terraform-drift-detection
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection (MEDIUM): The skill is vulnerable to indirect injection because it ingests data from external AWS resources into the agent's decision-making loop.
- Ingestion points:
drift.json(generated fromterraform show -json). - Boundary markers: Absent. The agent processes the raw JSON output to categorize drift and recommend actions.
- Capability inventory: Execution of
terraform applyandterraform apply -refresh-onlybased on agent recommendations. - Sanitization: Absent. Maliciously crafted AWS resource tags or names could influence the agent's categorization or impact assessment.
- Command Execution (MEDIUM): The skill explicitly executes system commands including
aws sts get-caller-identity,terraform init, andterraform plan. While essential for the skill's purpose, this grants the agent control over the cloud environment's state and configuration. - Data Exposure (MEDIUM): Terraform state and plan files (
drift.json) often contain sensitive plain-text information such as database passwords, API keys, or private environment variables. Reading this data into the LLM context constitutes a risk of sensitive data exposure to the model provider.
Audit Metadata