CLAUDE CODE AGENTS: FUNDAMENTALS - Chapter 11 (Capstone), Exercise 1
Why a Research Agent Handles the Convention Check, Not the Coding Agent
====================================================================================

QUESTION:
Explain why Step 1 (confirming the convention) uses a research agent rather
than simply having the coding agent check for an existing course itself
before writing.

SOLUTION / EXPLANATION:

This comes back to Chapter 1's own isolation principle and Chapter 10's own
description of what a research agent is specifically built for: locating
information and answering structural questions, with tool access
restricted to reading and searching only - no Edit or Write at all.

A coding agent, per Chapter 5, is designed and tooled for a genuinely
different job: implementing a specific, already-scoped task. If the same
agent were also responsible for first determining whether that task's
premise is even valid (does this course already exist under another name?
is the prefix actually free?), it would be doing two structurally different
kinds of work inside one agent - open-ended verification, and then
committed implementation - with no clean separation between "still
figuring out if this is the right thing to build" and "actively building
it."

Using a dedicated research agent for Step 1 keeps that verification step
cleanly separated and read-only: it can search course_folder_mapping.md,
course_name_index.md, and completed_courses.md thoroughly, with zero risk
of accidentally starting to write files before the premise is actually
confirmed, since it has no Edit or Write access to do so even if it wanted
to. Only once the research agent's findings confirm the convention is
genuinely free does the actual implementation step (Step 3, the coding
agent) begin - a clean, sequential separation between confirming the
premise and acting on it, rather than one agent doing both at once with
less discipline about the difference.

--------------------------------------------------------------------------
WHY THIS WORKS AS AN ANSWER:
It connects the separation directly to the isolation principle and the
specific tool restrictions each agent type carries, explaining why mixing
verification and implementation inside one agent would blur a distinction
this course treats as structurally important throughout.