p-image
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references installation instructions and documentation from the
inference.shofficial domains and GitHub repositories (github.com/inference-sh/*,inference.sh). It suggests installing thebelt-sh/clitool usingnpx, which is standard for this ecosystem. - [INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface for indirect prompt injection via the processing of external images.
- Ingestion points: The
imagesinput parameter in thepruna/p-image-editandpruna/p-image-edit-loramodels allows processing of remote URLs. - Boundary markers: No explicit boundary markers or instructions to ignore embedded content are present in the provided examples.
- Capability inventory: The skill uses the
Bashtool to executebeltCLI commands. - Sanitization: The skill instructions do not include validation or sanitization steps for the content of the external image URLs provided at runtime.
Audit Metadata