save-report
Pass
Audited by Gen Agent Trust Hub on Jul 1, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection. It processes untrusted user data (medical images) via a vision tool and uses the output to perform file operations.
- Ingestion points: The
mcp__4_5v_mcp__analyze_imagetool reads content from user-provided image paths inSKILL.md. - Boundary markers: The prompt templates provided in Step 3 do not include instructions to disregard potential commands or adversarial text embedded within the medical reports.
- Capability inventory: The skill has
Writepermissions to the filesystem to save JSON data and images. - Sanitization: While the output is validated against a JSON schema (
schema.json), this does not prevent the LLM from being influenced by instructions found inside the image during the analysis phase.
Audit Metadata