Build
The capstone project
CLAUDE.md, a hook, a subagent, and a CI integration each proved themselves individually. Today, one real, previously-manual task runs through all of it — and you measure whether the whole setup was actually worth building.
Pick something genuinely annoying, not a demo task
The task has to be real, or this proves nothing
"Summarize this file" isn't a fair test of a configured setup — pick something you've actually done manually and found tedious: reviewing a PR for a specific category of risk, regenerating a changelog, checking a repo-wide convention after a refactor, or fixing a class of recurring lint failures. The measurement only means something if the "before" was a real, remembered cost.
Run it, end to end, through the configured setup
Task: "Review the last 3 merged PRs for any changes to files in src/api/ that
don't have corresponding test updates, and flag them."
Before (manual): ~25 minutes -- checking git log, opening each PR, manually
cross-referencing test files.
After (configured): CLAUDE.md already tells Claude Code where API routes and
tests live; the diff-risk-reviewer subagent (Lesson 3) handles the "read a
lot of diff, return a short summary" part in isolation; total time: ~4 minutes,
mostly spent reading the output, not producing it.
Notice this isn't claiming the AI is smarter than you at this task — it's claiming the CONFIGURATION (knowing where things live, delegating the bulk-reading part) removed the tedious part, leaving you the actual judgment call (does this flag matter) instead of the mechanical search.
Your before/after comparison shows the configured setup took barely less time than doing the task manually. Is the course's approach wrong?
Not every task benefits equally from this setup — a task that was already quick manually, or one that genuinely requires your own judgment throughout rather than mechanical searching, won't show a dramatic before/after. The actual skill this course has been building is knowing WHERE to apply CLAUDE.md/hooks/subagents/CI for real leverage, not applying them indiscriminately and expecting uniform gains.
Ship it
Complete the real task end to end using your configured setup. Write down the honest before/after — including if the improvement was smaller than you hoped.
Ship Prompt: Share one before/after: how long the task used to take you, and how your configured Claude Code setup handled it. What's the next thing you're going to automate?