The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data, specifically Terraform plans, Kubernetes manifests, and Helm diff outputs. This creates a surface for indirect prompt injection if the processed files contain malicious instructions.
Ingestion points: Reads saved plans or diff output files (e.g., tfplan, kustomize build output) for tool-specific analysis.
Boundary markers: The skill instructs the agent to cite observed output or mark facts as unverified, which helps distinguish external data from internal logic, though explicit delimiters for external content are not mandated.
Capability inventory: The skill explicitly operates in a planning-only mode with no autonomous execution capabilities; it is prohibited from running CLI tools or reading credentials.
Sanitization: The skill does not explicitly mention sanitization of the plan output, but its "no-run" policy significantly mitigates the risk of data-driven command execution.
[COMMAND_EXECUTION]: While the skill contains numerous references to powerful CLI commands (e.g., terraform apply, kubectl rollout undo, pulumi up), it incorporates robust safety constraints to prevent autonomous execution.
Evidence: The instructions repeatedly state "OMH ran nothing," "OMH never runs terraform, tofu, pulumi, kubectl, or helm," and "every apply is the operator's."
[DATA_EXFILTRATION]: The skill demonstrates high awareness of sensitive data exposure risks inherent in IaC workflows.
Evidence: It provides specific warnings about Terraform plan files containing cleartext secrets and explicitly forbids the agent from attaching these files to PRs or pasting them into chat interfaces.

omh-iac-change