Part 5 of 6

Skills, Cron Jobs, and Automation

Transform your conversational AI into an autonomous agent with curated skills, scheduled tasks, webhook integrations, and real-time email events.

30 minutes

Automation-focused

Understanding the Skills System

Skills are capability packs — collections of tool definitions, prompts, and code that extend what your agent can do. Skills ship from three sources:

Bundled skills — included with OpenClaw, maintained by the core team. Safe to use without auditing.

ClawHub community skills — the public registry at clawhub.ai with 13,000+ packages. Not vetted by the OpenClaw team.

Workspace skills — skills you write yourself, stored in ~/.openclaw/workspace/skills/.

Safe Skill Installation Practices

⚠️ Security Warning

Security researchers have found that between 12 and 20 percent of ClawHub skills in audits contained malicious code, data exfiltration, or prompt injection payloads. Treat every ClawHub skill as executable code from a stranger, because that is exactly what it is.

Before installing any community skill:

Check the ClawHub page for the skill's VirusTotal scan report. A flagged report is a hard stop.
Review the source on GitHub. Read the actual tool definitions and any scripts it executes.
Pin the version. Installing @latest means the author can push a malicious update.
Review outbound network permissions. Unexpected external calls are a red flag.
Check install count and age. High count + long history = lower risk.

Skill management

# List bundled skills
openclaw skills list --bundled

# Install a community skill pinned to a version
openclaw skills install clawhub:steipete/github@2026.3.1

# Install from official skills repo
openclaw skills install github.com/openclaw/skills/github

# View installed skills
openclaw skills list

# Remove a skill
openclaw skills remove github

Curated Skills Worth Installing

Productivity & Communication

gmail — Read, search, label, archive, and send email via Gmail API (bundled)
google-calendar — Query and create calendar events
notion — Read from and write to Notion databases and pages
github — Query repos, create issues, review PRs, and search code

System & Infrastructure

shell — Executes shell commands on the VPS (bundled). Scope carefully — runs as openclawops
docker-control — List containers, read logs, restart services, check resources
system-health — Reports CPU, memory, disk, and load average

Research & Web

browser — Headless Chromium web browsing, form filling, screenshots
web-search — Search via Brave, Serper, or SerpAPI. Cheaper than browser for lookups
rss-reader — Monitor RSS feeds and surface new items matching keywords

Cron Jobs and Scheduled Tasks

OpenClaw includes a built-in cron scheduler. Tasks run inside agent sessions with full access to tools and skills.

Example cron jobs

# Daily morning briefing at 7:30 AM
openclaw cron add \
  --name "morning-briefing" \
  --schedule "30 7 * * *" \
  --agent personal-agent \
  --channel telegram \
  --prompt "Summarize unread important emails, list today's calendar events, and note GitHub notifications. Keep it concise."

# Weekly system health report every Monday at 9 AM
openclaw cron add \
  --name "weekly-health" \
  --schedule "0 9 * * 1" \
  --agent work-agent \
  --channel slack \
  --prompt "Run a system health check: disk, memory, failed services, top 3 CPU processes. Flag anomalies."

# Manage cron jobs
openclaw cron list
openclaw cron run morning-briefing    # Test immediately
openclaw cron remove morning-briefing

Timezone: Cron uses the system timezone. Confirm with timedatectl and set with sudo timedatectl set-timezone America/Chicago.

Webhook-Triggered Automations

Enable webhooks

openclaw webhooks enable
openclaw webhooks list

GitHub Actions Integration

Trigger an agent session when a CI build fails:

GitHub Actions workflow step

- name: Notify OpenClaw on failure
  if: failure()
  run: |
    curl -X POST https://agent.yourdomain.com/webhooks/github-ci \
      -H "Content-Type: application/json" \
      -H "X-OpenClaw-Token: ${{ secrets.OPENCLAW_WEBHOOK_TOKEN }}" \
      -d '{"event": "build_failed", "repo": "${{ github.repository }}"}'

Create webhook handler

openclaw webhooks create \
  --name "github-ci" \
  --agent work-agent \
  --channel slack \
  --prompt "A GitHub Actions build failed. Analyze the failure and suggest likely causes."

Gmail Pub/Sub for Real-Time Email

Deliver email events to the Gateway in real time rather than polling:

Gmail Pub/Sub setup

# Install and authenticate
openclaw skills install gmail
openclaw skills gmail auth

# Enable push notifications
openclaw skills gmail enable-push \
  --project-id your-gcp-project \
  --topic-name openclaw-gmail-events

Once connected, the agent receives a notification within seconds of a new email and can take action — summarize, categorize, draft a reply, or route based on sender rules.

Building Multi-Step Workflows

Skills can be chained. The agent handles chaining based on its system prompt and the task you give it. Example prompts:

Research-to-document: "Search for latest news about [topic], find the 3 most relevant articles using the browser skill, summarize each, and save to Notion."

Issue triage: "Check GitHub notifications for new issues, categorize each as bug/feature/question, draft a response, and show me the drafts before posting."

Server incident: "Check which Docker containers are running/stopped. Pull logs from stopped containers. Check disk and memory. Summarize the likely cause."

Monitoring Skill Activity

Track skill activity

# View recent skill invocations
openclaw skills log --tail 50

# View logs for a specific skill
openclaw skills log --skill gmail --tail 20

# View outbound network requests
openclaw skills log --outbound --tail 50

Automated skill audit

openclaw cron add \
  --name "skill-audit" \
  --schedule "0 */6 * * *" \
  --agent personal-agent \
  --channel telegram \
  --prompt "Review skill activity logs for the last 6 hours. Flag unexpected outbound requests, repeated errors, or unusual patterns. Only report if something looks wrong."

Part 4: LLM Config Part 6: Advanced Use Cases