previs-control-passes

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's scripts allow for the injection of user-supplied data into ComfyUI workflow templates, creating a surface for indirect prompt injection if malicious data is provided as input.\n
  • Ingestion points: CLI arguments in scripts/binding_apply.py (key-value pairs) and scripts/comfy_client.py (via the --set flag).\n
  • Boundary markers: The skill does not implement explicit delimiters or guardrail instructions around the injected values.\n
  • Capability inventory: The skill possesses capabilities to upload files, execute workflows (which may contain arbitrary processing logic on the ComfyUI server), and download output files.\n
  • Sanitization: While the scripts parse inputs as JSON when possible to ensure structural integrity, the content of the values is not sanitized for potential adversarial instructions targeted at LLM nodes within the workflow.\n- [EXTERNAL_DOWNLOADS]: The documentation and preflight scripts reference external resources and repositories for the associated studio application and specialized ComfyUI node packs.\n
  • Evidence: References to github.com/navarisun1982/comfyui-motion-previs-studio and github.com/stuubszzz/Stubelius-Ultimate-LTX2.5 in SKILL.md and references/requirements.md.\n
  • Context: These references are provided for user information and licensing compliance; the skill does not automatically download or execute content from these locations during its operation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 08:11 PM