Before diving into workflows, learn the four building blocks that power spec2cloud.
The orchestrator runs in a repeating 11-step cycle called the Ralph Loop. Each iteration: read current state → determine the next task → find or create the right skill → execute → verify → update state. This loop is the heartbeat of spec2cloud — it drives every phase from specification to deployment.
The loop maintains deterministic progress through state.json, committed to git after every action. If you stop, crash, or close your laptop — the loop picks up exactly where it left off.
Skills are reusable procedures that do the actual work. Each skill has a SKILL.md file with instructions, optional references, and scripts. The orchestrator discovers and invokes them automatically.
46 skills in 7 categories:
| Category | What they do | Count |
|---|---|---|
| Phase | Drive discovery (spec refinement, UI/UX, tech stack) | 3 |
| Delivery | Execute increments (tests, contracts, implementation, deploy) | 6 |
| Protocol | Ensure consistency (state, commits, audit, gates, resume, errors) | 6 |
| Utility | Support tools (validator, runner, build check, diagnostics, Aspire, Playwright) | 11 |
| Extraction | Scan existing codebases (scanner, deps, arch, API, data, tests) | 6 |
| Assessment | Evaluate paths (modernize, rewrite, cloud-native, security, perf) | 5 |
| Planning | Generate increments (modernize, rewrite, cloud-native, extend, security) | 5+4 |
Skills follow the agentskills.io standard. You can create your own or install community skills from skills.sh.
Two systems make spec2cloud reliable and safe:
State persistence — All progress lives in .spec2cloud/state.json and audit.log, committed to git after every action. This means:
- Resume from any interruption point
- Share progress via git push/pull
- Full audit trail through git history
- No external database required
Human gates — The orchestrator pauses at critical checkpoints and waits for your approval. Nothing ships without your sign-off: spec review, design review, tech stack, Gherkin scenarios, code review, deployment verification, and (for brownfield) the testability gate.
→ Deep dive: State & Human Gates
Work is delivered in increments — small, self-contained units that each leave the application fully working and deployed.
Every increment follows the same 4-step cycle:
- Tests — Generate e2e specs, Gherkin scenarios, and unit test stubs. Establish a red baseline (tests fail, as expected).
- Contracts — Generate OpenAPI specs, shared TypeScript types, and infrastructure contracts. These enable parallel implementation.
- Implementation — Write code in parallel slices (API + Web), then wire together (Integration). Red tests turn green.
- Deploy — Provision Azure resources, deploy, run smoke tests, verify with full regression.
After each increment: main is green, deployment is live, docs are current. Then the next increment begins.
→ Greenfield Guide | Brownfield Guide
Choose your path:
- Building something new? → Greenfield Guide
- Working with existing code? → Brownfield Guide
- Want examples? → Walkthroughs