photo-composition-critic
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes Python scripts in
references/analysis-scripts.mdintended for execution via theBashtool. These scripts use standard libraries such as NumPy, Pillow, and CLIP to perform legitimate image analysis tasks, including visual weight calculation, centroid detection, and Rule of Thirds alignment. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes untrusted data from local images and external web search results.
- Ingestion points: Local image files provided to analysis tools and search result snippets from the
mcp__firecrawl__firecrawl_searchtool. - Capability inventory: The skill utilizes the
Bashtool for script execution andWrite/Edittools for file operations. - Boundary markers: No explicit delimiters or instructions to ignore embedded content were found in the prompt templates.
- Sanitization: No specific sanitization or filtering logic is implemented for the data ingested from external sources.
- [EXTERNAL_DOWNLOADS]: The Python implementations in
references/analysis-scripts.mdutilizeclip.load(), which downloads pre-trained machine learning weights from well-known sources (OpenAI/GitHub repositories) to facilitate aesthetic scoring.
Audit Metadata