id-capture-cordova
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill enables the ingestion and processing of data from physical and mobile identity documents (passports, driver's licenses) via MRZ, VIZ, and barcode scanning. While processing external, untrusted document content represents an indirect prompt injection surface, the skill is scoped to structured data extraction and includes explicit guidance to avoid logging Personal Identifiable Information (PII), which mitigates common risks associated with this vector.
- [CREDENTIALS_UNSAFE]: The skill manages Scandit license keys using specialized tools. The instructions provide a secure handling policy by directing the agent to show only masked previews of keys and never print full credentials in the chat interface. Code examples correctly use placeholders for user-supplied keys.
- [PROMPT_INJECTION]: The skill includes a 'Do Not Trust Internal Knowledge' directive. Analysis confirms this is a benign instructional prompt intended to ensure the agent uses the provided SDK v8 documentation instead of potentially outdated training data, rather than an attempt to override safety filters or system prompts.
- [SAFE]: The external plugins and repositories referenced (e.g., Scandit's official documentation and samples) originate from a recognized vendor infrastructure and are used for their intended purpose in a developer-facing integration guide.
Audit Metadata