What You Get at the End
2 min read
What You Get at the End
The complete output
When your agent finishes the 13-step process, your project folder contains:
your-project/├── docs/│ ├── idea/idea-pack.md # Structured idea│ ├── prd/prd.md # Full spec with acceptance criteria│ ├── prd/acceptance-checklist.md # Testable checkbox list│ ├── adr/ADR-001-*.md # Every technical decision│ ├── adr/README.md # Decision index│ ├── implementation-plan.md # Build blueprint│ ├── design-system.md # Visual language│ ├── pages/*.md # Page specifications│ ├── screenshots/*.png # Every page captured│ └── build-log.md # What happened during build├── src/ # Application code├── tests/ # Test files├── state.json # Build state tracker├── README.md # Setup + screenshots├── .env.example # Required environment variables└── package.json # DependenciesWhat makes this different
Decision trail. Every "why?" has an answer in an ADR. Three months from now, you know exactly why PostgreSQL was chosen over MongoDB, why NextAuth instead of Clerk, why Tailwind instead of CSS modules.
Testable spec. The acceptance checklist isn't a wish list — it's a pass/fail scorecard with numbered criteria that map directly to user stories. You can hand this to QA, a client, or a new developer and they'll know exactly what the app should do.
Resumable. The state.json file tracks progress. If the agent crashes, loses context, or gets interrupted, it reads state.json and picks up where it left off. No "what were we building again?"
Documented from birth. Most projects start documented and end undocumented. This process generates documentation as a byproduct of the build — not as an afterthought.
Real numbers from the dogfood build
We built an uptime monitor (URL health checker) using this exact process:
Metric | Result |
|---|---|
Planning time (Steps 0-4) | ~45 minutes |
Build time (Steps 5-6) | ~90 minutes |
Total time | ~3 hours |
Lines of app code | ~850 |
Git commits | 5 |
Acceptance criteria | 20 defined, 18 passing |
API endpoints | 10 |
Pages | 5 (login, signup, dashboard, detail, settings) |
Features | Auth, CRUD, scheduler, charts, notifications |
The planning was 25% of the total time. That front-loaded thinking meant the build phase had almost zero backtracking.
Get the skill file
Ready to try it? Download the agent skill file and tell your OpenClaw agent to follow it:
[Download SKILL.md →]
Or tell your agent directly:
> Read the build workflow skill at [URL] and follow it to build me [your idea].
Built by Jiffi — Professional Problem Solvers. [jiffi.co](https://jiffi.co)
Sign up to read the full guide
Free access to all 12 workflow guides. No password needed.