The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute a variety of external infrastructure tools including helm, terraform, tflint, tfsec, trivy, syft, and actionlint to analyze the target environment. These tools run locally on the files provided in the repository.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection because it ingests and processes untrusted data from infrastructure manifests (Kubernetes, Helm, Terraform) and CI/CD workflows, which could contain malicious instructions hidden in comments or metadata.
Ingestion points: Reads and renders Helm charts, Kustomize overlays, Terraform configs, and GitHub Actions workflows.
Boundary markers: No explicit delimiters are specified for separating untrusted tool output from the agent's internal reasoning context.
Capability inventory: The agent can execute shell commands for linting, templating, and security scanning.
Sanitization: The skill mitigates injection risks by instructing the agent to summarize tool findings rather than outputting raw data streams.
[SAFE]: The skill mandates strict safety boundaries by explicitly prohibiting state-changing operations such as apply, upgrade, delete, or destroy without direct user authorization. It also enforces the use of temporary directories for state files to prevent unauthorized persistence in the repository.

tools-infra-operational