Merge pull request #56 from clawdocs/issue-55-getting-started-refresh

Refresh Getting Started section for mid-2026

Author: melvincarvalho

docs/getting-started/core-concepts.md

@@ -99,14 +99,26 @@ graph TD
     Channels["🔗 Channels<br/>WhatsApp · Telegram · Discord<br/>Slack · Signal · WebChat"] --> Gateway
 ```
 
-## Skills
+## SOUL.md (Identity & Personality)
+
+The **SOUL.md** file defines *who* the agent is — its name, personality, tone, rules, and behavioral boundaries:
+
+```markdown title="~/.openclaw/SOUL.md"
+# Agent Identity
+
+You are Jarvis, a helpful personal assistant.
+
+## Rules
+- Always respond politely
+- Never make purchases without explicit approval
+- Summarize long emails instead of forwarding them in full
+```
+
+SOUL.md is loaded into every conversation as system context. It's the primary way to customize the agent's behavior — and also the [#1 attack surface](/security/hardening#soulmd-protection) for prompt injection. Protect it with appropriate file permissions.
 
-Skills are OpenClaw's extension system. Each skill is a Markdown file with YAML frontmatter that defines:
+## Skills
 
-- **What** the skill does
-- **When** to activate it
-- **What tools** it needs
-- **How** to execute
+Skills are OpenClaw's extension system — Markdown files with YAML frontmatter that define reusable agent capabilities:
 
 ```yaml title="skills/weather.md"
 ---
@@ -122,10 +134,38 @@ and provide a concise briefing with temperature, conditions,
 and any weather alerts.
 ```
 
-Skills can be installed from [ClawHub](/guides/clawhub) or written from scratch. See [Skill Development](/guides/skill-development) for details.
+Over **10,700 skills** are available on [ClawHub](/guides/clawhub), the community marketplace. Skills can also be written from scratch — see [Skill Development](/guides/skill-development).
+
+:::warning
+ClawHub has been targeted by malicious actors. Always [verify skills](/guides/clawhub#security) before installing.
+:::
+
+## MCP (Model Context Protocol)
+
+**MCP** is the open protocol that connects OpenClaw to external tools and services. With 32,600+ MCP servers and 229,800+ tools available, it's how the agent accesses databases, APIs, cloud services, and more without custom code.
+
+```json title="~/.openclaw/openclaw.json (excerpt)"
+{
+  "mcp": {
+    "servers": {
+      "filesystem": { "command": "npx @anthropic/mcp-filesystem" },
+      "github": { "command": "npx @anthropic/mcp-github" }
+    }
+  }
+}
+```
+
+See [MCP Servers](/guides/mcp-servers) for setup and the full ecosystem.
+
+## Multi-Agent
+
+OpenClaw can spawn **sub-agents** for parallel work — an orchestrator delegates tasks to specialized workers, each with their own context and tools. This enables complex workflows like research pipelines, DevOps fleets, and virtual companies.
+
+Multi-agent comes with a 15x token cost multiplier, so it's best for tasks that genuinely benefit from parallelism. See [Multi-Agent Workflows](/guides/multi-agent) for patterns and cost management.
 
 ## Next Steps
 
 - [Architecture Overview](/architecture/overview) — Deep dive into each component
 - [Basic Usage](/guides/basic-usage) — Common patterns and workflows
 - [Configuration](/reference/configuration) — Tune every setting
+- [First 7 Days](/guides/first-7-days) — Hands-on structured learning path

docs/getting-started/installation.md

@@ -31,7 +31,22 @@ npm i -g openclaw
 openclaw onboard
 ```
 
-## Method 3: Git Clone (Developer Mode)
+## Method 3: Docker
+
+Run OpenClaw as a container — ideal for servers and VPS deployments:
+
+```bash
+docker run -d \
+  --name openclaw \
+  --restart unless-stopped \
+  -v ~/.openclaw:/root/.openclaw \
+  -p 18789:18789 \
+  ghcr.io/openclaw/openclaw:latest
+```
+
+One-click deployment is also available via **[Coolify](https://coolify.io)**, **[1Panel](https://1panel.dev)**, and **[Portainer](https://portainer.io)** templates. See [Deployment Options](/guides/deployment-options) for full Docker Compose, Kubernetes Helm, and cloud hosting guides.
+
+## Method 4: Git Clone (Developer Mode)
 
