vision-framework
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill provides code patterns for ingesting and processing untrusted data from visual sources.
- Ingestion points: The skill documents ingesting untrusted data via RecognizeTextRequest, DetectBarcodesRequest, and DataScannerViewController in SKILL.md and reference files.
- Boundary markers: The provided Swift code snippets do not include boundary markers or explicit instructions to treat recognized content as untrusted data.
- Capability inventory: The skill shows how to extract recognized strings and payloads, including a pattern in references/vision-requests.md that initializes URL objects directly from untrusted barcode payloads.
- Sanitization: No sanitization, escaping, or validation of the extracted text or payloads is performed in the example implementations.
Audit Metadata