Platform

Boundaries

Ownership and lane boundaries for the agents repo and the wider agent ecosystem

Why boundaries matter

If this repo starts owning mutable operator state or too much lower-layer shared logic, it stops being a clean control plane and becomes a confusing duplicate of the surrounding system.

This repo is the canonical home for the shared agents ecosystem, but it is not the entire ecosystem.

What this repo owns

CLI

packages/cli is the active agents CLI implementation.

Backend

packages/backend hosts the Convex deployment and wrapper surface for the shared component.

Dashboard

apps/app is the control-plane UI for operators.

Docs

apps/docs is the public explanation surface for the product and repo model.

Design support

packages/design provides repo-local shell support for the web surfaces.

What stays outside this repo

Local agent OS

~/.agents owns rules, skills, hooks, sessions, plans, events, and runtime state.

Shared Convex component

@lnittman/convex-agents remains the lower-layer shared backend component.

Other products

Product shells like Abbie stay separate rather than folding into this repo.

Fleet maintenance

Fleet-wide dependency and workspace maintenance happen elsewhere in Luke’s toolchain.

Practical rule of thumb

Product visibility
apps/app
Product docs
apps/docs
Local execution
packages/cli
Host wrappers
packages/backend/convex/agentSystem/**
Shared nouns
@lnittman/convex-agents
Mutable operator state
~/.agents

Source of truth

  • Repo guidance: AGENTS.md
  • Repo overview: README.md
  • Boundary note: docs/ecosystem-boundaries.md