HIL-DEC-01Tier IHuman-in-the-Loop Checkpoints

Approval Gate for Irreversible Actions

Require explicit human approval before the agent executes actions that are irreversible, high-value, or externally visible.

Runtime: DecideAgentOps: Design & ScopeSee on the matrix →
APPROVAL GATE FOR IRREVERSIBLE ACTIONS Agent Planproposedaction Classifierreversibility· $ Approval UIplan +reasoning + Humanapprove ·modify · ExecuteloggedApproval UI shows the full chain — never just "Approve?".
In Plain English

This page is the complete instruction page for one pattern called "Approval Gate for Irreversible Actions." It explains the problem this pattern solves, the idea behind the solution, when you should (and should not) use it, and what happens afterward — both the good effects and the costs.

This matters because building AI agents is not just about making them clever. It is also about making them safe and predictable. Following a well-tested pattern like this one helps avoid common mistakes, and shows you exactly which safety rules and regulations it connects to, listed under "Standards Mesh" on this page.

Context

Some actions cannot be undone — payments, public communications, data deletions, production deploys. Autonomy that includes these requires a meaningful checkpoint, not a theatrical one.

Problem

Autonomous execution of irreversible actions makes every model error a potential incident.

Forces

  • Autonomy and speed vs safety
  • Meaningful review vs rubber-stamp fatigue

Solution

Define an irreversibility threshold (e.g. any write outside sandbox, any external communication, any transaction above X). When the agent plans such an action, it halts and presents: proposed action, full reasoning chain, retrieved evidence, and risk classification. Human approves, rejects, or modifies. Decision and rationale are logged.

Applicability

  • Any write-capable agent in production
  • Financial, medical, legal workflows
  • External communications agents

Anti-Patterns

  • Approval UI showing only the action without reasoning
  • Auto-approving after N seconds of inactivity
  • Same person acting as operator and approver

Consequences

  • +Meaningful human control
  • +Last line of defence against model error
  • Workflow latency
  • Approval fatigue risk