Tool Registries and Permissioned Toolsets

This build lab focuses on the tool registry that turns raw capabilities into permissioned agent actions. The goal is not to copy a private machine setup. The goal is to learn the architecture pattern well enough to build a small, classroom-safe version.

Tools need names, schemas, descriptions, permission levels, validators, and runtime approval rules before a model can call them safely.

Build the small version

1. 1Draw or write a registry table for five tools with inputs, outputs, risk level, and approval behavior.
2. 2Mark which parts are user-facing, which parts are internal, and which parts require approval.
3. 3Choose one low-risk workflow and implement only that workflow first.
4. 4Add one failure case before adding a second feature.
5. 5Write a short operator note: what the agent may do, what it must ask about, and what it must never do.

tools:
  lesson_search:
    inputs: {query: string}
    risk: low
    approval: never

  send_email:
    inputs: {to: email, subject: string, body: string}
    risk: medium
    approval: draft_then_confirm

  run_shell:
    inputs: {command: string}
    risk: high
    approval: explicit_human_only

The big idea: allowlist is not decoration. It is part of the product architecture students need before an agent becomes safe enough to use with real people.