Claude — 30-Day Roadmap

Setup + scaffolding. Day 1: install Claude Code, write a project CLAUDE.md (200–300 lines, hard rules + tech stack + test commands), commit it. Day 2: scaffold an empty module via /claude module generator, run bin/magento module:enable + setup:upgrade. Day 3: add one Service Contract end-to-end (interface, repository, di.xml, REST route). Day 4: add the matching GraphQL mutation + resolver. Day 5: write the first MFTF test that proves both transports work. Weekend: review what you built, refactor anything you don't understand — if you can't explain a line, delete and rewrite it.

Right. By now your CLAUDE.md is stable, so it's time to build leverage. Day 8: add .claude/commands/scaffold-feature.md, add-graphql.md, write-mftf.md — pin them to your repo's actual conventions. Day 9: add .claude/commands/run-tests.md that wires php -l, phpcs, phpstan, mftf. Day 10: install the MySQL MCP server, point at your dev DB, ask Claude to introspect your tables before each new feature. Day 11: install the Magento Admin REST MCP, ask Claude to verify a product was created via API. Day 12: add a PostToolUse hook that runs setup:di:compile after di.xml edits. Now you're an order of magnitude faster than week 1.

Three sub-agents. code-reviewer: primed as an Adobe EQP reviewer, runs phpcs --standard=Magento2, phpstan --level=6, checks for ObjectManager + vendor edits + missing _isAllowed(). test-writer: knows your MFTF action-group conventions, writes one MFTF test per feature, runs it before reporting back. release-notes: parses conventional commits, groups by type, surfaces breaking changes. Each agent is a Markdown file in .claude/agents/<name>.md — system prompt + tools allowed + when to invoke. Test by deliberately writing bad code (ObjectManager::getInstance() in a controller) and watching the reviewer catch it.

Yes. Day 22: feature-freeze, run the full test suite, run the security-review skill. Day 23: write the CHANGELOG.md via /release-notes, bump SemVer in composer.json. Day 24: tag, push to GitHub, build the Composer package or Marketplace ZIP. Day 25: deploy to staging, run smoke tests by hand. Day 26: production deploy via your existing DEPLOYMENT.md flow, monitor for 24h. Day 27–28: write a retrospective — what worked, what didn't, what goes into the next CLAUDE.md. Day 29–30: open-source the slash-commands and sub-agents you built — you'll thank yourself in 6 months.

Don't restart. The plan is built around artifacts, not days — if you have the artifact, you have the skill. Audit: do you have a working module with a Service Contract, REST + GraphQL transports, and at least one MFTF test? If yes, jump to week 3 (sub-agents). If no, the gap is week 1 — go back. Common slip: people skip the MFTF test on day 5 because "it works manually". A week later they break the GraphQL resolver and don't know — the test would have caught it. Do not skip the test; reduce the scope of the test instead (one happy path is fine on day 5).