@eventmodelers/build-kit-cratis-csharp

Real-time Claude agent + skill kit that turns Eventmodelers board slices into Cratis vertical slices in a .NET / C# project. Connect your board to an autonomous coding agent that picks up slice status changes, implements the slice the Cratis way (Cratis Arc + Chronicle), runs dotnet build / dotnet test, and marks the work done — all without manual intervention.

Built on cratis.io. The Cratis conventions are distilled into the kit's skills so the agent stays self-contained — it does not need the Cratis repos to know how to build a slice.

What you get

• A ready-to-run Cratis starter app (Cratis Arc + Chronicle + MongoDB + React/PrimeReact) dropped into your project root, including an example vertical slice to learn from and a CLAUDE.md of conventions.
• The ralph loop — an autonomous agent that reacts to slice:changed events and implements slices.
• Build skills that encode the Cratis way:
  • /build-state-change — write slices → [Command] + Handle() + [EventType]
  • /build-state-view — read slices → [ReadModel] + projection/reducer + queries
  • /build-automation — automation / translation → IReactor + ICommandPipeline
• Platform skills (/connect, /load-slice, /update-slice-status, /learn-eventmodelers-api).

How it works

Board (Eventmodelers)
  │  slice status → "Planned"
  ▼
Realtime Agent  ──────────────────► tasks.json
  │  listens on a board channel          │
  ▼                                      ▼
ralph loop ◄────────────────────── Phase 1: load slice  (/connect + /load-slice → .slices/)
  │
  ▼
Phase 2: build slice
  → set status "InProgress" on the board
  → route by slice type to /build-state-change | /build-state-view | /build-automation
  → implement ONE .cs slice file the Cratis way
  → dotnet build (also regenerates TypeScript proxies)  +  dotnet test  (slice only)
  → implement the React component(s) if UI-triggered, register in the composition page
  → commit, merge to main
  → set status "Done" on the board

Prerequisites

• .NET SDK 9.0+ and Docker (for Chronicle + MongoDB via docker-compose).
• Node.js 18+ (the realtime agent and the starter frontend are Node/Vite).
• An Eventmodelers board (org ID, board ID, API token) — optional; the kit also works locally without a board (it just skips board sync).

Step 1 — Install

Run the installer in your project directory:

npx @eventmodelers/build-kit-cratis-csharp install

It will:

• Copy the Cratis starter app into your project root (CratisApp.csproj, Program.cs, docker-compose.yml, the .frontend/ shell, an example SomeModule/SomeFeature/ slice, CLAUDE.md).
• Install the loop machinery and skills into .build-kit-cratis-csharp/ (gitignored automatically).
• Prompt for board credentials (org ID, board ID, token) → .build-kit-cratis-csharp/.eventmodelers/config.json.
• Configure the Eventmodelers MCP server in .build-kit-cratis-csharp/.claude/settings.json.

| Path | Purpose | |------|---------| | .build-kit-cratis-csharp/.claude/skills/build-state-change | Write-slice skill (Cratis commands/events) | | .build-kit-cratis-csharp/.claude/skills/build-state-view | Read-slice skill (read models/projections) | | .build-kit-cratis-csharp/.claude/skills/build-automation | Automation/translation skill (reactors) | | .build-kit-cratis-csharp/.claude/skills/_shared/cratis-conventions.md | The distilled Cratis conventions | | .build-kit-cratis-csharp/.claude/skills/{connect,load-slice,update-slice-status,learn-eventmodelers-api} | Platform skills | | .build-kit-cratis-csharp/ralph-claude.js / ralph.sh | The agent loop | | .build-kit-cratis-csharp/lib/prompt.md / backend-prompt.md | Agent instructions | | .build-kit-cratis-csharp/lib/AGENT.md | Accumulated learnings across iterations |

Step 2 — Bring up the app

docker-compose up -d        # Chronicle + MongoDB + Aspire dashboard
dotnet build                # backend + TypeScript proxy generation
npm install                 # frontend deps

Step 3 — Start the loop

# Claude (default)
node .build-kit-cratis-csharp/ralph-claude.js

# Local Ollama model (run `ollama serve` first)
OLLAMA_MODEL=qwen3:8b node .build-kit-cratis-csharp/ralph-ollama.js

# Target a custom project directory
node .build-kit-cratis-csharp/ralph-claude.js /path/to/project

Optionally start the realtime agent for automatic board notifications:

cd .build-kit-cratis-csharp && npm install && node realtime-agent.js

Slice-type routing

The loop reads slice.json and routes by type:

| slice.json signal | Cratis slice type | Skill | Produces | |---|---|---|---| | has commands[] / events[] | State Change | /build-state-change | [Command] + Handle() + [EventType] | | has readModel / projections / queries | State View | /build-state-view | [ReadModel] + projection/reducer + static queries | | non-empty processors[] | Automation | /build-automation | IReactor | | sliceType === "TRANSLATION" | Translation | /build-automation | IReactor → ICommandPipeline.Execute(command) |

The Cratis way (what the skills enforce)

• ALL backend artifacts for a slice in ONE .cs file under <Module>/<Feature>/<Slice>/.
• [Command] records with Handle() on the record — never separate handler classes.
• [EventType] with NO arguments; past-tense, never-nullable events.
• ConceptAs<T> for every identity/value — no raw Guid / string in the domain.
• [ReadModel] records with public static query methods; observable queries return ISubject<T>.
• Reactors implement IReactor; new writes go through ICommandPipeline, never IEventLog.
• dotnet build generates the TypeScript proxies — Backend → build → Specs → Frontend → Composition.

Full detail: .build-kit-cratis-csharp/.claude/skills/_shared/cratis-conventions.md.

CLI commands

npx @eventmodelers/build-kit-cratis-csharp install    # install and configure
npx @eventmodelers/build-kit-cratis-csharp status     # check what is installed
npx @eventmodelers/build-kit-cratis-csharp uninstall  # remove installed files

Slice statuses

| Status | Meaning | |--------|---------| | Created | Slice exists on the board, not yet planned | | Planned | Queued for the agent — triggers a build | | InProgress | Agent is currently implementing | | Review | Implementation complete, awaiting review | | Done | Fully implemented and merged | | Blocked | Waiting on an external dependency |

License

MIT