Review & Reputation Monitoring with OpenClaw

# AGENTS.md — review-monitor agent

agents:
  review-monitor:
    description: "Fetch new reviews from all platforms and alert on negatives"
    tools:
      - trustpilot-api
      - google-mybusiness-api
      # fallback: http-crawler if API not available
    credentials:
      trustpilot_api_key: "${TRUSTPILOT_API_KEY}"
      trustpilot_business_unit_id: "${TRUSTPILOT_BUSINESS_UNIT_ID}"
      google_oauth_token: "${GOOGLE_OAUTH_TOKEN}"
    config:
      alert_on_stars: [1, 2]      # negative = 1 or 2 star reviews
      lookback_hours: 24          # check for reviews posted in last 24 hours
      state_file: "reviews_seen.json"  # track already-processed reviews
      deduplication: true         # skip reviews already in state_file
      platforms: [trustpilot, google, amazon, etsy]
    output:
      format: markdown
      sections:
        - new_negative_reviews    # 1-2 star, highest priority
        - new_positive_reviews    # 4-5 star, summarised count
        - overall_stats           # total count, avg rating
      sort_by: stars              # negatives first
      include_fields:
        - platform
        - reviewer_name
        - rating_stars
        - review_text
        - review_date
        - response_status         # has merchant responded?

# AGENTS.md — sentiment-tracker agent

agents:
  sentiment-tracker:
    description: "Calculate rolling sentiment and alert on significant drops"
    tools:
      - trustpilot-api
      - google-mybusiness-api
    config:
      rolling_window_days: 30
      metrics:
        - avg_rating                    # average star rating
        - review_count                  # number of reviews posted
        - sentiment_score               # POSITIVE / NEUTRAL / NEGATIVE classification
      classification: text_based        # classify by review text, not just stars
      alert_thresholds:
        rating_drop_week_over_week: 0.3 # alert if avg drops >0.3 points vs last week
        sentiment_shift: true            # alert if POSITIVE → NEUTRAL shift
      platforms: [trustpilot, google]
    prompt: |
      For each review, classify the text as POSITIVE, NEUTRAL, or NEGATIVE based on language.
      A 3-star review with positive language ("Great product, slow shipping") counts as POSITIVE.
      A 5-star review with concerns ("Works fine but wish it had X") counts as NEUTRAL.
      This text-based classification is separate from the star rating.

# AGENTS.md — response-drafter agent

agents:
  response-drafter:
    description: "Draft personalised responses to flagged reviews"
    tools:
      - trustpilot-api
      - google-mybusiness-api
    config:
      trigger: negative_review         # draft on reviews with 1-2 stars
      complaint_categories:
        - shipping_delay
        - product_quality
        - wrong_item
        - customer_service
        - packaging
        - missing_item
      state_file: "responses_drafted.json"
    prompt: |
      For each negative review, perform these steps:
      1. Classify the complaint into one of the categories above
      2. Acknowledge the specific issue mentioned in the review
      3. Apologise sincerely (tone: warm, professional, never defensive)
      4. Offer a concrete resolution:
         - For shipping delays: offer a discount on next order
         - For product quality: offer a replacement or refund
         - For wrong/missing item: offer replacement with priority shipping
         - For customer service: acknowledge and take responsibility
      5. Invite them to contact support directly for faster resolution
      6. End with gratitude for their feedback

      Output format for each response:
      ─────────────────────────────────────
      REVIEW ID: [id]
      PLATFORM: [platform]
      COMPLAINT TYPE: [category]
      REVIEWER: [name]
      RATING: [stars]
      ─────────────────────────────────────
      DRAFT RESPONSE:
      [your response here]
      ─────────────────────────────────────

      IMPORTANT: Add a note at the top:
      "⚠️ DRAFT — review before posting. Do not post automatically."

# HEARTBEAT.md — daily review digest

schedule:
  - name: "Daily Review Digest"
    cron: "0 7 * * *"             # every day at 7 AM
    agents:
      - review-monitor
    output:
      file: "daily_review_digest.md"
      notify: true
    prompt: |
      Summarise all reviews from the last 24 hours across all platforms.
      Format as:
      1. URGENT NEGATIVES (1-2 star reviews) — if any exist, list each with platform,
         reviewer name, and first 100 chars of review text
      2. NEW POSITIVES — summary line: "X new 5-star, Y new 4-star reviews"
      3. SUMMARY — "Total: X reviews, average rating: 4.5 across all platforms"
      If no new reviews: "No new reviews in the last 24 hours."

# HEARTBEAT.md — instant negative alert

schedule:
  - name: "Negative Review Alert"
    trigger: on_new_review
    filter: stars <= 2
    agents:
      - review-monitor
    output:
      notify: true
      notify_channels: [email, slack]
    prompt: |
      A negative review has just been posted. Send an immediate alert with:
      - Platform and reviewer name
      - Full review text
      - Current store rating (for context)
      - A link to the review (so the user can respond directly)

# HEARTBEAT.md — weekly sentiment report

schedule:
  - name: "Weekly Sentiment Report"
    cron: "0 8 * * 1"             # Mondays at 8 AM
    agents:
      - sentiment-tracker
    output:
      file: "weekly_sentiment_report.md"
      notify: true
    prompt: |
      Generate a 30-day rolling sentiment report with:
      1. Average rating (30-day window): [X.X stars]
      2. Week-over-week change: +/- [X.X] points
      3. Sentiment breakdown: X POSITIVE, Y NEUTRAL, Z NEGATIVE (% each)
      4. Top complaint category if sentiment is down
      5. Action suggestion if rating dropped >0.3 points vs last week