The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file evals/files/legacy_network.tf contains a hardcoded password Sup3rSecret! assigned to a variable. While this is provided as a 'bad' example for evaluation purposes, it represents a plaintext credential exposure.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to recommend or use external tools such as tflint, trivy, checkov, and terraform-docs. These are standard, well-known tools within the DevOps and infrastructure-as-code community used for linting, security scanning, and documentation generation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external Terraform and OpenTofu configuration files, which can contain untrusted instructions.
Ingestion points: Files such as evals/files/legacy_network.tf, evals/files/drift_backend.tf, and evals/files/artifact_bundle.tf are read into the agent's context for review and refactoring tasks.
Boundary markers: The skill does not define specific delimiters or instructions to prevent the agent from following directives embedded within the processed HCL files.
Capability inventory: The agent is instructed to execute a wide range of terraform and tofu commands, including plan, apply, test, and state operations, which interact with infrastructure and external providers.
Sanitization: The skill relies on standard HCL parsing and manual plan verification without additional sanitization of the input text or specific safety guardrails for the agent's interpretation of HCL comments or metadata.

terraform