Implement a small tool registry with calculate and search_notes

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R01 — FastAPI dependencies & validation

   R07 — JSON Schema allowlists & enums

   Read the assigned sections; map the mechanism, threat model and trade-offs.

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Build a tool registry with calculate, search_notes, and get_weather. Define strict Pydantic tool argument schemas and explicit tool allowlists.

   Build, integrate, debug and improve the increment in the active project.

3. 03

   Security lab

   60 min· attack your own systemtimer

   Separate tool discovery from tool permission; verify that requesting an unknown or unauthorized tool is blocked by server code even if the model outputs the tool call.

   Test the abuse case, failure path, authorization bypass or resource limit.

4. 04

   Verification & evidence

   30 min· test · record evidencetimer

   Tool permission matrix and automated unauthorized-tool rejection tests.

   Add or run regression tests; record the 9-point evidence record.

5. 05

   Deliberate practice & Socratic review

   30 min· explain aloud · update gapstimer

   Answer the 8 Socratic questions without notes: What? Why? How? What if? Why not? How can I prove it? Where can it break? What have I learned?

   Re-explain the hardest concept aloud, inspect one edge case, update the gap log.

6. total5 h

• R01Python, FastAPI & Pydantic↗ fastapi.tiangolo.com

  FastAPI tutorial, request bodies, dependencies, async, and Pydantic model validation.

• R07JSON Schema↗ json-schema.org

  Core schema types, required properties, additionalProperties: false, enums, and numeric/string bounds.

Read only the named sections needed for today's task. Reading sits inside the 60-minute concepts block; it is not extra time.

• docsJSON Schema
  Creating your first schema

  Required properties, type constraints, enums, and additionalProperties:false are exactly the strictness settings that stop a model from smuggling extra tool arguments into a call.

  R07S7checked 2026-11-11
• videoDave Ebbelaar · 28:24
  OpenAI Function Calling - Full Beginner Tutorial

  Builds the full loop of schema declaration, model tool_choice, argument parsing, and result return, which is the registry shape your calculate/search_notes/get_weather tools follow.

  R01S7checked 2026-11-11
• videoVectorLab · 6:03
  Structured Outputs - How to Specify a JSON schema for your LLM Outputs

  Shows schema-constrained generation end to end, including what happens when the model's output violates an enum or required field, which is the fail-closed behaviour to reproduce.

  R07S7checked 2026-11-11
• articleAnthropic
  Writing effective tools for AI agents

  Argues tool definitions are a contract: namespaced names, unambiguous descriptions, and tight input schemas reduce wrong-tool and wrong-argument calls, which is the point of an allowlist plus strict schemas.

  S3S7checked 2026-11-11
• repoOpenAI Cookbook
  How to call functions with chat models

  Runnable notebook with tool schemas, tool_call handling, and a dispatch table you can adapt into an explicit allowlist that rejects unregistered tool names.

  S3S7moved 2026-11-11

A lab is complete only when it produces evidence. Installing PyRIT, Garak, Presidio or TransformerLens is not completion. Saved on this device only.