The Agent Skills Directory

[Command Execution]: The skill directs the agent to use the agents-cli tool for managing infrastructure and deployments. To ensure safety, it explicitly mandates that the agent must obtain human approval before executing any deployment commands, which prevents unauthorized or accidental infrastructure changes.
[Credential Management Best Practices]: The instructions emphasize using secure methods for secret handling, such as Google Cloud Secret Manager and Workload Identity Federation (WIF). This approach follows security best practices by minimizing the risk of credential exposure compared to traditional methods like hardcoded keys or long-lived service account tokens.
[External Resource Retrieval]: The skill references and retrieves documentation from adk.dev and installation packages from established registries like PyPI. These are recognized vendor resources and well-known services used to provide up-to-date documentation and tools required for the deployment workflow.
[Indirect Prompt Injection Surface]: The skill involves a surface for indirect prompt injection because it processes data from project files (e.g., Dockerfile, .env) and external documentation to perform its tasks. While this is necessary for its function, it represents a potential vector if project files are maliciously crafted.
Ingestion points: Project configuration files and remote documentation URLs.
Boundary markers: The skill utilizes explicit human confirmation for high-stakes actions like deployment, which serves as a supervised boundary.
Capability inventory: Access to system CLI tools including agents-cli, terraform, gcloud, and kubectl for infrastructure management.
Sanitization: The skill does not implement automated sanitization of file content, relying instead on the agent's internal logic and the mandatory human review step before execution.

google-agents-cli-deploy