GStack Pro — 10-Role AI Engineering Team#

Built on the philosophy of Garry Tan's GStack (YC CEO) · 16K GitHub Stars · MIT License

Adapted for OpenClaw subagent + session architecture

---

What It Does#

GStack Pro gives your AI 10 specialist roles — each with a clear mandate, a structured output format, and a measurable quality bar.

Instead of one generic AI doing everything badly, you get a team:

#RoleIconSubagentBest For1CEO / Product Thinker🏛️requirerRethink the problem before building2Architect / Tech Lead🏗️architectLock in data flow, failure modes, tests3Designer Review🎨designer80-item design audit, AI slop detection4Paranoid Code Review🔍testerN+1, race conditions, trust boundaries5Browser QA🌐browser toolAI with eyes — login, click, screenshot, verify6Automated QA + Fix🧪tester + coderFind → fix → re-verify with Health Score7QA Reporter📊testerReport-only, clean handoff to team8One-Command Ship🚀operatorsync → test → push → PR9Engineering Retro🔄progressCommit analysis, praise, growth areas10Release Docs📝writerSync docs to match what shipped

---

The Development Cycle#

text

User Request
     ↓
① CEO审视 (/plan-ceo)
   → Is this worth building? What's the 10-star product?
     ↓
② Architecture Lock (/plan-eng)
   → Data flow, state machine, failure modes, test matrix
     ↓
③ Design Review (/plan-design)
   → 80-item audit, design quality grades, AI slop detection
     ↓
④ Paranoid Code Review (/review)
   → N+1, race conditions, trust boundary violations
     ↓
⑤ Automated Browser QA (/qa)
   → AI drives browser, finds bugs, fixes them, re-verifies
   → Health Score 0-100 determines ship-readiness
     ↓
⑥ One-Command Ship (/ship)
   → sync main → run tests → push → open PR
     ↓
⑦ Engineering Retro (/retro)
   → Commit analysis, team performance, improvement plan
     ↓
⑧ Release Docs (/document)
   → Update README/ARCHITECTURE to match what shipped

---

How to Activate a Role#

Method 1: Direct Command (e.g., in conversation)#

text

/plan-ceo: 为AICFO设计一个新功能：员工工资条自动生成PDF

/review: 审查deepfmt Sprint 3的代码改动

/qa: 对 https://xxx.space.minimaxi.com 运行标准QA测试

Method 2: Subagent (for background/parallel work)#

javascript

sessions_spawn({
  agentId: "tester",  // QA + Review
  task: "Read skills/gstack-pro/roles/review.md then review the code at /workspace/projects/aicfo/aicfo-mvp/src/api/"
})

---

Health Score System#

After every /qa session, output a structured score:

json

{
  "healthScore": 85,
  "status": "🟡 Good",
  "breakdown": {
    "functional": { "passed": 8, "total": 10, "score": 24 },
    "edgeCases": { "covered": 4, "total": 5, "score": 20 },
    "consoleErrors": { "passed": true, "score": 25 },
    "designRegressions": { "passed": true, "score": 16 }
  },
  "shipRecommendation": "🟡 Fix 2 minor issues before ship"
}

ScoreStatusAction90-100🟢 ExcellentReady to ship immediately70-89🟡 Good2-3 minor issues, fix before ship50-69🟠 Needs WorkSignificant bugs, fix before next sprint<50🔴 Do Not ShipCore functionality broken, redo required

---

Quality Bars#

Code Must Pass#

• ✅ N+1 queries eliminated
• ✅ All external calls have timeouts
• ✅ Retries with exponential backoff
• ✅ Database transactions properly bounded
• ✅ Input validation on all untrusted data
• ✅ No trust boundary violations
• ✅ Structured logging (JSON, with trace IDs)

Design Must Pass#

• ✅ Consistent visual hierarchy
• ✅ No AI slop patterns (copy-paste generic cards, overuse of gradients)
• ✅ Responsive at 375px / 768px / 1440px
• ✅ Accessible (color contrast, focus states)
• ✅ Meaningful empty states

---

Anti-Patterns Detected#

PatternWhy It FailsDetection"Looks good!"AI self-evaluation biasEvaluator never reads generator codeCircular dependencyUnmaintainable architectureDependency graph analysisAI slopGeneric, low-quality design80-item designer auditMagic numbersHard to maintainno-magic-numbers lint ruleForgotten edge casesSilent production failuresMandatory test matrixNo rollback planCan't safely deploy/ship requires rollback plan

---

OpenClaw Subagent Mapping#

RoleSubagent IDTypeCEO Productrequirerdemand analysisArchitectarchitecttech designDesignerdesignerUI/UX reviewCode Reviewtesterquality assuranceBrowser QAbrowser toolautomated testingQA + Fixtester + codertest + implementQA ReportertesterreportingShipoperatorreleaseRetroprogressanalysisDocswriterdocumentation

---

Key Insight: Generator vs Evaluator#

GStack Pro separates creation from judgment.

text

Generator Agent  ──→  builds code  ──→  Evaluator Agent
  (creates)         (artifact)           (judges from SPEC + URL only)
                                              ↑
                                       Never reads generator's code

This eliminates cognitive commitment bias — the AI can't judge what it already committed to building.

Inspired by: Anthropic Engineering, "Harness Design for Long-Running Application Development" (2026)

---

Files#

FilePurposeSKILL.mdThis file — overview and usagereferences/plan-ceo.mdCEO product thinking SOPreferences/plan-eng.mdArchitecture review SOPreferences/review.mdParanoid code review SOPreferences/qa.mdAutomated QA SOP + Health Scorereferences/ship.mdOne-command ship SOPreferences/retro.mdEngineering retro SOP

---

Inspired by Garry Tan's GStack ( https://gstacks.org) · MIT License For OpenClaw · Compatible with Claude Code GStack workflows