martian56 Using Claude-Engineer
Announce at start: "I'm using claude-engineer to drive this project's lifecycle."
Claude-Engineer is a lifecycle orchestrator for building full greenfield web apps end-to-end. It is opinionated: React or Next.js + Tailwind + a DESIGN.md-driven custom design system on the front, FastAPI + SQLAlchemy 2.0 + Alembic + Pydantic v2 on the back, a polyglot monorepo, Docker for stateful services only, and hard quality gates.
It does NOT reimplement general engineering discipline - it calls into superpowers for that and
adds the whole-project glue superpowers lacks (intake → phased plan → autonomous build → finish, plus
durable cross-session memory).
Iron Laws (violating the letter is violating the spirit)
- Plan-gated. No build work before the user signs off on the phased plan. Heavy alignment up front, then autonomous.
- Greenfield only. Claude-Engineer scaffolds NEW projects on its opinionated stack. It does not onboard existing codebases.
- Customer experience first. Build in this order: DESIGN.md → frontend shells/components → backend APIs → integration.
- Hard quality gates for "done." ALL of: tests + coverage; Playwright E2E with ZERO console errors; WCAG AA; strict types + lint clean. Enforced at the commit gate and at finish.
- Services in Docker, app runs native. Postgres/Redis/MinIO/Mailpit in Docker; frontend & backend run natively in separate shells.
- Trust the ledger.
.claude-engineer/progress.md+memory.md+git logover your own recollection (survives compaction).
The lifecycle and which skill to invoke
| Stage | claude-engineer skill | calls into superpowers |
|---|---|---|
| 0. Conventions/setup | claude-engineer:project-conventions |
- |
| 1. Intake (brief) | claude-engineer:intake-project |
brainstorming |
| 2. Plan (sign-off gate) | claude-engineer:plan-project |
writing-plans |
| 3. Memory init | claude-engineer:project-memory |
- |
| 4. Build (autonomous) | claude-engineer:execute-build |
using-git-worktrees, subagent-driven-development (each task: test-driven-development), dispatching-parallel-agents, executing-plans (fallback), systematic-debugging |
| - design system | claude-engineer:design-system |
- |
| - frontend | claude-engineer:frontend-development |
- |
| - backend | claude-engineer:backend-development |
- |
| - monorepo/tooling | claude-engineer:monorepo-and-tooling |
- |
| - local services | claude-engineer:docker-dev-environment |
- |
| 5. Verify/review | claude-engineer:testing-and-quality |
requesting-code-review, receiving-code-review, verification-before-completion |
| 6. CI + finish | claude-engineer:ci-cd-pipeline, then finishing-a-development-branch |
finishing-a-development-branch |
Guided path: the /claude-engineer:engineer command walks stages 0→6 in order.
Per-stage path: invoke any skill above directly (e.g. /claude-engineer:plan-project).
Rules of engagement
- If a stage's skill exists, USE it; don't freelance the stage.
- Reference a skill's
references/docs on demand (read by path) - do not paste them wholesale. - Update
.claude-engineer/progress.mdafter each task and.claude-engineer/memory.mdon key decisions. - When unsure whether a claude-engineer skill applies: it does. Invoke it.