intern-interview-intake
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses sensitive user media in the Photos library and exports it to the /tmp directory for processing.
- [COMMAND_EXECUTION]: The skill instructs the agent to perform OCR using Tesseract and to programmatically delete processed images from the Photos library, which carries an inherent risk of accidental data loss.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from screenshots via OCR without sanitization or boundary markers. 1. Ingestion points: Photos library media (SKILL.md). 2. Boundary markers: None identified. 3. Capability inventory: File system writes to $WORLD_MODEL_VAULT and deletion of Photos library items (SKILL.md). 4. Sanitization: None identified.
Audit Metadata