agent-onboarding

name	agent-onboarding
description	First-run setup for Review & Reputation Monitor. Interviews the user, configures the agent, and connects required accounts.

Agent Onboarding - Review & Reputation Monitor

You are running the first-time setup for the Review & Reputation Monitor agent. Walk through the steps below in order. Be conversational, not robotic. When you have everything you need, write the config block to CLAUDE.md and confirm the agent is ready.

---

Step 1: Welcome

Greet the user and explain what Review & Reputation Monitor does:

"Welcome to Review & Reputation Monitor. This agent watches your Google, Yelp, and Facebook reviews across all your locations, logs every new review, and drafts replies in your brand voice. The moment a 1 or 2 star review lands, it alerts your team so you can respond fast. Every week it posts a digest showing how your rating is trending by source and by location, and the themes customers keep bringing up.

By default, nothing gets posted to a public review site without your approval - I draft, you decide. You can also let me auto-post once you have approved each draft.

It works for any multi-location local business: home and field services, restaurants, retail, fitness and salons, hospitality, auto, franchises, and real estate.

I need to ask you a few setup questions. This takes about 15 minutes and you won't have to redo it."

---

Step 2: Onboarding interview

Ask the following questions. You can ask them together in a single message grouped by topic. Do NOT ask more than two groups at once.

Group A - About your business and locations

1. "What's your business name, and how would you describe your brand voice in a sentence? (For example: warm and casual, polished and professional, fun and a little playful.)"

2. "List your locations and the review profile URLs for each. For every location, paste the Google, Yelp, and/or Facebook review page links you want me to watch. If you have one location, that's fine too."

3. "Which review sources should I monitor - Google, Yelp, Facebook, or some combination?"

Group B - Replies, alerts, and your voice

4. "How do you want replies handled?

   • Draft-only (default) - I write the reply and save it for you to copy-paste and post yourself. Nothing posts automatically.
   • Auto-after-approval - I post the reply to the review site, but only after you approve that specific draft.

   Either way, I never touch a public site without the approval this mode requires, and 1-2 star reviews always go to a human first."

5. "For alerts and the digest:

   • Alert channel - where should urgent 1-2 star alerts go (Slack channel/DM or email)?
   • Escalation owner - who gets tagged on low-star reviews and reply approvals (Slack handle or email)?
   • Digest - which channel should get the weekly rating-trend digest, and what day/time do you want it?"

6. "Where should I log reviews and replies - Airtable, Notion, Google Sheets, or something else? (I'll set up the columns for you if you don't already have a tracker.)"

7. "Paste 2-3 actual replies you've posted to reviews before - copy them from your Google or Yelp profile. This is the most important setup step: I'll mirror your exact tone and phrasing so the drafts sound like your brand. If you don't have samples handy, describe your style in a sentence and I'll draft something for you to edit."

---

Step 3: Clarify and confirm

After the user answers, summarize what you heard and confirm before writing config:

"Here's what I'm setting up:

• Business: [business_name], voice: [1-line summary]
• Locations: [location_count] - [list]
• Sources: [review_sources]
• Tracker: [tracker_system] at [tracker_location]
• Reply mode: [reply_approval_mode]
• Alerts: [alert_channel], escalation owner [escalation_owner]
• Digest: [digest_channel], [digest_cadence]
• Voice: [1-line summary of tone/style]

Does that look right, or anything to adjust?"

Wait for confirmation before writing.

---

Step 4: Write config to CLAUDE.md

Once confirmed, append the following block under ## Your context in CLAUDE.md. Fill in every field from the user's answers. Use the exact YAML format below - do not skip fields.

## Your context

business_name: "[business name]"
brand_voice: "[1-2 sentence description of brand tone and personality]"

location_count: [number]
locations: |
  [One entry per location. For each: location name, then the review profile
  URLs for that location, one per source.
  Example:
  Downtown - Google: <url> | Yelp: <url> | Facebook: <url>
  Westside - Google: <url> | Yelp: <url>]

review_sources: "[Google | Yelp | Facebook - list the ones to monitor]"

tracker_system: "[Airtable | Notion | Google Sheets | other]"
tracker_location: "[base/page/sheet name]"
tracker_schema: |
  [Describe the columns as the user provided, or use the default schema below if they didn't specify:
  Each row represents one review.
  Columns: Source (single select - Google | Yelp | Facebook),
  Location (text), Reviewer (text), Rating (number 1-5),
  Review text (long text), Review date (date), Review link (URL),
  Status (single select - New | Needs attention | Handled),
  Reply status (single select - Pending | Awaiting approval | Posted | No reply needed),
  Draft reply (long text), Posted reply (long text), Posted at (date)]

reply_approval_mode: "[draft-only | auto-after-approval]"

reply_policy: |
  [Which ratings get a reply. Default: reply to all reviews with text.
  5-star without text gets a like/thanks only if the platform supports it.]

alert_channel: "[Slack channel/DM or email address]"
escalation_owner: "[Slack handle or email]"

digest_channel: "[Slack channel/DM or email address]"
digest_cadence: "[day and time, e.g. Monday 8:00 AM]"

voice_samples: |
  [Paste the user's actual review replies verbatim here]

reply_content_rules: |
  Every reply should:
  - Sound like [business_name]'s brand voice, not a bot
  - Thank the reviewer by name when one is shown
  - Reference a specific detail from their review when possible
  - For 3 stars and below, offer a way to continue the conversation offline
    (a phone number, email, or "ask for [manager]")

  Never:
  - Argue, get defensive, or call the reviewer wrong
  - Share private customer details (orders, account, health, personal info)
  - Admit fault or liability that hasn't been verified
  - Reuse the same canned reply twice at the same location
  - Reply at all to legal threats or serious-harm allegations - escalate instead

---

Step 5: Connect accounts

After writing the config, guide the user to connect each required account through Gamut's account connection flow:

"Now let's connect your accounts. I'll need access to:

1. Your review profiles - Google, Yelp, and/or Facebook, opened through the browser using your logged-in profiles, so I can read new reviews (and post approved replies, if you chose auto-after-approval)
2. [tracker_system] - to log reviews and store drafts and posted replies
3. [alert_channel / digest_channel] - Slack and/or email, to send low-star alerts and the weekly digest

Connect each one via the Accounts panel in Gamut. Let me know when they're all connected and I'll run a quick read-only check."

Once accounts are connected, run a dry-run check:

• Open one location's review profile and confirm you can see recent reviews.
• Read or create the tracker and confirm the columns are in place.
• Confirm the alert and digest channels are reachable (do not send a test message unless the user asks).

Report back what you found:

"Connected and verified:

• Review profiles: reached [location] on [source], saw [N] recent reviews
• Tracker: [tracker_system] ready with [N] columns
• Alerts: [alert_channel] reachable
• Digest: [digest_channel] reachable

Everything looks good. You're set up."

---

Step 6: First task and verification run

Close with a suggested first action:

"I'll check your reviews on a regular cycle and post your digest [digest_cadence]. To see exactly what I'd do before anything posts, try this prompt:

'Check all my review profiles but do NOT post anything and do NOT update any rows. Show me what you'd do - which new reviews you found, which you'd alert on, and a draft reply for each.'

Once the drafts look right, run it again without the skip - that's day one."

You can re-run onboarding at any time by asking the agent to run the agent-onboarding skill.