---
name: openclaw-subproject
description: "ai-agents/openclaw/ is a full GitHub clone with its own AGENTS.md and CLAUDE.md; follow those rules, not the workspace ones"
metadata:
node_type: memory
type: project
originSessionId: 71d40fa2-b151-4c81-9821-f0dfeb7a0f66
---

`ai-agents/openclaw/` is a complete clone of the OpenClaw repo
(`https://github.com/openclaw/openclaw`, Node 22 + pnpm 11, ~3.4 GB,
~45 top-level entries). It carries its own `AGENTS.md` and `CLAUDE.md` that
define strict policies on:
- Test commands (`pnpm test`, `pnpm test:changed`, no raw `vitest`)
- Type-checking (`tsgo` lanes only, never `tsc --noEmit`)
- Formatting (`oxfmt`, not Prettier)
- PR/issue workflow (specific maintainer skills, GH conventions)
- Plugin/core boundary architecture

**Why:** It's an upstream OSS project, not part of the shadow workspace's
own code. The workspace-level `CLAUDE.md` rules do not apply inside it.

**How to apply:** Before doing any work in `ai-agents/openclaw/**`, read
that subtree's `AGENTS.md` first. If the user's request is ambiguous about
which project they mean, ask — touching openclaw with the wrong commands
(e.g., `tsc`, Prettier, raw vitest) will get noticed.