Capstone: Designing an Agent Team for website-content
Claude Code Agents: Fundamentals
Chapter 11 · Capstone: Designing an Agent Team for website-content
Every prior chapter covered one agent type in isolation. This capstone applies all seven — teaching, coding, code-review, testing, refactoring, documentation, and research — to one real, already-familiar project: website-content, the very course-generation system this course itself was built inside. Rather than a hypothetical scenario, this walks through which of these agents would genuinely help with the real work this project actually requires, and why.
The Scenario
A new course chapter needs to be generated: written in full compliance with the project's own rules (the required banner comment, the code-block/copy-button convention, the correct wrapper CSS class and folder placement), checked for accuracy against existing cross-references, and — once the course is actually finished — the project's own tracking files need updating to reflect what was really built. This is the exact workflow this project follows for every chapter, done here deliberately with a real agent assigned to each step.
Step 1 — Confirming the Convention Before Writing Anything (Chapter 10)
Before generating anything new, a research agent, given a thorough search breadth, checks the project's own mapping and index files for an existing entry under a different name or prefix — exactly the kind of overlap check that catches a proposed course accidentally duplicating one that already exists, before a single line of content gets written.
Step 2 — Explaining an Unfamiliar Convention (Chapter 4)
If someone new to this project needs to understand why every chapter requires a specific banner comment, or what the copy-button convention actually requires and why, a teaching agent — briefed with what that person already knows about HTML/CSS generally — walks them through the project's own specific conventions, rather than requiring them to read every rules file cover to cover unassisted.
Step 3 — Writing the Actual Chapter (Chapter 5)
The chapter itself is implemented by a coding agent, briefed with the exact course name, chapter number, accent color, wrapper class, and content outline — matching this project's own established per-course template precisely, the same way every chapter in this very course was actually produced.
Step 4 — Checking Rules Compliance (Chapter 6)
Rather than assuming the newly-written chapter is compliant, a code-review agent — restricted to reading only, no editing — checks it against the project's actual rules: is the banner comment present and correctly formatted, does every code block use the site's established copy-button markup, is the wrapper class consistent with the rest of the course. It reports findings; it never quietly fixes them itself.
Step 5 — Verifying Nothing Broke During a Reorganization (Chapter 7)
When files get moved between folders — exactly as happened during this project's own recent folder-structure migration — a testing-agent-style verification pass confirms the numbers genuinely reconcile: every file accounted for, none silently lost, matching the kind of before/after reconciliation check that migration actually relied on.
Step 6 — Restructuring Old Content Without Changing Its Meaning (Chapter 8)
Upgrading an old single-line course-header comment to the project's current, fuller banner format, or wrapping an existing code block in the copy-button markup, is a refactoring task in exactly this course's own sense: the lesson's actual content and meaning must stay identical — only its formatting structure changes, verified by confirming the visible lesson content reads the same before and after the change. This is a genuine refactoring agent job, not a coding-agent one.
Step 7 — Updating the Project's Own Tracking Files (Chapter 9)
Once a course is genuinely finished, the project's tracking files need updating to reflect the real, current state — describing what was actually built, verified against the real generated files, not against whatever a prior outline said before the course started. This is a documentation agent's job specifically: describing current reality accurately, not modifying the course content itself.
| Scenario Step | Agent Used | Chapter |
|---|---|---|
| Step 1 — Confirming the convention first | Research/Explore Agent | Chapter 10 |
| Step 2 — Explaining an unfamiliar convention | Teaching Agent | Chapter 4 |
| Step 3 — Writing the chapter | Coding Agent | Chapter 5 |
| Step 4 — Checking rules compliance | Code-Review Agent | Chapter 6 |
| Step 5 — Verifying a reorganization | Testing Agent | Chapter 7 |
| Step 6 — Restructuring old formatting | Refactoring Agent | Chapter 8 |
| Step 7 — Updating tracking files | Documentation Agent | Chapter 9 |
Hands-On Exercises
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.
📄 View solutionExplain why Step 6 (upgrading an old banner comment format) is described as a refactoring task rather than a coding task, using this course's own Chapter 8 distinction.
📄 View solutionA developer decides to build and deploy all seven of these agents before generating even a single course chapter for this project. Using this chapter's own warning box, explain what consideration they should weigh first.
📄 View solutionClaude Code Agents: Fundamentals — 11 of 11 chapters complete. Claude Code Agents: Advanced Orchestration remains outstanding as this two-course track's own next planned course.
Chapter 11 Quick Reference
- Every agent type from this course maps to a real, specific step in this project's own actual workflow
- Research agent confirms conventions first, teaching agent explains them, coding agent implements, review agent checks compliance, testing agent verifies reorganizations, refactoring agent restructures old formatting, documentation agent keeps tracking files current
- None of these steps requires a new capability beyond what Chapters 1–10 already covered — this capstone only applies them together
- Formalizing a full agent team has its own real setup cost — it pays off at repeated volume, not for a single one-off task