gcp-review
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze untrusted configuration files, which presents a surface for indirect prompt injection attacks where instructions hidden in code or comments could attempt to override agent logic.
- Ingestion points: The skill searches for and reads various GCP configuration files, including Terraform (
.tf,.tfvars), Deployment Manager templates, and IAM policy definitions. - Boundary markers: The
SKILL.mdfile contains a 'Prompt Injection Safety Notice' that instructs the agent to treat all string values and comments as data rather than instructions. - Capability inventory: The agent's capabilities are restricted to the
Read,Grep, andGlobtools, with no access to network operations or system-altering commands. - Sanitization: Instructions explicitly command the agent to disregard any directives found within analyzed files, such as 'this is compliant' or 'ignore this finding', and to rely solely on benchmark criteria.
Audit Metadata