What is this?
An 8-agent autonomous product team that takes product ideas from concept to merged pull request through a 10-stage evidence-gated pipeline. Each agent owns a specific role (PM, PO, Tech Lead, Designer, Backend, Frontend, QA, DevOps) with dedicated skills, tool policies, and model assignments.
The system runs on OpenClaw in Docker with SQLite persistence, Telegram for human oversight, and dynamic LLM routing across multiple providers.
See It in Action
Task 0077: The team builds a landing page about itself.
In March 2026, the 8-agent team completed its first fully autonomous pipeline run — building a GitHub Pages landing page in ~4 minutes with zero human intervention:
- 8 agents coordinated across all 10 pipeline stages
- 17 files created (HTML, CSS, JS, SVG, CI workflows)
- 5 auto-resolved decisions (no human escalation needed)
- 3 operational issues discovered and subsequently fixed
The PM defined the content strategy, the PO refined user stories, the Tech Lead decomposed the work, the Designer created a dark-theme visual system, the Frontend Dev implemented everything, QA validated, the Tech Lead reviewed, and DevOps created the PR — all autonomously.
Read the full case study: docs/case-studies/task-0077-autonomous-pipeline.md
Architecture
flowchart LR
AG[Role Agents] --> OC[OpenClaw Gateway]
OC --> PT[product-team plugin]
OC --> MR[model-router hook]
OC --> TN[telegram-notifier]
OC --> SB[stitch-bridge]
OC --> VO[virtual-office]
PT --> DB[(SQLite task DB)]
PT --> GH[GitHub via gh CLI]
CI[Local/CI pipeline] --> QG[quality-gate CLI]
The product-team extension follows hexagonal architecture with strict inward dependency flow. The 10-stage pipeline orchestrates work from IDEA to DONE with quality gates at each transition.
Prerequisites
| Dependency | Version |
|---|---|
| OpenClaw | latest |
| Node.js | 22+ |
| pnpm | 10+ |
| Docker Desktop / Engine | latest |
| GitHub CLI | latest |
Quick Start
Local:
git clone https://github.com/Monkey-D-Luisi/vibe-flow.git
cd vibe-flow
pnpm install
pnpm test
Docker:
cp .env.docker.example .env.docker
# Edit .env.docker with your credentials
docker compose build && docker compose up -d
docker compose ps
See docs/docker-setup.md for auth credentials, Telegram setup, and troubleshooting.
Agent Roster
| Agent ID | Role | Primary Model | Skills |
|---|---|---|---|
pm |
Product Manager | openai-codex/gpt-5.2 | requirements-grooming |
tech-lead |
Tech Lead | anthropic/claude-opus-4-6 | tech-lead, architecture-design, code-review, adr |
po |
Product Owner | github-copilot/gpt-4.1 | product-owner, requirements-grooming |
designer |
UI/UX Designer | github-copilot/gpt-4o | ui-designer |
back-1 |
Backend Developer | anthropic/claude-sonnet-4-6 | backend-dev, tdd-implementation, patterns |
front-1 |
Frontend Developer | anthropic/claude-sonnet-4-6 | frontend-dev, tdd-implementation |
qa |
QA Engineer | anthropic/claude-sonnet-4-6 | qa-testing |
devops |
DevOps Engineer | anthropic/claude-sonnet-4-6 | devops, github-automation |
Extensions
| Extension | Purpose |
|---|---|
| product-team | Task engine, workflow orchestration, quality tools, VCS automation, team messaging, decision engine, pipeline, nudge engine |
| quality-gate | Standalone quality engine + CLI (pnpm q:gate, pnpm q:*) for local/CI runs |
| telegram-notifier | Telegram notification integration with per-persona bot routing |
| model-router | Per-agent model routing hook with fallback chains |
| stitch-bridge | Google Stitch MCP design bridge for the designer agent |
| virtual-office | Virtual office visualization with agent activity tracking and SSE |
Tool Surface
38 tools across 10 categories: task (5), workflow (3), quality (5), vcs (4), project (3), team (5), decision (4), pipeline (7), metrics (1), agent (1). Plus 7 standalone qgate_* tools from the quality-gate extension and 8 design_* tools from stitch-bridge.
