/SKILL.md
Multi-agent collaborative development. Gemini designs UI/UX (design spec + presentation code), Claude Code fills business logic (TODO functions only), Codex reviews code quality. Use when user says "协作开发", "collab dev", "multi-agent build", or requests multi-agent collaboration for building web apps. Best for visually intensive projects (dashboards, data viz, animated UIs). For simple CRUD apps, solo development may suffice.
npx skillsauth add laolin5564/collab-dev collab-devInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
| Agent | Does | Does NOT |
|-------|------|----------|
| Gemini | design.md + all visible code (CSS/HTML) | write business logic |
| OpenClaw | spec.md + server skeleton + TODO markers + validation + fixes | write UI styles or fill TODOs |
| Claude Code | fill // TODO function bodies only | change CSS/HTML structure |
| Codex | write review report | modify any code |
On first trigger, check if ~/.openclaw/skills/collab-dev/.ready exists. If yes, skip to Phase 0. If not, run:
READY=1
# 1. Gemini CLI
GEMINI=$(which gemini 2>/dev/null)
if [ -n "$GEMINI" ]; then
echo "✅ Gemini CLI: $($GEMINI --version 2>&1 | head -1)"
else
echo "❌ Gemini CLI 未安装"; READY=0
fi
# 2. Claude Code CLI
CLAUDE=$(which claude 2>/dev/null)
if [ -n "$CLAUDE" ]; then
echo "✅ Claude Code: $($CLAUDE --version 2>&1 | head -1)"
else
echo "❌ Claude Code 未安装"; READY=0
fi
# 3. Codex CLI
CODEX=$(which codex 2>/dev/null)
if [ -n "$CODEX" ]; then
echo "✅ Codex CLI: $($CODEX --version 2>&1 | head -1)"
else
echo "❌ Codex CLI 未安装"; READY=0
fi
# 4. ACP config (optional but recommended)
ACPX_CFG="$HOME/.acpx/config.json"
if [ -f "$ACPX_CFG" ]; then
echo "✅ ACP 配置: $ACPX_CFG"
else
echo "⚠️ ACP 配置未找到(可选,OpenClaw 内置 ACP 也可用)"
fi
If any ❌, show install guide and stop:
| Tool | Install | Auth |
|------|---------|------|
| Gemini CLI | npm i -g @google/gemini-cli | gemini (首次运行自动引导登录) |
| Claude Code | npm i -g @anthropic-ai/claude-code | claude (首次运行自动引导登录) |
| Codex CLI | npm i -g @openai/codex | codex (首次运行自动引导登录) |
Tell user: "装好后再说一次「协作开发」即可。"
安装遇到问题? → 读 references/troubleshooting.md(涵盖 OAuth、ACP 权限、双 auth 不同步、session 堆积等 6 个常见坑)
If all ✅, write marker and proceed:
touch ~/.openclaw/skills/collab-dev/.ready
If user only provides a brief idea (e.g. "做一个番茄钟"), clarify requirements before generating PRD.
Step 1: Requirement Clarification (ask until clear)
Ask user these questions one round at a time (don't dump all at once). Stop asking when answers are sufficient:
Skip questions the user already answered. If user says "你决定" or "随便", make reasonable defaults and state them.
Step 2: Generate PRD
Read references/prd-guide.md for template. Generate PRD with user stories, feature list (P0/P1/P2), data model, interaction patterns, acceptance criteria.
Step 3: User confirms PRD → convert to spec.md
Skip Phase -1 entirely if user already provides a detailed spec or PRD.
Write spec.md with:
.hidden class toggle for modalsDo not write UI style hints — leave design freedom to Gemini.
gemini -p outputs design.md. Prompt must include:
1. :root {} CSS variables (colors, sizes, shadows, transitions)
2. HTML class naming for every component
3. Interaction states: .hidden class for all JS show/hide
4. Mobile: keep all P0 features accessible (off-canvas/drawer, no display:none)
gemini -p outputs public/index.html. Prompt rules:
1. Names must exactly match design.md
2. No business logic
3. Use .hidden class for JS show/hide, never inline display:none
4. Mobile: drawer/off-canvas for sidebars
Clean output: sed -i '' '1{/^```/d}; ${/^```/d}' public/index.html
Quality gate: verify design.md names appear in code.
Write server/backend with:
Integrate Gemini's UI code:
.hidden class mechanism worksSave Gemini's original UI: cp <ui-file> <ui-file>.gemini-original
Inject skeleton with TODO markers:
// TODO: functionName — description
// Input: params
// 1. step one
// 2. step two
// Output: return value
3a. Claude Code fills TODOs (ACP, 5min timeout):
Only modify // TODO function bodies.
Do not change CSS/HTML/DO NOT MODIFY blocks.
Fallback: if no file change in 3min → kill, OpenClaw fills.
3b. Codex pre-reviews skeleton (ACP, simultaneous).
Phase 4 checks (adapt commands to project's tech stack):
4a. Completeness
4b. Design consistency
4c. CSS/JS integration
display:none outside .hidden class and media queries4d. Security
4e. DOM/UI structure
4f. Spec coverage
diff index.html index.html.gemini-original — Claude didn't touch presentationDimensions: functionality, design consistency, security, code quality, performance, bug risk (1-10 each).
Fix Critical/High from review → re-validate → deliver.
| Issue | Workaround |
|-------|-----------|
| Gemini ACP unavailable | Always use gemini -p CLI |
| Gemini markdown wrapping | Phase 1b: sed cleanup |
| Claude ACP hangs | 5min timeout + fallback |
| Claude modifies presentation | Phase 5 diff check |
| Standard tool projects: Gemini adds little value | Best for visually intensive projects |
development
Maintainer-only workflow for handling GitHub Secret Scanning alerts on OpenClaw. Use when Codex needs to triage, redact, clean up, and resolve secret leakage found in issue comments, issue bodies, PR comments, or other GitHub content.
development
Maintainer workflow for OpenClaw releases, prereleases, changelog release notes, and publish validation. Use when Codex needs to prepare or verify stable or beta release steps, align version naming, assemble release notes, check release auth requirements, or validate publish-time commands and artifacts.
development
Run, watch, debug, and extend OpenClaw QA testing with qa-lab and qa-channel. Use when Codex needs to execute the repo-backed QA suite, inspect live QA artifacts, debug failing scenarios, add new QA scenarios, or explain the OpenClaw QA workflow. Prefer the live OpenAI lane with regular openai/gpt-5.4 in fast mode; do not use gpt-5.4-pro or gpt-5.4-mini unless the user explicitly overrides that policy.
development
End-to-end Parallels smoke, upgrade, and rerun workflow for OpenClaw across macOS, Windows, and Linux guests. Use when Codex needs to run, rerun, debug, or interpret VM-based install, onboarding, gateway smoke tests, latest-release-to-main upgrade checks, fresh snapshot retests, or optional Discord roundtrip verification under Parallels.