ops-engine
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes local shell scripts distributed within the skill package for validating output structure and testing logic.
- Evidence:
scripts/validate-output.shandtests/run-validation.shuse standard shell utilities (grep,cat,echo) to enforce the risk/plan/checks output contract. - The scripts perform no network operations, access no sensitive user directories, and are focused purely on instructional linting.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external infrastructure manifests and CI/CD configurations, which could potentially contain malicious instructions.
- Ingestion points: The skill uses
read_fileto ingest configuration files and deployment manifests from the project workspace. - Boundary markers: Workflow step 1 mandates defining system boundaries and change surfaces before proceeding with analysis.
- Capability inventory: The skill generates commands and plans for the user to review and execute; it does not autonomously execute destructive shell commands or network operations.
- Sanitization: The skill enforces safety through mandatory human-in-the-loop gates (Workflow step 5) and explicit abort criteria for unacceptable risks (as demonstrated in
assets/examples/edge-case.md). - [SAFE]: The skill implements extensive operational safety rules and best practices.
- Evidence: Mandatory human approval is required for all production deployments, destructive operations, and irreversible rollbacks.
- Evidence: Guidelines in
rules/kubernetes-safety.mdandrules/deployment-safety.mdemphasize least privilege, health verification, and blast-radius control.
Audit Metadata