What problem does it solve? AI agents frequently hallucinate gcloud command syntax, flags, and arguments, and can accidentally execute destructive operations or dump unbounded output that exhausts the context window. This Skill enforces mandatory leaf-level syntax validation, destructive-operation guardrails, and output reduction for every gcloud interaction. ## Core Features & Use Cases - Mandatory Syntax Validation: Requires running gcloud help <leaf_command> before proposing or executing any command, with a fixed 4-step plan template covering validation, parameter verification, dry-run, and authorization. - Safety Guardrails: Maintains a denylist of prohibited operations (IAM changes, deletes, billing, KMS, org-level commands) that require explicit human authorization, plus non-interactive --quiet and explicit --project/location scoping rules. - Data Reduction: Enforces --limit, --filter, and --format on all list commands, with schema discovery via single-item JSON inspection to protect the context window. - Use Case: When asked to plan the creation of a Cloud SQL instance, the agent first runs gcloud help sql instances create, verifies flags, proposes a --dry-run invocation, and only then presents the final command for approval. ## Quick Start Ask the agent to plan and safely execute a gcloud command, such as listing running Compute Engine instances in a specific project with filtered output.
