Browse AI Agents & MCP SECTION 3: Tool Definitions & Schemas

The Permission Boundary Designer

3 fill-in slots · from The AI Prompt Handbook for AI Agents & MCP

Your details

0/3 filled

The Permission Boundary Designer

Design permission boundaries for this tool surface.

Tools: [LIST, MARKING READ VS WRITE]
Agent context: [WHO IT ACTS ON BEHALF OF]
Sensitive operations: [LIST]

Produce:
1. Tools grouped by permission tier: read-only, reversible write,
   irreversible write, administrative
2. For each irreversible operation: what gates it
3. Which tools should never be available in the same agent, and why
4. A least-privilege configuration: the smallest permission set
   that lets the agent do its job
5. What the agent should be told about its own permissions, so it
   does not attempt what it cannot do
6. What happens on a permission denial — the message the agent
   receives and what it should do
7. Whether permissions are per-agent, per-user, per-session, or
   per-request, and the implications of each

Flag anything where the agent acting "on behalf of" a user could
exceed what that user could do directly. That confusion is where
privilege escalation lives.

Highlighted [slots] are still empty. Downloads are plain .txt — paste into any AI assistant.