refactor: hexagonal architecture, DDD domain layer, CD service rename, CI tests

[ronenhamias]

Summary

Complete architectural overhaul of Robokit from a flat, monolithic layout into a hexagonal (ports & adapters) architecture with a clean domain-driven design layer. Alongside the structural refactor, the "environment service" concept is consistently renamed to Continuous Deployment (CD) service throughout the codebase.

---

What changed

Architecture — Hexagonal / DDD

The entire app/ directory is replaced by src/ with strict layering:

src/
  domain/          ← pure business logic, zero I/O
  application/     ← use cases, orchestration
  adapters/        ← GitHub, CD service, Vault
  templates/       ← markdown renderer
  config.js        ← single validated config object
  logger.js        ← structured pino logger
  server.js        ← entry point
  webhook.js       ← Probot wiring

Deleted (dead / replaced): app/cache.js, app/config.js, app/environments.js, app/github-service.js, app/http-gateway.js, app/utils.js, app/vault-api.js, app/statuses/templates.js, index.js, robokit.js

---

Domain layer (src/domain/)

File	What it does
PipelineStatus.js	Single source of truth for all CD-status → GitHub status/conclusion/marker mappings. Fixes the silent CANCELLED vs CANCELED typo bug.
DeploymentContext.js	Immutable value object (Object.freeze). All deploy state in one place; withCheckRunName() / withNamespace() return new instances.
DeploymentPolicy.js	Pure functions: evaluate(checkRun) decides trigger type; evaluateUserAction(action) maps GitHub button clicks. Zero I/O.

---

Application layer (src/application/)

File	What it does
DeploymentOrchestrator.js	Replaces coordinator.js. Drives the full deploy pipeline: build immutable context → acquire lock → create check run → stream CD service → update GitHub on every status change → finalize. Uses local state object instead of mutable shared object.
deploymentLock.js	Per-owner/repo:branch concurrency guard. A second deploy for the same key while one is in-flight is skipped with a warning rather than running in parallel.

---

Adapters (src/adapters/)

Adapter	Notes
cd-service/CdServiceClient.js	Renamed from environments.js. Fixed WebSocket double-connect race with #connecting private field (reuses in-flight promise). Uses rxjs directly (Subject) — removes rxjs-compat.
cd-service/CdServiceMock.js	Extracted mock for LOCAL_DEV=true. Emits RUNNING×2 then SUCCEEDED; no real WS needed.
github/CheckRunAdapter.js	Replaces check-run.js. Uses PipelineStatus — fixes the CANCELLED conclusion bug. Removed obsolete ant-man-preview / flash-preview headers.
github/DeploymentAdapter.js	Replaces deployment.js. Same cleanup.
github/OctokitRegistry.js	Replaces cache.js. Typed register/get with warn-on-miss.
vault/VaultAdapter.js	Renamed from vault-api.js.

---

Infrastructure

• src/config.js — validates all required env vars at boot (APP_ID, PRIVATE_KEY, WEBHOOK_SECRET), exits with a clear message on missing values. Groups all config in one object.
• src/logger.js — pino structured logger; child loggers carry deployId, repo, branch, sha correlation fields across the entire pipeline.
• src/server.js — selects real vs mock CD service based on LOCAL_DEV. Adds GET /health endpoint ({ status, uptime, cdService }).
• src/templates/renderer.js — safe string interpolation (only substitutes string | number values). Fixed "Continues Delivery" → "Continuous Delivery" typo in templates.
• package.json — added pino, removed rxjs-compat. Updated scripts: start → Probot, robokit → direct node for IntelliJ.
• Dockerfile — updated COPY paths for new src/ layout.

---

Tests

• test/unit/domain/PipelineStatus.test.js — 20 tests covering all status mappings, markers, isTerminal, unknown fallbacks, and the CANCELED/CANCELLED typo guard.
• test/unit/domain/DeploymentContext.test.js — 6 tests: field assignment, immutability, withCheckRunName, withNamespace, chaining.
• test/unit/domain/DeploymentPolicy.test.js — 14 tests: all evaluate() branches and all evaluateUserAction() cases.
• test/unit/application/deploymentLock.test.js — 6 tests: basic run, skip duplicate, release after completion, release after error, key isolation, isLocked state.
• test/e2e.test.js — added GET /health test; cleaned up stale commented-out tests.

---

CI

.github/workflows/publish-docker.yml — added a test job that runs npm test before the Docker build. The build job now has needs: test, so a failing test blocks the image publish.

---

.env cleanup

Removed unused variables (GITHUB_INSTALLATION_ID, GITHUB_OWNER, GITHUB_REPO, GITHUB_SHA). Regrouped remaining variables into four labelled sections: Server, GitHub App, Continuous Deployment Service, Vault.

---

Test plan

• npm test passes (all unit tests green)
• CI test job runs before build on push to develop
• LOCAL_DEV=true npm run robokit starts without errors; /health returns { status: 'ok' }
• Sending a valid check_run webhook (smee.io) triggers the mock CD pipeline and completes the GitHub check run
• A second simultaneous deploy for the same branch logs deploy.skipped.already-in-progress and returns without error