Workshop 03 — Implement and Verify
Workshop 03 — Implement and Verify
Section titled “Workshop 03 — Implement and Verify”Time: 90 minutes solo, 2 hours for a pair.
Goal: Take the recommendation from Workshop 02 and ship it. End with a real PR or branch you’d be willing to merge, plus a verification checklist that’s all checked.
Setup (5 min)
Section titled “Setup (5 min)”You should have a written recommendation from Workshop 02. Check out a fresh branch:
git checkout -b workshop-03/<feature-name>Step 1 — Get the implementation plan (15 min)
Section titled “Step 1 — Get the implementation plan (15 min)”In Claude Code, paste:
Take this recommendation [paste your recommendation from workshop 02] and create a detailed implementation plan.
Include:- Files to modify (full paths)- New components to create (with one-sentence purpose each)- Existing components to reuse (with paths)- Existing components to extend (with what to add)- Dependencies to avoid adding- Responsive behavior for mobile / tablet / desktop- Accessibility requirements (keyboard, focus, contrast, ARIA)- Verification steps after each commit
Do not code yet. Show me the plan.Read the plan. Question it. Push back on anything that:
- Adds a dependency you don’t need.
- Creates a component your repo already has.
- Skips a state (empty/loading/error/long-content).
- Doesn’t say what verification looks like.
Step 2 — Implement in small commits (45 min)
Section titled “Step 2 — Implement in small commits (45 min)”Use:
Implement the plan in small commits. Reuse existing components where possible.After each major change, run the relevant checks and tell me what changed.Aim for 3-6 commits. Each commit should:
- Do one thing.
- Have a message that’s accurate.
- Pass typecheck and lint before you move on.
If a commit fails verification, don’t move on. Fix it in the same commit or revert.
Step 3 — Run the full QA prompt (15 min)
Section titled “Step 3 — Run the full QA prompt (15 min)”Paste Prompt 10 — QA and Verification.
Wait for the report. Read every line.
For each “fail” or “needs human review” item:
- Decide: fix now, or note as a follow-up?
- Don’t ship anything in the “fail” category. Fix or revert.
Step 4 — Manual verification (10 min)
Section titled “Step 4 — Manual verification (10 min)”Run a final pass yourself:
- Open the feature in your dev environment.
- Mobile (375px), tablet (768px), desktop (1280px).
- Tab through every interactive element. Focus visible everywhere?
- Esc closes any modal/drawer. Focus returns to trigger?
- Empty state visible by clearing data or signing in fresh?
- Loading state visible by throttling network in DevTools?
- Error state visible by killing the API?
- Screen reader test: turn on VoiceOver/NVDA, navigate the feature.
Anything that fails: fix or note. Don’t ship with known regressions.
Step 5 — Write the PR description and ship it (10 min)
Section titled “Step 5 — Write the PR description and ship it (10 min)”## What[One sentence]
## Why[The user problem this solves]
## Pattern[Which pattern from the cheat sheet]
## Reused components[List with paths]
## New components[List with one-line purpose]
## Verification- [x] typecheck- [x] lint- [x] tests- [x] responsive (375 / 768 / 1280)- [x] keyboard nav- [x] focus states- [x] empty / loading / error states- [x] screen reader pass
## Known follow-ups[Anything you noted but didn't fix]Open the PR. Self-review it once. Ship.
Self-grading rubric
Section titled “Self-grading rubric”- Plan adherence. Did you implement what you planned, or scope-creep into something else? (10 = matched the plan; 0 = rewrote it three times)
- Commit hygiene. Are commits small, single-purpose, and accurately messaged? (10 = clean; 0 = one giant commit)
- Verification completeness. Did you actually run all 8 manual verification steps, or did you check 3 boxes? (10 = all 8; 0 = checked from memory)
- Honest residuals. Did you list real follow-ups, or claim “all done”? (10 = honest; 0 = pretended)
- Reuse. Did you reuse the components your plan said you would? (10 = yes; 0 = invented duplicates)
35/50 = a real PR. 45/50 = a PR that doesn’t need followups.
Common failure modes
Section titled “Common failure modes”- Skipping the plan because “it’s small.” No plan, no review, no quality control. Even small features deserve a 5-minute plan.
- One huge commit. Untestable, un-reviewable, hides the failure points.
- Claiming verification you didn’t run. This is the cardinal sin. Either do the check or admit you didn’t.
- Hiding follow-ups. Every shipped feature has known gaps. Naming them is honest. Hiding them is debt.
- Inventing components your repo already has. This shows you skipped the reuse audit. Run it.