Skip to content

AGT-05 · Human-in-the-loop: approve, edit, reject, time out and audit

Approval steps are bolted on late, so decisions get lost, threads cannot resume after a deploy and risk teams cannot see who approved what. Queues back up and agents wait forever because nobody planned for timeouts.

  • Who reads it: GenAI engineers building approval flows, DevOps engineers running them, and risk and compliance reviewers
  • Outcome: Every risky action approved, audited and never stuck
  • Pack owner: James (catalog owner role: Agent Lead)

Master catalog entry

  • Catalog name: Human-in-the-Loop Interrupt Patterns (P1)
  • Why it matters: The control that makes agents acceptable to risk and compliance teams
  • What it removes: Approval, escalation and resume flows already designed
  • Stack: LangGraph interrupts, checkpointers, Teams and email approval
  • Maturity target: Level 3 — approve, edit, reject and timeout paths with audit trail

Playbooks

ID Story Title Kind Complexity Urgency Status Owner Planned publish
PB-AGT-05-01 AGT-05-S1 Build approve, edit, reject and timeout paths with LangGraph interrupts How-to Medium 4 draft Sanjay Sarvesh 2026-11-24
PB-AGT-05-02 AGT-05-S2 Deliver approvals in Teams or email and resume the agent from a callback How-to High 3 draft James 2026-12-03
PB-AGT-05-03 AGT-05-S3 Clear an approval queue backlog and handle timeouts Runbook Medium 5 draft James 2026-11-05
PB-AGT-05-04 AGT-05-S4 Resume or recover paused threads after a deploy or graph change Runbook High 4 draft Deeparani 2026-11-16
PB-AGT-05-05 AGT-05-S5 HITL audit trail and compliance sign-off checklist Checklist Low 3 draft Sanjay Sarvesh 2026-12-03