A clean Eve starter monorepo with Atlas AI documentation, memory, skills, and review gates built in.

Atlas Eve Starter is a small, reviewable template for building production-style Eve agents. It gives you the starter wiring, repository structure, and AI-facing documentation defaults up front, so your team can replace the example agent with real product behavior instead of first inventing the project shape.

The repository is initialized with Atlas, Blazity's framework-agnostic AI documentation scaffold. Atlas adds the .ai/ workspace, shared agent instructions, durable memory files, managed skills, artifact paths, and doctor/review gates that help coding agents work from explicit project context.

• 🗺️ Atlas from the first commit: best-practice AI documentation, memory, skills, and structural gates are already part of the repo.
• 🤖 Eve-native setup: the example app includes a replaceable Eve agent, deterministic tool, HTTP channel, and opt-in eval.
• 🧪 Initial eval setup: model-backed evals are ready for pre-PR or pre-publication confidence checks without becoming a hidden local workflow cost.
• 🧭 Reviewable agent boundaries: app behavior stays under apps/*/agent, while shared contracts live in explicit packages/* imports.
• 🧰 Production-style defaults: pnpm, Turborepo, TypeScript, Biome, Vitest, Husky, and lint-staged are wired from the start.
• 📦 Multi-agent context setup: Atlas keeps shared instructions, memory, skills, and artifacts in one place for every coding agent in the repo.

• pnpm 11 workspace
• Turborepo
• Node 24.x
• TypeScript 6
• Biome
• Vitest
• Eve via eve
• Atlas AI documentation scaffold

Use Node 24 (.node-version and .nvmrc are included).

pnpm install

Routine checks do not require model credentials:

pnpm check
pnpm typecheck
pnpm test

Pre-PR or pre-publication Eve check:

pnpm eve:build

pnpm eval is opt-in. It runs Eve eval sessions, can require provider credentials, and may spend tokens. Keep it as a manual confidence check instead of a Husky hook, pre-commit check, or automatic local workflow.

A replaceable echo agent that demonstrates the starter wiring:

• agent/agent.ts for the root Eve agent definition
• agent/instructions.md for concise agent instructions
• agent/tools/echo.ts for deterministic tool behavior
• agent/channels/http.ts for a local HTTP ingress example
• evals/example.eval.ts for an opt-in Eve eval

The HTTP channel uses auth: null only because this is a local domain-neutral example. Add an explicit auth/session model before adapting the route for production.

Run it locally:

pnpm --filter @repo/example-agent dev

Atlas-managed AI context for the repository:

• config.json as the source of truth for artifact paths
• LANGUAGE.md for project vocabulary
• memory/ for stable project context
• plans/, research/, decisions/, and results/ for agent artifacts
• skills/ for setup, review, and Eve-agent guidance

A tiny shared contract package used by the example app. Replace it with your own shared schemas or remove it if your starter does not need a local package.

• A working Eve example app with local HTTP ingress.
• A shared package pattern for contracts and schemas.
• Atlas-managed AI instructions, memory, skills, and artifact paths.
• Deterministic checks for formatting, linting, types, and tests.
• Manual Eve build and eval commands for higher-confidence review.