kuroco-security-audit
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is provided by the platform vendor (Diverta) and operates as a read-only auditor. It explicitly forbids tool calls that could modify site settings and contains strict instructions to avoid retrieving or logging actual secret values like tokens or passwords.
- [INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface as it processes external data from the Admin MCP to perform audits. Findings: 1. Ingestion points: The skill ingests configuration data via whoami, site_setting-get, api-list, and member-list tool calls. 2. Boundary markers: The skill uses a pre-defined security checklist (checklist.md) and instructs the agent to report unconfirmed or missing data as 'Manual Check' rather than assuming default safe values. 3. Capability inventory: The skill's scope is strictly limited to read-only Admin MCP tool calls; it does not possess file-writing or command-execution capabilities. 4. Sanitization: It enforces the use of specific columns filters when retrieving member data to avoid accidental PII collection and prohibits the reporting of sensitive credentials.
Audit Metadata