What problem does it solve? AI agents often hallucinate gcloud commands, flags, and syntax, run destructive operations without authorization, or flood the context window with unfiltered output. This Skill enforces validated, safe, and minimal gcloud CLI usage for managing Google Cloud resources. ## Core Features & Use Cases - Mandatory Command Validation: Requires running gcloud help <command> at the leaf level before executing any command to prevent hallucinated syntax. - Safety Guardrails: Denylists destructive operations (deletes, IAM changes, billing, KMS) and mandates --dry-run previews and explicit user authorization for risky actions. - Data Reduction: Enforces --format, --filter, and --limit flags plus schema discovery via single-item list queries to protect the context window. - Use Case: When asked to investigate why a Compute Engine instance is unreachable, the agent first validates the exact describe command, scopes it with --project and --zone, and projects only the needed fields instead of dumping full resource JSON. ## Quick Start Ask the agent to list all running Compute Engine instances in your project using the gcloud skill with filtered, projected output.
