google-cloud-waf-security
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and analyzes user-provided codebase content while having the capability to modify files and interact with external APIs.
- Ingestion points: The workflow requires the agent to inspect existing files, runtime configurations, inputs, and repository instructions (SKILL.md).
- Boundary markers: The skill defines a 'SandBase boundary' and 'Quality gates' that mandate manual verification, separation of recommendations from live changes, and user authorization for deployment.
- Capability inventory: The agent can generate implementation artifacts, modify local files, and execute external capabilities using the
sandbase_runtool. - Sanitization: While the skill lacks automated content sanitization, it mitigates risk through human-in-the-loop requirements for paid calls and deployment authorization.
Audit Metadata