regulatory-guard
Warn
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: In SKILL.md, the 'File read' capability is explicitly intended to access 'env' and configuration files. While this is justified by the skill's purpose to check for regulatory compliance (e.g., privacy posture), accessing environment files is a high-risk activity as they typically contain sensitive credentials and system secrets.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external websites and local application source code without explicit sanitization or boundary markers.
- Ingestion points: Browser-based inspection via Playwright, network capture via curl/fetch, and local source code reads (SKILL.md).
- Boundary markers: The instructions lack specific delimiters or directions for the agent to ignore instructions embedded within the data it audits, which could allow malicious content to hijack the agent's behavior.
- Capability inventory: The skill employs browser navigation, network requests, file reading, and sub-agent dispatching (SKILL.md, references/privacy.md).
- Sanitization: There is no evidence of filtering, escaping, or validation mechanisms applied to the untrusted content gathered during the audit phases.
Audit Metadata