Openclaw Starter Kit

作者 @wiziswiz

Quickly set up a proactive, memory-rich OpenClaw agent with prebuilt skills, scheduled tasks, and persistent learning in just five minutes.

🤖

OpenClaw Starter Kit

Turn your AI assistant from a stateless chatbot into a proactive, memory-rich agent that learns your patterns, tracks your work, and gets better every day.

QuickstartWhat's InsideMemorySkillsCronsPhilosophy


This kit distills months of real-world iteration into a ready-to-use workspace. It's opinionated — because the defaults matter more than the options.

🧠 The Problem

Out of the box, AI agents are goldfish. Every conversation starts from scratch. They don't remember your preferences, can't track your open threads, won't check on things proactively, and repeat the same mistakes you've already corrected.

This starter kit fixes that with:

  • Three-layer memory — facts, daily events, and learned patterns that persist across sessions
  • Self-correcting behavior — corrections you make are permanently stored and loaded every session
  • Proactive analysis — your agent checks things on its own (system health, stale threads, upcoming meetings)
  • 19 pre-built skills — from GitHub repo analysis to morning digests to Twitter research
  • 10 cron job templates — scheduled tasks that make your agent feel alive

⚡ 5-Minute Quickstart

# 1. Clone this repo
git clone https://github.com/wiziswiz/openclaw-starter-kit.git
cd openclaw-starter-kit

# 2. Run the bootstrap script
chmod +x setup.sh
./setup.sh

# 3. Fill in your identity
#    The script will tell you which files to edit

# 4. Start (or restart) OpenClaw
openclaw gateway restart

That's it. Your agent now has a memory system, proactive heartbeats, thread tracking, and security-audited skill installation.


📦 What's in the Box

openclaw-starter-kit/
│
├── 🏠 workspace/                    ← Your agent's brain
│   ├── AGENTS.md                    # Core rules: memory, search discipline, security
│   ├── SOUL.md                      # Persona + ACTIVE RULES (self-correcting behavior)
│   ├── IDENTITY.md                  # Agent name, personality, vibe
│   ├── USER.md                      # Your name, timezone, preferences
│   ├── HEARTBEAT.md                 # What to check proactively
│   ├── PATTERNS.md                  # Tacit knowledge (grows over time)
│   ├── TOOLS.md                     # Notes about your local tools
│   ├── pending-threads.json         # Open loop tracker
│   ├── memory/                      # Daily event logs
│   └── life/areas/                  # Knowledge graph
│       ├── people/                  #   → Person entities
│       ├── companies/               #   → Company entities
│       └── projects/                #   → Project entities
│
├── 🛠️ skills/                       ← 19 pre-built skills
│   ├── repo-analyzer/               # GitHub repo trust scoring & due diligence
│   ├── spawn-agent/                 # Sub-agent orchestration
│   ├── fact-extraction/             # Memory fact extraction & management
│   ├── morning-digest/              # Daily morning briefing
│   ├── company-research/            # Structured BD research briefs
│   ├── x-research/                  # Twitter/X research & search (free)
│   ├── x-research-api/             # Twitter/X research via paid API
│   ├── adversarial-review/          # Cross-model adversarial plan review
│   ├── consensus-researcher/        # Multi-source weighted consensus research
│   ├── google/                      # Gmail, Calendar, Drive integration
│   ├── health-tracking/              # Wearable health data (WHOOP, Oura, etc.)
│   ├── healthcheck/                 # System security hardening
│   ├── post-update-health-check/    # Post-update regression verification
│   ├── baseline-ui/                 # UI quality baseline enforcement
│   ├── source-library/              # Auto-file & cross-reference shared URLs
│   ├── fixing-accessibility/        # WCAG accessibility fixes
│   ├── fixing-metadata/             # SEO & Open Graph metadata
│   ├── fixing-motion-performance/   # Animation performance optimization
│   ├── vercel-react-best-practices/ # 57 React/Next.js performance rules
│   └── web-design-guidelines/       # Web interface compliance review
│
├── ⏰ crons/                        ← 10 scheduled job templates
│   └── starter-crons.json           # Morning digest, nightly review, etc.
│
├── 📚 docs/                         ← Guides
│   ├── MEMORY-SYSTEM.md             # How the 3-layer memory works
│   ├── SKILLS-GUIDE.md              # Creating and using skills
│   ├── CRON-GUIDE.md                # Setting up proactive crons
│   └── PHILOSOPHY.md                # The 10 principles behind great agents
│
└── 💡 examples/                     ← Reference configs
    ├── example-patterns.md          # Common tacit knowledge patterns
    ├── example-heartbeat.md         # Heartbeat variants (dev/founder/researcher)
    └── example-cron-setup.md        # Step-by-step cron setup commands

🧠 The Memory System

Most AI agents forget everything between sessions. This kit gives your agent three layers of persistent memory:

