This chapter gives you 3 actionable checklists: starting a project, shipping a feature, ending a session. Print them, keep them within sight. Reading time: 5 minutes.
Checklist 1, before starting a project
Everything that needs to be in place before you write the first line of code. Each box should be tickable with a concrete proof (a command that passes, a file that exists, a manual test that works).
Installation
- VS Code open, Claude Code extension visible in the sidebar
-
claude --versionprints a recent version in the terminal - Active Anthropic subscription (Pro or Max), confirmed on console.anthropic.com
- First prompt sent, Claude replies with no auth error
Project setup
-
CLAUDE.mdat the repo root with the 8 sections (chapter 3) -
.claude/commands/exists and holds at least one agent -
.claude/settings.local.jsonpresent, pre-authorized permissions listed -
docs/plans/created for design docs -
MEMORY.mdinitialized with at least one entry (chapter 10) -
.gitignoreexcludes.env.localand.claude/settings.local.json
Advanced configuration (optional)
- Prettier post-edit hook active, an edited file comes out formatted
- Typecheck pre-commit hook active, a commit with a TS error is blocked
- Global deny list in
~/.claude/settings.json(rm -rf, force push, db reset) tested - MCP servers connected,
claude mcp listshows the right names - Exploration sub-agents present in
~/.claude/agents/ - Status line configured, context alert fires at 60%
Infrastructure (if SaaS)
-
node --versionshows v22 or higher - Supabase project created,
npx supabase startboots with no error - Vercel linked to the repo, a
git pushtriggers a deploy - Every variable from
env.examplecopied into.env.localand filled - Domain purchased and DNS configured (if going to production)
Checklist 2, for every feature
The full cycle, from brainstorm to commit. Nothing is ticked until you have proof that it works.
Planning phase
- 3 to 4 options explored in writing before picking one
- Design doc written: vision, user stories, business rules
- Implementation plan written: numbered tasks, files touched, per-task verification criteria
- Plan reviewed and validated before any code
Execution phase
- Claude runs the plan task by task, you supervise
-
npm run typecheckpasses (zero errors) after each task -
npm run lintpasses (zero warnings) after each task - Local build OK:
npm run buildfinishes with no error
Validation phase
- Visual test on localhost:3000, happy path confirmed
- Visual test of error cases: invalid input, empty data, 404
- Responsive test: window shrunk to 375px, nothing overflows
- Copy proofread, zero typo, accents correct (if French)
- 2 to 3 visual or functional iterations done
Shipping phase
-
/reviewrun, every flag handled (fixed or justified) - Unit tests written and green if business logic
- Lighthouse > 90 on the touched page (if public)
-
/commitwith a conventional message (feat:, fix:, refactor:) -
git pushdone, Vercel preview deployed and tested
Checklist 3, at the end of a session
What needs to be done before you close VS Code.
Save
-
git statusclean, no uncommitted changes -
git pushdone, remote is up to date - "Update the project memory" run,
MEMORY.mdreflects today's decisions
Prepare for the next session
- Next priorities noted in
MEMORY.mdorTODO.md - Open bugs documented with repro steps
- Working branch identified (
mainor feature branch) - Local servers stopped if you're powering down (Docker,
npm run dev)
"Quick" version for days when you're in a rush
If you only have 30 minutes, here's the bare minimum.
Express (30 min)
- Open the project in VS Code + Claude Code
- Describe the feature in a sharp prompt (chapter 4)
- Let Claude code, supervise
- Quick visual test on localhost
-
/commit+git push - "Update the memory" before closing
Takeaway: these checklists are your safety net. Every box is a verifiable action, not a vague principle. If you can't tick it, you're not done.