Agent permissions, actions and agents at work
Agent instructions
Writing instructions an agent can run
- Turning a Written Instruction into a Runnable Checklist
- Writing a Handoff Brief an Agent Can Follow End to End
- Building a Small Golden Set for Your Instructions
- Checklists That Run as Code
- Routing for the First Pass: Which Cheap Model Handles What
- Structured Outputs for Tool Calls: Schema and Strict Mode
- Which Model Can Follow Your Brief: A Short Test Protocol
- Writing a Policy Document That Both People and Agents Can Follow
Tool permissions
Tool permissions and scopes
- Auditing a Manifest Change: What an Agent May Do Differently
- Authority Levels for Human and Agent Work: A Table You Can Paste
- How to Write a Tool Permission Manifest for an Agent
- Scoping an Agent's Tools: Read, Write and Destructive
- Separating Data from Commands in a Tool Call
- Tool Arguments as an Attack Surface: Validating What Agents Pass
- Designing Narrow Tool Interfaces for Agents
- Manifest Review in CI: Blocking Merges When Reach Widens
- All of Tool permissions
Destructive actions
Safe destructive actions
- Splitting a Destructive Tool into Prepare and Commit
- Expiring Prepared Actions: TTLs and Revalidation Before Commit
- Idempotency Keys for Agent Writes: Retries Without Duplicates
- Reversing a Committed Agent Action: The Compensation Step
- Revoking an Approval Before Commit: Cancel, Expire, Confirm
- Setting Call Caps and Budgets for Agent Tools
- Writing a Dry-Run Mode Every Destructive Tool Can Reuse
Approvals and audit logs
Reviewers, approvals and audit logs
- Recording an Approval as Structured Data: Who, When, Against What
- Who Can Approve: Mapping Risk Tiers to Reviewers
- A Log Schema for Tool Arguments and Outcomes
- Approval Queues for Agent Work: Design, Backlog and Throughput
- Auditing an Agent's Approval Chain from Logs Alone
- Auditing What an Agent Did: Reconstructing Decisions from Logs
- Escalation Paths at an Agent's Authority Limit
- Fields to Log When an Agent Calls a Tool
- All of Approvals and audit logs
Graders and checklists
Graders and checklists
- Grading With a Rubric vs a Reference: Which Catches More
- Routing Between a Cheap First Pass and an Expensive Second Look
- When to Let an Agent Mark Its Own Checklist
- Writing the Grader Prompt for a Second-Model Check
- A Prompt Injection Test Suite for Inbox and Ticket Agents
- Building a Labeled Set When No One Has Time to Label
- Calibrating a Confidence Threshold: From Scores to Decisions
- Choosing a Second Model: Cost, Latency, Independence for Graders
- All of Graders and checklists