that lacks authentication for local processes. If the AI agent is induced to generate or execute malicious Blender scripts through untrusted data input, it could result in unauthorized file access or system data exfiltration.
Ingestion points: The agent generates Python code for the bridge based on user instructions and scene data processed from the live session (e.g., in scripts/blender-send.sh and the various example scripts).
Boundary markers: There are no explicit boundary markers or instruction-ignoring wrappers applied to the generated code before it is passed to the bridge for execution.
Capability inventory: The bridge possesses full access to the Blender Python API (bpy), the local file system via the os module, and the ability to return stdout/tracebacks to the caller.
Sanitization: While the bridge includes a cross-origin request check (Origin and Sec-Fetch-Site) to prevent browser-based attacks, it does not require a token or authentication for local requests.

blender