pixel-art

Pass

Audited by Gen Agent Trust Hub on Jul 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local Python script, scripts/quantize.py, to process generated images. This script performs grid-alignment and palette quantization using the Pillow library.
  • [EXTERNAL_DOWNLOADS]: The skill requires the Pillow image processing library, which is installed from the official Python Package Index (PyPI) as specified in scripts/requirements.txt.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests user-provided text briefs and reference images to generate prompts.
  • Ingestion points: User-supplied image briefs and reference screenshots (documented in SKILL.md).
  • Boundary markers: Employs a universal block structure (e.g., [STYLE], [PALETTE], [SUBJECT]) within templates to separate instruction components (documented in templates/scene-prompt.md and others).
  • Capability inventory: Execution of the scripts/quantize.py script for file processing and invocation of image generation tools via host-native capabilities or MCP tools.
  • Sanitization: No explicit sanitization of user-provided text is described before interpolation into the prompt blocks.
  • [SAFE_PRACTICE]: Implements a mandatory Phase 0 gate that explicitly declines requests requiring exact text, layout, or brand tokens, ensuring the agent does not attempt out-of-scope composition tasks.
  • [SAFE_PRACTICE]: Includes a non-negotiable IP guardrail that forbids referencing specific artists in prompts and instructs the agent to rewrite any user requests that name artists into descriptive craft markers.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 11, 2026, 07:29 AM
Security Audit — agent-trust-hub — pixel-art