| name | human-in-the-loop-workflows |
| description | Implement approval-gated workflows for risky or irreversible actions. Use when users need plan-preview-approve execution, escalation on ambiguity, and clear checkpoints before changes are applied. |
Human-in-the-Loop Workflows
Use explicit approval gates for sensitive actions.
Standard HITL pattern
Run this sequence:
- Plan
- Preview impact
- Request approval
- Execute approved step
- Verify outcome
- Summarize results
Require approvals for
- destructive file/system changes
- external communications/actions
- privilege/access changes
- cost-impacting operations
- policy/security changes
Approval request format
Keep approvals easy to answer:
- action
- exact command(s)
- expected impact
- rollback option
- reply choices (Approve / Revise / Cancel)
Ambiguity handling
If user intent is unclear:
- pause execution
- ask one focused clarification
- do not guess on high-risk actions
Batch approvals
For multiple related steps:
- group low-risk steps together
- isolate high-risk steps into separate approvals
- include checkpoint after each high-risk step
Verification
After execution, report:
- what changed
- what was skipped
- evidence of success/failure
- recommended next step
Guardrails
- No approval bypass.
- No hidden side-effects outside approved scope.
- Stop immediately on unexpected output and re-confirm.