skills/terminalskills/skills/gcp-auth/Gen Agent Trust Hub

gcp-auth

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions provide templates for CLI commands that process external data such as project IDs and service account emails.
  • Ingestion points: Variable placeholders like my-project, alice@example.com, and prod-deploy@my-project.iam.gserviceaccount.com throughout SKILL.md.
  • Boundary markers: No explicit delimiters or instructions are provided to the agent to treat these interpolated values as untrusted data.
  • Capability inventory: The instructions involve high-privilege command execution via the gcloud CLI, including IAM policy modifications (gcloud iam service-accounts add-iam-policy-binding).
  • Sanitization: No sanitization or validation logic is defined for the variables before they are used in shell commands.
  • [METADATA_POISONING]: The author field in the skill's metadata is set to google-cloud, which is a trusted organization, while the skill's actual author context is terminalskills. This discrepancy in author identification can lead to incorrect assumptions regarding the official status or provenance of the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 06:01 PM
Security Audit — agent-trust-hub — gcp-auth