gstack for OpenClaw

AI engineering team simulator — ported from garrytan/gstack to OpenClaw-native architecture.

One developer, 28 specialized skills, entire development lifecycle.

What It Does

gstack gives your OpenClaw agent the knowledge of a full engineering team:

Stage	Skills	What They Do
Think	/office-hours	YC-style brainstorming, premise challenges
Plan	/plan-ceo-review, /plan-eng-review, /plan-design-review, /autoplan	Multi-persona plan review (CEO, engineer, designer)
Review	/review, /second-opinion	Two-pass code review + cross-model comparison
Design	/design-consultation, /design-review, /design-shotgun	Design system creation, visual audit, rapid variants
Test	/qa, /qa-only, /benchmark, /browse, /investigate	Browser-based QA, performance, root-cause debugging
Guard	/careful, /guard, /cso	Destructive command warnings, edit boundaries, security audit
Ship	/ship, /land-and-deploy, /canary, /document-release	PR creation, deploy, monitoring, doc updates
Reflect	/retro	Data-driven retrospective from git history

Installation

1. Copy workspace files

cp -r workspace/ ~/.openclaw/workspace/

2. Merge config

Merge openclaw.json into your existing ~/.openclaw/openclaw.json:

• Add the gstack agent definition
• Add the openai provider (for /second-opinion)
• Add the skills directory

3. (Optional) Install browser automation

For full QA and testing capabilities:

git clone https://github.com/garrytan/gstack.git /tmp/gstack
cd /tmp/gstack && bun install && bun run build
cp browse/dist/browse /usr/local/bin/browse

Or run /setup-browser to walk through it interactively.

4. Configure your project

Run /setup-deploy to configure deployment, or edit USER.md manually.

File Structure

~/.openclaw/
├── openclaw.json              # Agent config (merge with yours)
├── openclaw.plugin.json       # Plugin manifest
└── workspace/
    ├── AGENTS.md              # Operating instructions
    ├── SOUL.md                # Principles & multi-persona voice
    ├── IDENTITY.md            # Agent identity
    ├── USER.md                # Your preferences (edit this)
    ├── TOOLS.md               # External tool documentation
    ├── MEMORY.md              # Long-term memory (auto-maintained)
    ├── memory/                # Daily logs
    └── skills/                # 28 skills (25 invocable + 3 reference)

Porting Notes

This is a port from Claude Code to OpenClaw. Key differences from the original:

• No template resolvers — Skills are static markdown (original uses TypeScript build-time generation)
• Guard/freeze is advisory — OpenClaw has no tool-call interception hooks; freeze lists are memory-based conventions
• Browser is external — The browse binary must be installed separately (not bundled)
• Review logging uses memory — Original uses JSONL files; this uses OpenClaw's daily memory logs
• Second-opinion replaces codex — Provider-agnostic; uses whatever secondary provider is configured

Credits

Original gstack by Garry Tan. MIT License.