Browse / General-purpose / agent-workspace

agent-workspace

Vainstel

Template for a Claude Code entry-point workspace: layered CLAUDE.md context (workspace/org/product/repo), authoring skills, and a manager agent

Open Vainstel/agent-workspace → Checked against the four rules.

Built for Claude Code

ships .claude CLAUDE.md and nothing for other agents.

In its own words

README.mdVainstel/agent-workspace

A template for a Claude Code entry-point workspace: one directory that holds all your organizations, products and repositories, with layered CLAUDE.md context files, authoring skills, and a manager agent that helps you set everything up and keep it all up to date.

The story and reasoning behind it — in the article: One Entry Point for Your Coding Agent.

Structure

The point: an agent started here quickly finds what it needs to work on, orients itself across the tree, and understands how the repos of a product relate to each other and how products relate between themselves — no repeated intro prompts to bring it up to speed each session.

Context is layered — the agent reads the CLAUDE.md chain top-down along its task path and finds the right depth at the right place, with nothing duplicated:

  • Workspace level — single global CLAUDE.md: base agent rules, independent of any company.
  • Organization level — one directory + CLAUDE.md per org; private/ (personal projects) is an org too.
  • Product level — CLAUDE.md per product: what it is, repo map, how services talk. A single-repo product may sit directly under an org (no product layer).
  • Repo level — each repository's own CLAUDE.md, self-contained, living inside that repo's git (not tracked here). Optionally a personal, gitignored CLAUDE.local.md next to it for env-specific details.

Quoting 46% of the README. Read the rest on GitHub ↗

Read from the repository's git tree on 2026-08-17.