cloud-infrastructure — agentic threat model

Not certain from the listing — relies on Claude Code (Anthropic Claude models). Vulnerable to prompt injection and adversarial jailbreaks that could force the model to generate malicious IaC templates containing backdoors, overly permissive IAM policies, or hidden exfiltration routes.

Not certain from the listing — likely ingests local repository files, existing Terraform state, and cloud configuration schemas. Vulnerable to data poisoning if an attacker injects malicious metadata or manipulated state files into the workspace, leading to corrupted architecture generation.

The agent orchestrates subagents for AWS, Azure, GCP, and OCI. Vulnerable to tool misuse and insecure tool integration; if the planning loop is hijacked, the agent could execute destructive Terraform commands (e.g., terraform destroy) or call APIs that expose sensitive infrastructure.

Runs as a Claude Code plugin, likely executing in the user's local terminal or CI/CD environment. If the execution environment lacks strict sandboxing, the agent has direct access to local cloud credentials (~/.aws/credentials, kubeconfig), posing a severe risk of credential theft and host compromise.

Not certain from the listing — there is no mention of built-in dry-run validation, policy-as-code guardrails (like OPA/tfsec), or real-time monitoring of the generated IaC before it is applied to production environments.

Not certain from the listing — lacks explicit mention of identity federation, role-based access control (RBAC), or compliance auditing. The agent operates with whatever ambient cloud permissions the host system or user session possesses.

Bundles multiple subagents across AWS, Azure, GCP, and OCI. Vulnerable to cascading failures and trust abuse where a compromise in one cloud subagent (e.g., OCI) propagates malicious configurations or lateral movement to another (e.g., AWS) via hybrid networking setups.