lgpd-incident-response
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input when responding to reports of security incidents such as data breaches.
- Ingestion points: User descriptions of incidents (triggered by phrases like "vazamento de dados" or "tivemos um incidente") are ingested directly into the response workflow and templates.
- Boundary markers: The instructions do not define clear delimiters or specific warnings to ignore potential embedded instructions within user-provided incident details.
- Capability inventory: The skill involves creating and updating files in the
.lgpd/incidents/directory to log incidents. No specific external tools are invoked in the provided documents. - Sanitization: The skill lacks explicit instructions for sanitizing or validating user input before it is recorded in compliance logs or used to generate notification templates.
- [NO_CODE]: The skill is composed entirely of markdown documentation and templates. It contains no executable scripts (Python, JavaScript, shell) or dynamic context injection commands, which significantly reduces the potential attack surface.
- [SAFE]: The skill references official government websites for legitimate regulatory compliance purposes.
- Evidence: Links to the official ANPD portal (gov.br) are provided for incident reporting as required by Brazilian law.
Audit Metadata