Project Bootstrap
Turn a project idea into a running multi-agent development pipeline in one session.
Overview
This skill codifies the workflow for:
- Agent Team Design — break complex work into specialized agents
- Taskboard Setup — CLI-based task management across agents
- GitHub Repo + CI/CD — TDD pipeline with Discord notifications
Phase 1: Agent Team Design
Analyze the Project
Before creating agents, answer these questions:
- What are the 3-5 major workstreams? (e.g., frontend, backend, research, design)
- Which workstreams need different expertise or thinking styles?
- What's the dependency graph between workstreams?
Design Agent Roles
For each workstream, create an agent with a SOUL.md following this structure:
# Agent Name — Nickname Emoji
You are **Nickname**, the [Role] — [one-line mission].
## 🧠 Identity & Memory
- **Role**: [specific expertise]
- **Personality**: [3-4 traits that affect work style]
- **Memory**: [what context files they track]
- **Experience**: [what failure modes they've seen]
## 🎯 Core Mission
[2-4 responsibility groups with specifics]
## 🚨 Critical Rules
[Non-negotiable constraints — security, process, boundaries]
## 📋 Deliverables
[Concrete outputs this agent produces]
## 🎯 Success Metrics
[How to measure if the agent is doing well]
## 💬 Communication Style
[How the agent communicates — tone, format, language]
## 🔗 Workflow Position
[Where in the pipeline: who feeds input, who receives output]
Register Agents in Config
For each agent, add to openclaw.json:
{
"id": "agent-id",
"name": "agent-id",
"agentDir": "/path/to/workspace/agents/agent-id",
"model": "model-alias",
"tools": {
"profile": "full",
"deny": ["gateway"]
}
}
Key decisions:
- Model selection: Use cheaper models (Haiku/Sonnet) for routine work, expensive (Opus) for architecture/review
- Tool access: Deny
gateway for all agents except main. Deny message for pure code agents.
- Subagent allowlist: Main agent lists which agents it can spawn
Wire Discord Bindings (if multi-bot)
If agents have separate Discord bots, add bindings:
{
"bindings": [
{ "agentId": "tech-lead", "match": { "channel": "discord", "accountId": "bot-name" } }
]
}
Phase 2: Taskboard Setup
Install Taskboard CLI
See references/taskboard-setup.md for the full taskboard CLI setup guide including:
- Task schema (id, title, status, assignee, priority, dependencies)
- CLI commands (create, list, assign, update, close)
- Cross-agent task handoff protocol
- Integration with cron jobs for status checks
Task Lifecycle
📋 backlog → 🔄 in-progress → 👀 review → ✅ done
↓ ↓
🚫 blocked ❌ rejected → 🔄 in-progress
Cross-Agent Handoff
When an agent completes a task that feeds into another agent's work:
- Update task status to
review
- Create a new task for the downstream agent referencing the completed task
- Send notification to the downstream agent's Discord channel
Phase 3: GitHub Repo + CI/CD
Repository Setup
# Initialize repo
gh repo create <org>/<project> --private --clone
cd <project>
# Branch protection
gh api repos/<org>/<project>/rulesets -X POST --input .github/ruleset.json
# Required structure
mkdir -p .github/workflows tests src docs/adr
TDD Pipeline
See references/ci-cd-templates.md for GitHub Actions workflow templates:
- test.yml: Run tests on every PR and push to main
- lint.yml: Code style checks
- deploy.yml: Deploy on merge to main (if applicable)
Discord Notifications
Add Discord webhook to GitHub repo:
# Create webhook in Discord channel (Server Settings → Integrations → Webhooks)
# Add to GitHub: Settings → Webhooks → Add webhook
# Or use GitHub Actions:
See references/ci-cd-templates.md for the Discord notification action template.
ADR (Architecture Decision Records)
Every significant technical decision gets an ADR:
# ADR-NNN: [Title]
## Status: [proposed | accepted | deprecated | superseded]
## Context: [Why this decision is needed]
## Decision: [What we decided]
## Consequences: [Trade-offs and implications]
Execution Checklist
Run through this for every new project:
