| name | human-in-the-loop |
| description | Activate this skill before opening pull requests, modifying production configurations, or executing deployment and destructive commands to ensure proper human-in-the-loop verification. |
Human-in-the-Loop Integrations
To prevent destructive automated actions and ensure test-driven stability, the AI agent MUST strictly adhere to the following stop-and-wait verification gates before taking critical actions.
1. Pull Request Verification Gates
Before finalizing and opening a Pull Request on GitHub, you MUST obtain explicit confirmation from the human developer that local checks are passing.
- The Test Check: Before executing
gh pr create or submitting a pull request, you MUST explicitly ask the user: "Are all tests passing locally?"
- The Wait Constraint: You MUST END your turn immediately after asking the question. You are FORBIDDEN from executing the PR creation command until the human user responds explicitly with an affirmative (e.g., "Yes").
2. Deployment & Destructive Task Gates
Any operations that irreversibly alter external state, infrastructure, or databases MUST be explicitly verified.
- The Deployment Check: Before executing any command that deploys code, drops databases, runs structural migrations, or pushes to non-feature branches, you MUST explicitly ask the user: "Are you sure you want to proceed with [action]?"
- The Wait Constraint: You MUST END your turn immediately after asking the question and await explicit approval.
Last Updated: 2026-07-22 | Last Reviewed: 2026-07-22