 For contributors or anyone who wants to hack on OpenClaw itself:
 

docs/getting-started/introduction.md

@@ -27,7 +27,7 @@ Unlike chatbots that wait for you to type, OpenClaw is an **autonomous agent**:
 
 - **Autonomous operation** — The [heartbeat system](/architecture/heartbeat) checks for pending tasks every 30 minutes and takes action without prompting
 - **Multi-platform messaging** — Operate through WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Teams, Feishu/Lark, and more
-- **100+ skills** — Extensible via [ClawHub](/guides/clawhub), the community skill marketplace
+- **10,700+ skills** — Extensible via [ClawHub](/guides/clawhub), the community skill marketplace with security scanning
 - **Model-agnostic** — Works with any LLM provider or [local models](/guides/local-models) via Ollama/vLLM
 - **Self-improving** — Can write code for its own new capabilities
 - **Private by default** — All data stored locally as Markdown files
@@ -74,10 +74,11 @@ OpenClaw itself is free (MIT license), but LLM API costs can add up quickly:
 | Heavy (many channels, complex skills) | $20–50+ |
 | Local models (Ollama/vLLM) | $0 |
 
-Some users have reported bills of $600+/month with heavy use. See [Local Models](/guides/local-models) for the zero-cost alternative.
+Some users have reported bills of $600+/month with heavy use. See [Local Models](/guides/local-models) for the zero-cost alternative and [Performance Tuning](/guides/performance-tuning) for techniques that have cut costs by 97%.
 
 ## Next Steps
 
 - [Install OpenClaw](/getting-started/installation) on your machine
 - Follow the [Quick Start](/getting-started/quick-start) to get running in 5 minutes
 - Understand the [Core Concepts](/getting-started/core-concepts) behind the architecture
+- Walk through the [First 7 Days](/guides/first-7-days) for a structured learning path

docs/getting-started/quick-start.md

@@ -8,6 +8,8 @@ description: Get OpenClaw running and talking to you in under 5 minutes
 
 Get from zero to a working OpenClaw agent in under 5 minutes.
 
+**What you'll have at the end:** A running OpenClaw gateway with CLI chat, able to execute shell commands, read/write files, and answer questions using your chosen LLM. Optionally connected to a messaging app so you can talk to it from your phone.
+
 ## Step 1: Install
 
 ```bash
@@ -99,16 +101,26 @@ The heartbeat makes OpenClaw truly autonomous — it checks for tasks periodical
 
 Edit `~/.openclaw/HEARTBEAT.md` with your preferences, then restart the gateway. OpenClaw will proactively act on these instructions.
 
+:::caution Cost awareness
+Each heartbeat tick costs tokens. With a 30-minute interval and Claude Opus, expect ~$5-20/day in API costs. See [Performance Tuning](/guides/performance-tuning) to reduce this — model routing alone can cut costs by 60x.
+:::
+
 ## What's Next?
 
 You now have a working OpenClaw agent. Here's where to go from here:
 
 | Goal | Guide |
 |------|-------|
+| Structured learning path | [First 7 Days with OpenClaw](/guides/first-7-days) |
 | Understand how it works | [Architecture Overview](/architecture/overview) |
 | Connect more platforms | [Channels & Integrations](/guides/channels) |
+| Browse 10,700+ community skills | [ClawHub Marketplace](/guides/clawhub) |
 | Build custom skills | [Skill Development](/guides/skill-development) |
+| Add voice and vision | [Voice & Multimodal](/guides/voice-multimodal) |
+| Build integrations | [API & Webhooks](/guides/api-webhooks) |
+| Reduce API costs | [Performance Tuning](/guides/performance-tuning) |
 | Run without API costs | [Local Models](/guides/local-models) |
+| Run multi-agent fleets | [Multi-Agent Workflows](/guides/multi-agent) |
 | Lock down security | [Security Hardening](/security/hardening) |
 | Fine-tune behavior | [Configuration Reference](/reference/configuration) |