┌─────────────────────────────────────────────────────┐
│                  LAYER 1: Knowledge Graph            │
│                                                      │
│   life/areas/people/alice/summary.md                │
│   life/areas/companies/acme/items.json              │
│   life/areas/projects/my-app/summary.md             │
│                                                      │
│   → Entity-based facts with typed lifecycle          │
│   → "Who is Alice?" → instant structured answer      │
└─────────────────────────────────────────────────────┘
                         │
┌─────────────────────────────────────────────────────┐
│               LAYER 2: Daily Notes                   │
│                                                      │
│   memory/2026-03-07.md                              │
│   memory/2026-03-06.md                              │
│   memory/2026-03-05.md                              │
│                                                      │
│   → What happened today (raw event log)              │
│   → Written continuously, not batched                │
│   → "What did we do last Tuesday?" → date search     │
└─────────────────────────────────────────────────────┘
                         │
┌─────────────────────────────────────────────────────┐
│            LAYER 3: Tacit Knowledge                  │
│                                                      │
│   PATTERNS.md                                        │
│                                                      │
│   → Patterns about how YOU work                      │
│   → "User works late, don't suggest breaks"          │
│   → "Test config on 1 system before batch rollout"   │
│   → Grows organically from corrections               │
└─────────────────────────────────────────────────────┘

Key insight: Manual memory writes > auto-capture. Deliberate saves are higher quality than automated logging. The agent decides what's worth remembering, not a parser.

→ Full details in docs/MEMORY-SYSTEM.md


🔄 The Self-Correction Loop

This is the single most impactful pattern in the kit:

You correct your agent
        │
        ▼
Agent adds correction to SOUL.md ACTIVE RULES
        │
        ▼
SOUL.md is loaded EVERY session
        │
        ▼
Correction persists forever ✨
        │
        ▼
PATTERNS.md captures the "why" for deeper understanding

Example: You tell your agent "don't use markdown tables on Telegram, they render as garbage." The agent:

  1. Immediately adds to SOUL.md: NO TABLES ON TELEGRAM — bullets only
  2. Adds to PATTERNS.md: the reasoning and context
  3. Never makes that mistake again

The difference between a good agent and a great one is how it handles corrections.


🛠️ Skills Library

Skills are specialized instruction files that teach your agent how to handle specific tasks. Install by copying to ~/.openclaw/skills/.

Research & Analysis

SkillWhat it does
repo-analyzerTrust-score any GitHub repo (security, code quality, contributor patterns)
company-researchStructured research briefs for outreach or due diligence
x-researchSearch Twitter/X for real-time perspectives and expert opinions
x-research-apiPaid X/Twitter API research (higher engagement results). Companion to x-research — use when you need tweets with 1K+ likes vs free version's ~23 likes.
consensus-researcherMulti-source research (Reddit, HN, Amazon, YouTube, expert sites) with weighted consensus scoring. Great for product/market research.
adversarial-reviewCross-model adversarial plan review — two models alternate writer/reviewer until convergence. Use before any major build or implementation.

Productivity

SkillWhat it does
morning-digestDaily briefing: calendar, threads, weather, action items
spawn-agentDeploy sub-agents for parallel or long-running tasks
fact-extractionExtract and organize facts from conversations into memory
googleGmail search/send, Calendar events, Drive file access
source-libraryAuto-file shared URLs with structured summaries, cross-referencing, confidence decay, and conflict detection. Your agent's second brain for everything you read.

Health & Wellness

SkillWhat it does
health-trackingWearable integration (WHOOP, Oura, Apple Health, Garmin) — recovery scores, sleep trends, fitness streaks

💡 Your agent can pull your daily recovery score and weave it into your morning digest. Green day? Stack the hard meetings. Red day? Go easy. See skills/health-tracking/SKILL.md for setup.

System & Security

SkillWhat it does
healthcheckSecurity hardening audit (firewall, SSH, updates, exposure)
post-update-health-checkVerify nothing broke after an OpenClaw update

Development Quality (6 skills)

SkillWhat it does
baseline-uiEnforce consistent, accessible UI patterns
fixing-accessibilityWCAG-compliant keyboard, screen reader, focus fixes
fixing-metadataSEO tags, Open Graph, canonical URLs, structured data
fixing-motion-performanceFix janky animations and rendering bottlenecks
vercel-react-best-practices57 prioritized React/Next.js performance rules from Vercel
web-design-guidelinesWeb Interface Guidelines compliance review

🔄 Keeping Skills Updated

Third-party skills can receive upstream updates. A weekly cron checks for new commits automatically and pings you if anything's worth pulling.

To manually check or pull updates for a skill:

cd ~/.openclaw/skills/<skill-name>
git fetch && git log HEAD..origin/main --oneline
git pull  # if updates look safe

For SME (Structured Memory Engine), updates require a rebase — see ~/Projects/Structured-Memory-Engine/README.md for the merge guide.


