visual-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The helper script
scripts/concat-images.mjsexecutes system commands for image processing (magick,convert,montage). While it utilizes a shell (sh -c) for command discovery, the inputs to these calls are hardcoded, and the primary execution logic avoids shell interpolation.\n- [DYNAMIC_EXECUTION]: The script useschild_process.spawnSyncto manage image manipulation tasks. It mitigates injection risks by passing arguments as arrays and enforcing strict regex validation on parameters like--crop.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes image filenames and captions, which represent an attack surface for indirect prompt injection. The security posture is maintained through the following implementation details:\n - Ingestion points: Image file paths and caption strings provided to
scripts/concat-images.mjs.\n - Boundary markers: None explicitly used for caption text in labels.\n
- Capability inventory: Process execution of
magick,convert, andmontageviaspawnSync.\n - Sanitization: Arguments are passed via array to
spawnSyncto prevent shell injection, and geometry inputs are validated against a strict regular expression.
Audit Metadata