OpenClaw Chip Agent Team

English | 简体中文

Open-source role templates for building a serious chip and SoC multi-agent team in OpenClaw.

A reusable OpenClaw agent squad for chip architecture, RTL, DV, STA, PD, DFT, analog, and layout collaboration.

Start Here

• One-line install: bash -c "$(curl -fsSL https://raw.githubusercontent.com/kichy-ge/openclaw-chip-agent-team/main/scripts/install-from-github.sh)"
• OpenClaw chat install: paste one sentence and let the agent do the setup
• Full guide: see INSTALL.md / INSTALL.zh-CN.md

OpenClaw One-Sentence Install

Please install the GitHub repo `kichy-ge/openclaw-chip-agent-team` into my local OpenClaw environment, run its installer, and generate the config template under `~/.openclaw/`.

Best for teams who want:

• role-separated chip discussion instead of one blended AI voice
• requirement, architecture, RTL, DV, STA, PD, DFT, analog, and layout perspectives in one workflow
• reusable OpenClaw agent templates instead of a one-off demo
• Chinese-first technical collaboration with explicit assumptions and ownership

This repository packages a reusable collaboration structure for:

• Chip Architect
• Chip Product Manager
• RTL Frontend Designer
• Verification Engineer
• DFT Engineer
• Physical Design Engineer
• STA Signoff Engineer
• Analog Designer
• Custom Layout Engineer

The goal is not to ship a fake demo. The goal is to give you a practical team that can turn requirement debates into role-based architecture, implementation, verification, timing, and backend decisions.

At A Glance

• 9 reusable role templates
• OpenClaw config template included
• local installer included
• safe-to-publish structure with secrets removed
• designed for real chip collaboration, not only orchestration screenshots

Team Overview

flowchart LR
    classDef input fill:#0e2236,stroke:#4aa3ff,color:#eef7ff,stroke-width:1.5px;
    classDef core fill:#15354f,stroke:#79d2ff,color:#f5fbff,stroke-width:1.5px;
    classDef specialist fill:#1e243d,stroke:#8c93ff,color:#f7f7ff,stroke-width:1.2px;
    classDef output fill:#173b32,stroke:#63d2a1,color:#f3fff8,stroke-width:1.5px;

    U["Product / Chip Question"]
    SUM["Role-Based Decision Output"]

    subgraph CORE["Core Decision Layer"]
        direction LR
        PM["Chip PM"]
        ARC["Architect"]
    end

    subgraph SPEC["Specialist Review Mesh"]
        direction TB
        RTL["RTL"]
        DV["DV"]
        DFT["DFT"]
        STA["STA"]
        PD["PD"]
        ANA["Analog"]
        CL["Custom Layout"]
    end

    U --> PM
    U --> ARC
    PM --> ARC

    ARC --> RTL
    ARC --> DV
    ARC --> DFT
    ARC --> STA
    ARC --> PD
    ARC --> ANA
    ARC --> CL

    PM --> SUM
    ARC --> SUM
    RTL --> SUM
    DV --> SUM
    DFT --> SUM
    STA --> SUM
    PD --> SUM
    ANA --> SUM
    CL --> SUM

    class U input;
    class PM,ARC core;
    class RTL,DV,DFT,STA,PD,ANA,CL specialist;
    class SUM output;

Loading

This team is designed to turn one incoming chip question into role-separated outputs instead of one blended answer. PM frames scope, Architect judges partition and ownership, and specialists challenge feasibility from implementation, verification, timing, backend, test, analog, and layout angles.

Why This Repo Exists

Most “multi-agent” demos stop at pretty orchestration. Real chip work is different:

• requirements and version scope must be explicit
• architecture and implementation ownership must be separated
• verification and signoff cannot be hand-waved
• backend and timing consequences must surface early

This template tries to make those role boundaries real.

Team Roles

Core

• Architect: partition, interface, gate decision. Output: architecture judgement, risk split, next owner.
• Chip PM: requirement baseline, version scope. Output: scope decision, must/should/later, acceptance.

Frontend Delivery

• RTL: micro-architecture, protocol, datapath. Output: implementation verdict, interface assumptions.
• DV: proof obligation, coverage, closure. Output: verification gate, missing evidence.
• DFT: testability and manufacturability. Output: scan, ATPG, and MBIST implications.

Signoff and Backend

• STA: timing integrity and fix ownership. Output: timing verdict, signoff risk.
• PD: backend feasibility and QoR closure. Output: floorplan, CTS, and route recommendations.

