This is a comprehensive safety harness for Claude when running gcloud commands. It enforces mandatory command validation via help text before execution (to prevent hallucinations), explicit project and location scoping to avoid interactive prompts, and data reduction strategies like projections and filters to save context window. The denylist blocks destructive IAM operations unless explicitly authorized. What stands out is the authentication guidance, especially the recommendation to use service account impersonation for local pair programming, which keeps agent permissions separate from human credentials. If you're building an agent that needs to query or manage GCP resources without accidentally nuking something, this skill provides the guardrails and workflow patterns to do it safely.
npx -y skills add google/skills --skill gcloud --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills