blender-superskill
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python script, "scripts/blender_orchestrator.py", to manage modeling workflows. This script further executes system commands, specifically the "codex" CLI tool, to perform modeling steps via "subprocess.Popen".
- [INDIRECT_PROMPT_INJECTION]: The orchestrator script processes untrusted data which can influence the behavior of the sub-agent worker.
- Ingestion points: User-provided strings via the "--task" argument and external images via the "--reference" argument are ingested into the script's state and subsequently included in prompts generated within "scripts/blender_orchestrator.py".
- Boundary markers: The prompt construction in functions like "planning_prompt" and "build_prompt" uses simple text headers (e.g., "Task:\n") but lacks strong delimiters or instructions to ignore potential injections within the task content.
- Capability inventory: The script invokes "codex exec" with the "--full-auto" flag, granting the sub-agent the ability to perform file operations and execute shell commands without human intervention.
- Sanitization: There is no evidence of input validation, escaping, or sanitization of the "task" description before it is interpolated into the prompts for the worker agents.
Audit Metadata