Analog and Custom

• Analog: analog intent and characterization. Output: PVT, noise, and performance risks.
• Custom Layout: matching, parasitics, extraction readiness. Output: DRC, LVS, and PEX-oriented guidance.

What Is Included

• Role contracts in templates/agents/*/SOUL.md
• Human-readable role playbooks in templates/agents/*/PLAYBOOK.md
• Reusable workspace bootstrap files in templates/common/
• A sanitized OpenClaw config template in openclaw.template.json
• An installer in scripts/install-chip-team.mjs
• A sample multi-agent discussion in examples/discussions/product-branch-evaluation.md

Design Principles

• Chinese-first role collaboration
• direct judgement before long explanation
• explicit assumptions instead of polished ambiguity
• clear ownership between PM, Architect, RTL, DV, STA, PD, DFT, Analog, and Layout
• version and scope awareness
• no API keys, logs, private memories, or runtime transcripts in the repo

Repository Layout

templates/
  common/
  agents/
scripts/
examples/
openclaw.template.json
README.md

Quick Start

1. Install OpenClaw locally.
2. Clone this repository.
3. Run:

node scripts/install-chip-team.mjs

4. Review the generated workspaces under ~/.openclaw/.
5. Merge or adapt openclaw.chip-team.template.json into your own ~/.openclaw/openclaw.json.
6. Replace placeholder provider keys with real credentials.
7. Restart the OpenClaw gateway.

If you want to install into another OpenClaw home:

node scripts/install-chip-team.mjs /path/to/openclaw-home

One-Line Install

If you just want to install the team into your local OpenClaw home:

bash -c "$(curl -fsSL https://raw.githubusercontent.com/kichy-ge/openclaw-chip-agent-team/main/scripts/install-from-github.sh)"

If you want to install into a custom OpenClaw home:

bash -c "$(curl -fsSL https://raw.githubusercontent.com/kichy-ge/openclaw-chip-agent-team/main/scripts/install-from-github.sh)" -- /path/to/openclaw-home

Install From Inside OpenClaw

You can also paste this directly into an OpenClaw chat:

Please install the GitHub repo kichy-ge/openclaw-chip-agent-team into my local OpenClaw environment, run its installer, and generate the config template under ~/.openclaw/.

Chinese version:

请把 GitHub 仓库 kichy-ge/openclaw-chip-agent-team 安装到我本地的 OpenClaw 环境里，执行安装脚本，并在 ~/.openclaw/ 下生成配置模板。

For all three installation paths, see INSTALL.md and INSTALL.zh-CN.md.

What You Get After Install

The installer generates:

• workspace-architect
• workspace-chip-pm
• workspace-rtl
• workspace-dv
• workspace-dft
• workspace-pd
• workspace-sta
• workspace-analog
• workspace-custom-layout
• openclaw.chip-team.template.json

What You Should Customize

• model provider and API keys
• USER.md project context for each role
• workspace paths if you use a different OpenClaw home
• which specialist agents the architect is allowed to dispatch
• whether your product direction is DDR-less, always-on, multi-camera, NPU-heavy, or something else

Suggested Project Context Fields

The most important thing to customize is each role’s USER.md. Recommended fields:

• current product line
• current version or milestone
• locked constraints
• key risks
• what is explicitly out of scope
• preferred output language and style

Recommended Use Cases

• requirement boundary review
• architecture gate discussion
• whether a request is v1.0, v2.0, or a new SKU
• RTL feasibility and interface review
• verification scope definition
• timing/signoff risk surfacing
• backend implication review before implementation starts

Example Questions This Team Should Handle Well

• Is this request still a v1.0 change, or already a new product branch?
• Which assumptions block RTL from starting implementation?
• What exactly must DV prove before we call the scope verifiable?
• Is this a local timing fix, or an MMMC/signoff-model change?
• Can PM lock the product boundary now, or is an architect gate still required?

Launch Copy

See docs/launch/launch-copy.md for bilingual GitHub launch copy and social post drafts.

Contributors

• kichy_ge (@kichy-ge): project direction, role design, chip collaboration framing, and OpenClaw integration.
• Codex: implementation partner for documentation polish, installation flow refinement, assets, and launch materials.

Non-Goals

This repository does not include:

• your real OpenClaw runtime logs
• your private memory files
• real provider keys
• full turnkey GitHub Actions or deployment automation
• a replacement for a real design flow

It is a team template, not a full EDA platform.

Contributing

See CONTRIBUTING.md or CONTRIBUTING.zh-CN.md.

License

MIT