⏰ Proactive Crons

Cron jobs are what make your agent feel alive — it does things without being asked.

The kit includes 10 starter cron templates (crons/starter-crons.json):

CronScheduleWhat it does
🌅 Morning Digest7:30 AM dailyCalendar, threads, weather, action items
🔍 Nightly Self-Review1:00 AM dailyGrade today's performance, find improvement areas
🛡️ Config WatchdogEvery 15 minDetect silent config changes or drift
💾 Git Backup3:00 AM dailyAuto-commit and push workspace changes
📞 Meeting PrepEvery 30 min (business hours)Research upcoming call participants
💪 Health Summary7:30 AM dailyPull wearable recovery/sleep, flag low days
🧠 Weekly SynthesisSunday 10 AMConsolidate the week's facts and patterns
📋 Thread Cleanup7:00 PM dailySurface stale open threads needing follow-up
📝 Fact Extraction9:00 PM dailyExtract durable facts from today's conversations
🧹 Session CleanupSunday 4 AMPrune old cron/sub-agent session files (they accumulate forever and bloat memory)

→ Step-by-step setup in examples/example-cron-setup.md


🏗️ Workspace Files Explained

AGENTS.md — The Rulebook

Your agent's operating manual. Contains:

  • Search Before Speaking — never claim ignorance without grepping first
  • Three-Layer Memory System — how to store and retrieve information
  • Continuous Memory Writes — write immediately, don't wait for batch flushes
  • Security Audit Protocol — mandatory review before installing any new skill
  • Tool Failure Recovery — try 3 approaches before reporting failure
  • Correction Protocol — how to handle behavior corrections

SOUL.md — Persona + Active Rules

Two sections:

  1. Persona — your agent's personality (confident, concise, playful, whatever you want)
  2. Active Rules — the 5-10 most important behavior corrections, loaded every session

HEARTBEAT.md — Proactive Checklist

What your agent checks on heartbeat polls:

  • System health (OpenClaw updates, cron errors)
  • Open threads older than 48 hours
  • Upcoming calendar events needing research
  • Patterns suggesting new skills to create

See examples/example-heartbeat.md for variants tailored to developers, founders, and researchers.

PATTERNS.md — Living Knowledge

Starts empty. Grows as your agent learns:

  • Debugging patterns ("always check HTTP method before calling")
  • Deployment lessons ("test on 1 system before batch rollout")
  • Your preferences ("user works late, don't suggest breaks")
  • Tool quirks ("this API returns 429 after 3 calls, back off")

See examples/example-patterns.md for common starter patterns.


🧭 Philosophy

Ten principles that separate great agent setups from mediocre ones:

  1. Manual memory beats auto-capture. Quality > quantity.
  2. Search before speaking. grep costs 1 tool call. Guessing costs 3+.
  3. Corrections go to SOUL.md. It's the only file loaded every session.
  4. Evening cleanup produces artifacts. Quiet time = drafts, research, thread updates.
  5. Benchmark before implementing. Evaluate first, build second.
  6. Isolate before fixing. Confirm which system is broken before changing anything.
  7. Test on one before rolling out. One cron, one config, one system.
  8. Deliver, don't prepare. Internal prep ≠ completion. User-facing delivery = done.
  9. Write immediately. Don't wait for batch flushes. Facts decay in RAM.
  10. Build skills from repetition. 3+ times = create a skill.

→ Full breakdown with examples in docs/PHILOSOPHY.md


🚀 Getting Started

After running setup.sh:

  1. Edit ~/clawd/IDENTITY.md — Give your agent a name and personality
  2. Edit ~/clawd/USER.md — Tell your agent who you are
  3. Edit ~/clawd/SOUL.md — Customize the persona (or keep the default)
  4. Start talking — Your agent will begin building memory from day one
  5. Correct freely — Every correction makes it permanently smarter
  6. Set up crons — See examples/example-cron-setup.md for the essentials

Optional Power-Ups

  • Install skills → Copy from skills/ to ~/.openclaw/skills/
  • Enable Google integration → Follow skills/google/SKILL.md setup
  • Add Obsidian sync → Dual-write to an Obsidian vault for graph visualization
  • Create custom skills → See docs/SKILLS-GUIDE.md

🤝 Contributing

Found a pattern that makes your agent better? A skill that's universally useful? Open a PR.

The best contributions are battle-tested patterns — things you've run for weeks and proven work.


📝 License

MIT — use it, fork it, make it yours.


Built from months of real-world iteration. Every pattern earned, not invented.

Workspace

更新时间 2026/03/22 13:18:09发布方式 clawlodge-cli/0.1.8
README.mdtext · 17.5 KB

还没有评论。

相关推荐