Security

Agent Actions and Approvals

• A Policy File for Routing Agent Actions by Risk Tier
• Approval Gates for Agents: Which Actions Need a Human
• How to Write the Approval Request a Human Can Actually Decide On
• One Approval Box for Two Actions: When a Batch Is Too Big to Send
• Routing Agent Actions by Risk Tier With a Policy File
• Turning a Risk Tier Into an Approval Timeout
• When an Agent Action Is Not in the Policy File
• Which Repairs Need a Human: A Policy for Meaning Changes
• A Cost Matrix for False Tool Calls Versus Missed Actions
• A Dry-Run Mode That Uses the Policy File on Old Traces
• A Monthly Review That Shrinks the Fallback
• A Policy File That Splits Batches Before Approval
• A Queue Owner's Guide to Rejecting a Repair
• A Schema for the Action Log the Policy File Writes
• An Agent Incident Runbook: Kill Switch, Rollback and Blast Radius
• Approval Fatigue: Counting How Often Reviewers Say Yes
• Binding Two Approvers to One Agent Payment Action
• Business Hours and Time Zones in Timeout Tests
• Dual Approval for Agent Transfers
• Dual Control for Agent Payments: Two Approvers, One Transfer
• From Unlisted Action to a Rule: A Decision Record Template
• Kill Switches and Rate Limits for Agents That Act on Their Own
• Logging Abstain Decisions: Why No Tool Was Chosen
• Logging an Approval That Timed Out
• Logging Original and Repaired Arguments Side by Side
• Parking an Action Without Letting the Queue Rot
• Routing Agent Actions to Different Queues by Risk and Amount
• Routing Repeated Iterations to a Cheaper Model
• Sampling Approved Repairs to Find Rules That Overreach
• The Audit Event Fields for a Human Approval of an Agent
• Tuning the Abstain Threshold With a Cost Matrix
• Two Actions, One Approver: When One Reply Covers a Batch
• What a Pending Action Does to the Agent Loop
• What the Model Sees When an Action Waits for a Human
• What the Model Should Say When No Tool Fits
• What the Policy File Should Say About Money and Limits
• What to Log So the Next Unlisted Action Is Easy to Find
• What to Put in an Agent's System Prompt About Its Own Limits
• When an Approver Says No: Turning a Human Decline Into Eval Cases
• When the Agent Should Ask, Not Abstain
• Where the Approver's Signature Lives: Binding a Yes
• Where the Card Sits: Chat, Email or Dashboard
• Writing a Decline That Teaches the Agent Something
• Writing a Test Case for Every Rule in Your Policy File
• Writing Up an Agent Action That Fell Outside the Role