Skip to content
Cratis

AI skills and plugins

The Cratis Stack

Teach your assistant the Cratis way

A general-purpose assistant knows C# and TypeScript, but it does not automatically know why a Cratis command lives beside its fact, why a projection consumes events instead of another read model, or why generated proxies should never be hand-edited. Cratis skills provide that missing framework context as reviewed workflows instead of another prompt you have to remember.

One behavior, several native packages

Section titled “One behavior, several native packages”

Cratis authors canonical skills in Cratis/AI. Four committed marketplace manifests — for Claude Code, Codex, GitHub Copilot, and Cursor — point hosts straight at the skills/ and engineering/ directories on the repository’s default branch, so what a host installs is exactly what a reviewer reads there.

Cratis/AI

canonical skills and profiles
Claude Code / Codex
Copilot / Cursor
npm and Pi packages
Your repository
.cratis/PROJECT.md

project-owned context

This avoids behavior forks. A projection skill should not teach one architecture in Claude and another in Copilot merely because their package manifests differ.

Public packages stay passive

Section titled “Public packages stay passive”

The planned public package contains portable Agent Skills and their static references, assets, and licenses. It excludes executable extensions, lifecycle scripts, credentials, project facts, authoring evals, engineering hooks, agents, prompts, and repository tooling.

Executable Chronicle operations belong to the documented CLI and Chronicle MCP surfaces, with their own authorization and confirmation boundaries. They are not smuggled into a passive coding-skills package.

Skills guide complete workflows

Section titled “Skills guide complete workflows”

Skills are the high-value pieces: your assistant invokes one when a request matches and follows a Cratis workflow instead of improvising.

  • event-modeling — settle commands, facts, streams, state views, reactions, compliance, and scenarios before implementation.
  • new-vertical-slice — build an application behavior end to end through backend, specs, frontend, documentation, and gates.
  • add-concept, add-projection, add-reactor, add-business-rule, add-ef-migration — implement one focused Cratis artifact by convention.
  • write-specs, write-documentation — preserve behavior or explain the product through the repository’s preferred structure.
  • review-code, review-security, review-performance — review against Cratis architecture and quality boundaries.
  • diagnose-slice and inspect-running-chronicle — separate source-code diagnosis from authorized running-store inspection.

Framework and client repositories select a different profile from Cratis applications. A skill reads the repository profile before applying application vertical-slice conventions.

Internal engineering behavior is separate

Section titled “Internal engineering behavior is separate”

Cratis maintainers currently have repository-local rules, agents, prompts, and hooks in Cratis-owned repositories. Those engineering artifacts can coordinate specialists, enforce repository gates, and carry contribution conventions, but they are not part of the planned public passive package.

See Using AI as a Cratis maintainer for that workflow.

Do not copy configuration between repositories

Section titled “Do not copy configuration between repositories”

The old all-to-all propagation model is frozen. Copying .ai, .claude, .github, .agents, or .pi trees creates mixed versions and lets a consuming repository become an accidental distributor.

The replacement is ordinary versioned distribution: hosts install the plugin from the Cratis/AI marketplace, or repositories subscribe to profiles at exact versions. A consuming repository owns only its project context and minimal host bootstraps. Updating or rolling back changes the shared version without overwriting project facts.

Profiles explains how to select a narrower scope than the whole bundle, and Trust and distribution explains the approval, release, and rollback gates in detail.

Get the skills and the operating tools today

Section titled “Get the skills and the operating tools today”

Everything is installable now, in two halves:

  1. Coding skills — add Cratis/AI as a marketplace and install the cratis plugin (commands per host), or install the published @cratis/pi package through Pi.
  2. Operating tools — run cratis init in your project to generate the supported CLI context, then connect Chronicle MCP when your host and authorization policy allow it.

Read Start using AI with Cratis for the guided setup and AI ecosystem support for host-by-host status.

Where to go next

Section titled “Where to go next”
Start using AI

Install the skills through your host’s marketplace and set up the operating tools.

 Profiles

Subscribe a repository to exactly the products, languages, and architectures it needs.

 Ecosystem support

Compare host-by-host installation status and the gates behind support claims.

 Cratis maintainers

Use repository-local skills and gates without restarting propagation.

 Code analysis

See the compiler and lint gates that verify contracts independently of the assistant.