Tool Registry and Schema Generation

Module 6, Lecture 6.3 | Building the Coding Agent

This lecture is a software-engineering detour from agent behavior into agent maintainability. The if/elif dispatch from Lecture 6.2 works for three tools, but it duplicates information across the function definition, the API schema, and the dispatch chain — a pattern guaranteed to drift as the agent grows. The fix is a tool registry built with a Python decorator, paired with auto-generated API schemas derived from each function's signature and docstring through the inspect module. Along the way, the lecture introduces the Python features many students may be seeing for the first time — first-class functions, decorators, structured docstrings, and runtime introspection — because they will reappear in every agent framework students encounter later in the course. The lecture closes with defensive dispatch, the principle that the model uses error messages to self-correct, and a centralized dispatch function that becomes the natural home for logging and observability.

Read the full lecture narrative

Additional Resources

• Lecture slides
• Python docs — PEP 318: Decorators for Functions and Methods — the language proposal that introduced the @decorator syntax. Explains the rationale and the equivalence to func = decorator(func).
• Real Python — Primer on Python Decorators — accessible, example-driven walkthrough of decorators from "functions are objects" through real-world patterns. Recommended for students new to the @ syntax.
• Python docs — inspect module — the standard-library reference for runtime introspection of Python objects. Covers signature, Parameter, annotations, and the Parameter.empty sentinel used in the lecture.
• Python docs — PEP 257: Docstring Conventions — the canonical guide to docstring conventions in Python: triple quotes, the first-line summary, and how __doc__ is populated.
• Google Python Style Guide — Docstrings — the docstring format the lecture uses: Args:, Returns:, Raises: blocks with structured per-parameter descriptions.
• Sphinx — Napoleon extension — shows how Google-style and NumPy-style docstrings are widely consumed by tooling, reinforcing why the convention is worth following.
• Anthropic — Tool use with Claude — official reference for the tool-use protocol, including the JSON Schema requirements for input_schema, properties, and required.
• JSON Schema — Understanding JSON Schema — the underlying specification the Anthropic tool schema follows. Useful background for students writing their own type mappings beyond string/integer/boolean.
• Pydantic docs — Models — the production-grade alternative to hand-rolled inspect parsing. Shows how frameworks generate JSON Schema with full validation from typed Python classes.