SupportBot — OpenClaw 24/7 Customer Support Agent

SupportBot is a fully autonomous OpenClaw customer support agent that handles inquiries around the clock, resolves common issues using your knowledge base, and escalates complex or sensitive cases to your human team with full context — all while speaking in your brand voice.

It integrates natively with Zendesk, Freshdesk, Linear, Slack, Intercom, and email. It monitors open ticket queues proactively, flags stale escalations, tracks active incidents from your status page, and sends your team a daily briefing every morning. It is the agent that works while your team sleeps.

---

What It Does

Resolves autonomously:

• Guides customers through password resets, account lockouts, 2FA issues
• Answers billing questions, explains refund policy, logs refund requests under your threshold
• Walks customers through technical troubleshooting using your KB articles
• Identifies active incidents and proactively informs affected customers

Escalates intelligently:

• Detects legal threats, safety concerns, media threats, fraud, and churn risk before responding
• Routes each escalation to the right team (billing, technical, legal, VIP) via both ticket and Slack alert
• Includes full conversation context in every ticket — no human has to ask "what happened?"
• Sends structured Block Kit Slack alerts with ticket links, customer details, and priority color coding

Monitors proactively:

• Every 5 minutes: checks for customers waiting too long without a reply
• Every 15 minutes: audits open ticket queue for stale urgent/high-priority tickets
• Every 30 minutes: fetches your status page and updates the known-issues KB automatically
• Every morning: sends a full overnight summary to your Slack

Maintains brand voice:

• BRAND.md defines every aspect of how it communicates — prohibited phrases, required templates, per-channel formatting, tone-by-situation
• The same warmth and consistency at 3am on a Sunday as at 10am on a Monday

---

File Overview

support-agent/
├── AGENTS.md             ← Decision pipeline: classify → escalation check → KB → resolve or escalate
├── SOUL.md               ← Empathy model: how to handle frustration, anger, confusion, gratitude
├── BRAND.md              ← ★ Voice, prohibited phrases, message templates, channel formatting
├── ESCALATION_RULES.md   ← ★ Every trigger, SLA, team route, on-call config — edit these
├── KB.md                 ← Knowledge base index mapping topics to kb/ files
├── HEARTBEAT.md          ← Proactive monitoring schedule (5-min to daily)
├── MEMORY.md             ← Customer profiles, known issues, active escalations
├── TOOLS.md              ← CLI reference for every integration
├── IDENTITY.md           ← Agent name, persona, AI disclosure policy
├── memory/               ← Daily interaction logs (auto-created)
├── escalation-queue/     ← Date-stamped JSON escalation records
├── kb/                   ← Knowledge base articles (one file per topic)
│   ├── account-access.md
│   ├── billing-payments.md
│   └── known-issues.md
└── skills/support-hub/
    └── scripts/
        ├── create-ticket.js    ← Unified ticket creation: Zendesk, Freshdesk, Linear
        ├── update-ticket.js    ← Update status, add internal notes or customer replies
        ├── list-tickets.js     ← Query open/urgent tickets for heartbeat monitoring
        ├── slack-alert.js      ← Priority-aware Block Kit alerts with @mention support
        ├── intercom-reply.js   ← Reply to or assign Intercom conversations
        ├── send-email.js       ← Gmail App Password or SMTP, no external packages
        ├── check-status.js     ← Fetch active incidents from Statuspage.io / Instatus
        └── log-interaction.js  ← Write to daily log, escalation queue, and MEMORY.md

---

Key Design Decisions

Escalation triggers check before every reply. ESCALATION_RULES.md defines two tiers: immediate_escalation (legal, safety, media, VIP) fires before SupportBot sends any response; conditional_escalation (churn risk, billing disputes, repeated failure) fires after the initial acknowledgment. This means no legal threat ever accidentally gets a chatty bot response.

The knowledge base is the boundary. SupportBot is explicitly instructed: if the answer isn't in a kb/*.md file, escalate rather than guess. This keeps it from manufacturing policy explanations or technical advice that might be wrong or contradict what your team says.

Brand voice is non-negotiable. BRAND.md includes a prohibited phrases list, required phrase templates (exact wording for escalations, apologies, AI disclosure), per-channel formatting rules, and tone-by-situation guidance. The bot reads this every session before responding to anything.

Escalation includes full context. Every ticket created by SupportBot includes the full conversation history, the classification, the trigger that fired, and which KB files were consulted. Human agents don't need to ask what happened — it's all in the ticket.

AI disclosure is hardcoded. IDENTITY.md contains a non-negotiable rule: if a customer sincerely asks "Are you a bot?", SupportBot must answer honestly using the exact template from BRAND.md. There is no configuration option to instruct it to claim to be human.

Zero external dependencies. All eight Node.js scripts use only the standard library. Drop the files anywhere Node 18+ is installed and they work. No npm install, no package.json, no version management headaches.

---

Configurable Parameters

In ESCALATION_RULES.md:

---

Integrations

Slack is the only required integration. All others are optional — unused ones are skipped gracefully.