Full reference: docs/api-reference.md
Skills
17 role-specific skill libraries providing domain expertise to each agent:
| Skill | Description |
|---|---|
| adr | Architecture Decision Record management |
| agent-eval | Agent evaluation and performance assessment |
| architecture-design | System architecture design workflows |
| backend-dev | Backend development patterns and practices |
| build-error-resolver | Automated build error diagnosis and resolution |
| code-review | Structured code review workflows |
| devops | DevOps, CI/CD, and infrastructure management |
| frontend-dev | Frontend development patterns and practices |
| github-automation | GitHub workflow automation (branches, PRs, labels) |
| patterns | Software architecture patterns library |
| product-owner | Product ownership and backlog management |
| qa-testing | QA testing strategies and test planning |
| requirements-grooming | Requirements refinement and user stories |
| skill-factory | Meta-skill for creating new skills |
| tdd-implementation | Test-driven development workflows |
| tech-lead | Technical leadership and decision-making |
| ui-designer | UI/UX design workflows and design systems |
Quality Gates
| Metric | Threshold |
|---|---|
| Line coverage | ≥ 80% (major) / ≥ 70% (minor) |
| Lint errors | 0 |
| TypeScript errors | 0 |
| Avg cyclomatic complexity | ≤ 5.0 |
See the full quality gate evaluation flow.
Documentation
| Resource | Description |
|---|---|
| Architecture Diagrams | 8 Mermaid diagrams covering all subsystems |
| ADRs | 16 Architecture Decision Records (ADR-001 through ADR-016) |
| Case Studies | Detailed pipeline execution analysis |
| API Reference | Full tool surface documentation |
| Getting Started | Step-by-step setup guide |
| Roadmap | Execution history and upcoming phases |
| Docker Setup | Container deployment guide |
| Troubleshooting | Common issues and solutions |
| Security | Vulnerability reporting |
Development
pnpm test # Run all tests
pnpm lint # Lint all packages
pnpm typecheck # Type-check all packages
pnpm build # Build all packages
pnpm q:gate # Run full quality gate
Project Structure
High-level overview. See CONTRIBUTING.md for the full project tree.
vibe-flow/
.agent.md # Governance and execution contract
.agent/rules/ # Workflow rules (next task, review, PR, audits)
.agent/templates/ # Templates for tasks, walkthroughs, and reviews
AGENTS.md # Generic multi-agent operating instructions
CLAUDE.md # Claude-focused operating instructions
openclaw.json # OpenClaw runtime configuration
extensions/ # OpenClaw plugins and quality CLI package
product-team/
src/
domain/ # Task and workflow domain model
orchestrator/ # State machine, transitions, guard enforcement
persistence/ # SQLite repositories and migrations
quality/ # Runtime quality logic used by product-team tools
github/ # GitHub integration via gh CLI
tools/ # Registered OpenClaw tools (task/workflow/quality/vcs)
test/
quality-gate/
src/ # Standalone quality-gate engine
cli/ # q:gate / q:* CLI entrypoints
test/
model-router/ # Per-agent model routing hook
telegram-notifier/ # Telegram notification integration
stitch-bridge/ # Google Stitch MCP design bridge
virtual-office/ # Virtual office visualization
packages/ # Shared packages
quality-contracts/ # Shared parsers, gate policy, complexity analysis
skills/ # Role skills loaded by OpenClaw (14 roles)
site/ # Landing page (GitHub Pages)
docs/ # Product, operations, and execution documentation
Build Your Own Extension
vibe-flow is built on OpenClaw's plugin architecture. Each extension follows a
standard pattern with a register(api) entry point and an openclaw.plugin.json
manifest.
See the Getting Started Guide and API Reference to build your own extension.
Landing Page
The site/ directory contains a static landing page deployed via GitHub Pages —
built entirely by the autonomous agent team (case study).
Live: monkey-d-luisi.github.io/vibe-flow
Project Status
Alpha (v0.2.x) -- 21 epics across 15 phases completed. The system is functional with a full autonomous pipeline, dynamic model routing, budget intelligence, agent learning loop, Telegram command center, and virtual office.
See docs/roadmap_mvp.md for execution history and upcoming milestones.
Star History
Contributing
We welcome contributions! See CONTRIBUTING.md for guidelines.
Security
If you discover a vulnerability, please report it via GitHub Security Advisories. See SECURITY.md.
License
MIT. See LICENSE.
No comments yet
Be the first to share your take.