The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill exposes an execute_python(code) tool that allows for arbitrary Python code execution within the Blender environment.
Evidence: The 'Quick reference' section defines execute_python(code) and notes it is 'Off until the user enables Allow connected agents to run Python' in the Scenario preferences.
[DATA_EXFILTRATION]: The skill includes capabilities to read local files and send data to external services.
Evidence: The upload_reference(path, kind) tool accesses file paths, though it is described as restricted to user-chosen files. Tools like capture_reference and generate interact with the remote Scenario API for asset processing.
[EXTERNAL_DOWNLOADS]: The documentation references external sources for extension installation and updates.
Evidence: Mentions https://blender.scenario.com/repo/index.json and https://github.com/scenario-labs/blender-plugin for extension setup. These are vendor-owned resources.
[INDIRECT_PROMPT_INJECTION]: The skill processes complex scene data and external job statuses which could potentially contain malicious instructions if the source data is compromised.
Ingestion points: Scene data via scene_summary, job metadata via job_status, and result applications in SKILL.md.
Boundary markers: None explicitly defined in the provided instructions.
Capability inventory: Arbitrary Python execution (execute_python), file reading (upload_reference), and scene modification tools found in SKILL.md.
Sanitization: No specific sanitization or validation logic is described for data returned from the MCP server or Blender API.

scenario-blender