OpenClaw Setup & Customization

Provisioning

1. Navigate to Settings → OpenClaw
2. Click Enable OpenClaw
3. Heard provisions a dedicated container for your organization

The container runs Heard’s two built-in skills:

• heard-brain — The conversational interface (query Heard from any connected channel)
• heard-proactive-scan — The scheduled scan engine

---

Connecting a Messaging Channel

The OpenClaw communicates through any channel supported by OpenClaw. Slack is the most common, but WhatsApp, Telegram, and other connectors are available.

Slack (Primary Example)

1. In Settings → OpenClaw → Channels, click Connect Slack
2. Authorize Heard in your Slack workspace
3. Select the notification channel — where proactive alerts and action plan notifications will be posted (e.g., #customer-intelligence)
4. Click Save

The OpenClaw appears as a bot user in your workspace. Team members can DM it or mention it in channels with @OpenClaw.

Other Channels

To connect additional channels:

1. In Settings → OpenClaw → Channels, click Add Channel
2. Select the channel type (WhatsApp, Telegram, etc.)
3. Follow the OAuth flow for that platform
4. Configure the notification destination

The OpenClaw works identically across all channels — same queries, same approvals, same proactive alerts. Your team can use whichever channel they prefer.

---

Proactive Scan Configuration

Configure what the OpenClaw checks for and when:

Schedule

Setting	Default	Description
Frequency	Weekdays at 9:00 AM	When the scan runs (cron schedule)
Timezone	Your organization’s timezone	So 9:00 AM means 9:00 AM for your team
Channel	The channel selected during setup	Where alerts are posted

What It Scans For

Check	Description	Default
Stale Actions	Triage cards with no activity for N days	✅ On (3 days)
Sentiment Drops	Customers whose sentiment declined over recent interactions	✅ On
Time Bomb Warnings	Actions approaching auto-execution	✅ On
Unread Triage	Triage cards not viewed within N hours	✅ On (48 hours)
Weekly Metrics	Team performance summary (sentiment avg, actions completed, response time)	✅ On (Fridays)

Customizing Thresholds

1. Navigate to Settings → OpenClaw → Scan Rules
2. Adjust the defaults:
   • “Flag stale actions after 5 days instead of 3”
   • “Only alert on sentiment drops of 0.3 or more”
   • “Disable weekly metrics”
3. Click Save

---

Conversational Commands

The OpenClaw responds to natural language. Here are common commands:

Command	What Happens
”What are the open issues for [customer]?”	Shows the customer’s profile and open issues
”Show me at-risk customers”	Lists customers with risk score > 0.7
”What happened with [customer] this week?”	Shows recent interactions and analysis
”Run a scan now”	Triggers an immediate proactive scan
”Approve plan #1234”	Approves the specified action plan
”Hold plan #1234”	Pauses a Time Bomb countdown
”Show my team’s stats”	Shows current team metrics

---

Custom Instructions

Standing instructions are persistent rules that shape how the OpenClaw handles requests and scans. They apply alongside your team’s playbook.

Adding Instructions

1. Navigate to Settings → OpenClaw → Custom Instructions
2. Click Add Instruction
3. Write the instruction in plain language:

“When asked to research a company, always check HubSpot for existing data first, then search the web. Format the result as: Company Overview, Recent News, Key Contacts, Relationship History.”

4. Set the scope:

Scope	When It Applies
Always	Every OpenClaw interaction and scan
On request	Only when the user asks about a relevant topic
Scheduled	Runs on a cron schedule (e.g., “Every Friday at 5pm”)

5. Click Save

Instruction Priority

When multiple instructions could apply, the OpenClaw follows this order:

1. Custom instructions (configured here — most specific)
2. Team playbook (configured in Team Setup)
3. Default behavior

If a custom instruction conflicts with a team playbook rule, the custom instruction wins.

---

Custom Skills

Extend your OpenClaw with additional OpenClaw skills from ClawHub.

Installing a Skill

1. Browse available skills at clawhub.io
2. Navigate to Settings → OpenClaw → Skills
3. Click Install Skill
4. Enter the skill ID (e.g., web-research, spreadsheet-sync)
5. If the skill requires credentials (e.g., a Google Sheets API key), enter them
6. Click Install

The skill is now available to your OpenClaw. You can test it immediately:

@OpenClaw Use web-research to find recent news about Acme Corp

Managing Skills

In Settings → OpenClaw → Skills, you can:

• View installed skills — see version, last used, and status
• Update skills — apply new versions from ClawHub
• Remove skills — uninstall with one click
• Configure credentials — update API keys or tokens per skill

Security

Your OpenClaw runs inside an OpenShell sandbox — an open-source runtime that enforces strict security boundaries:

• Filesystem isolation — skills can only access designated paths; they cannot read other tenants’ data or system files
• Network control — only your Heard API and registered MCP servers are reachable; all other outbound connections are blocked
• Credential protection — API keys and tokens are injected at runtime as secure Providers, never stored on the container’s filesystem
• Each skill only accesses the credentials you explicitly configure
• The OpenClaw’s HITL rules still apply — a custom skill can’t bypass approval requirements