psd-to-preview
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes Python scripts (psd_layers.py, export_slices.py, and generate_preview.py) to process design files. These scripts are invoked from the local filesystem within the .claude/skills/ directory.
- [EXTERNAL_DOWNLOADS]: The agent is instructed to check for and install the psd-tools and scikit-image libraries using the pip package manager from the Python Package Index (PyPI) if they are missing.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from PSD design files. It interprets layer and group names as layout instructions (e.g., [flow-y], [fixed]), which could be manipulated to influence the generated output or the agent's interpretation of the design.
- Ingestion points: PSD files specified by the user in the command arguments.
- Boundary markers: None identified in the parsing commands; names are processed as-is to derive CSS and layout properties.
- Capability inventory: Execution of Python scripts, file system writes, and package installation via pip.
- Sanitization: None; the skill explicitly includes a --preserve-names flag that maintains the original (potentially malicious) layer names in the output.
Audit Metadata