The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell tools specifically restricted to terraform:* and gcloud:* commands for provisioning and validating infrastructure components.
[DYNAMIC_EXECUTION]: The skill dynamically generates Terraform HCL files (.tf) and shell scripts, which are then processed or executed by the agent to manage state and infrastructure.
[PRIVILEGE_ESCALATION]: The documentation in references/errors.md provides troubleshooting advice to temporarily grant roles/owner to service accounts, which encourages a deviation from least-privilege security practices.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes infrastructure configurations that could be modified by external actors. 1. Ingestion points: Infrastructure requirement descriptions from the user and existing .tf files found in the project directory via Grep and Glob tools. 2. Boundary markers: None implemented in the instructions or reference templates. 3. Capability inventory: The skill possesses file write access (Write, Edit) and restricted command execution (Bash). 4. Sanitization: No validation or sanitization mechanisms are described for the content of processed configuration files.
[EXTERNAL_DOWNLOADS]: The skill uses Terraform, which by design downloads provider plugins from the Hashicorp registry (e.g., hashicorp/google). These are well-known and trusted sources.

adk-infra-expert