Twitter: Keyword Sentiment Alert

Created by

B3 Team

Root

Triage

Check

Format Sentiment Alert

Send Alert

About

Monitors Twitter/X for tweets matching your configured keywords, then uses AI to filter each tweet by sentiment. Only alerts you when the tweet contains something worth acting on — bug reports, UX complaints, service issues, or user frustration — and silently drops everything else (price talk, hype, casual mentions).

Quick Setup

1. Select your notification channel.

2. Adjust the polling interval if needed (default: 5 minutes).

3. Test and publish.

The AI classifies each tweet into one of seven categories (bug report, UX complaint, service issue, data error, security concern, user frustration, support question) and assigns a severity level (critical / high / medium / low). Non-actionable tweets are never forwarded.

Preview

check

Action

Condition:{"$and":[{"{{triage.result.output.alert}}":{"$eq":true}}]}

Format Sentiment Alert

Action

Code:function transform(input) { var t = input.tweet || {}; var ai = input.ai || {}; var catMap = {'bug_report':'Bug Report','ux_complaint':'UX Complaint','service_issue':'Service Issue','data_error':'Data Error','security_concern':'Security Concern','frustration':'User Frustration','support_question':'Support Question','not_actionable':'Not Actionable'}; var sevMap = {'critical':'[CRITICAL]','high':'[HIGH]','medium':'[MEDIUM]','low':'[LOW]'}; var cat = catMap[ai.category] || (ai.category || 'Unknown'); var sev = sevMap[ai.severity] || '[?]'; function fmtN(n) { if (!n || n < 1) return '0'; return n >= 1000000 ? (n/1000000).toFixed(1).replace(/\.0$/,'')+'M' : n >= 1000 ? (n/1000).toFixed(1).replace(/\.0$/,'')+'K' : String(n); } var verified = t.authorIsBlueVerified ? ' \u2713' : ''; var lines = []; lines.push(sev + ' ' + cat); lines.push(''); lines.push(ai.summary || ''); lines.push(''); lines.push('<b>@' + (t.authorUsername || 'unknown') + verified + '</b> \u00b7 ' + fmtN(t.authorFollowers || 0) + ' followers'); lines.push(''); lines.push(t.text || ''); var m = []; if (t.replyCount > 0) m.push(fmtN(t.replyCount) + ' replies'); if (t.retweetCount > 0) m.push(fmtN(t.retweetCount) + ' RTs'); if (t.likeCount > 0) m.push(fmtN(t.likeCount) + ' likes'); if (t.viewCount > 0) m.push(fmtN(t.viewCount) + ' views'); if (m.length > 0) { lines.push(''); lines.push(m.join(' \u00b7 ')); } lines.push(''); lines.push('<i>AI: ' + (ai.reasoning || '') + '</i>'); if (t.tweetUrl) { lines.push(''); lines.push('<a href="' + t.tweetUrl + '">View on X</a>'); } return lines.join('\n'); }

Input:{"ai":"{{triage.result.output}}","tweet":"{{root.result}}"}

root

Action

Interval Seconds:Prop: interval_seconds

Keywords:Prop: keywords

send_alert

Message configured

triage

Action

Json Schema:{"type":"object","properties":{"alert":{"type":"boolean"},"category":{"type":"string","enum":["bug_report","ux_complaint","service_issue","data_error","security_concern","frustration","support_question","not_actionable"]},"severity":{"type":"string","enum":["critical","high","medium","low"]},"summary":{"type":"string"},"reasoning":{"type":"string"}},"required":["alert","category","severity","summary","reasoning"],"additionalProperties":false}

Max Tokens:512

Model:openai/gpt-4o-mini

Prompt:root:text

Response Format:json_schema

System Prompt:You are a sentiment analysis agent monitoring tweets that match these keywords: Prop: keywords. Your job is to decide whether each tweet is worth flagging for the project team. Set alert=true ONLY for tweets about: - Bug reports: errors, crashes, failed transactions, features not working - UX complaints: confusing UI, bad user experience, feature behaving unexpectedly - Service issues: downtime, slowness, missing or wrong data, incorrect balances - Data accuracy: wrong prices, stale data, incorrect on-chain values - Security or funds: any concern about lost funds, unauthorized access, exploits - Strong user frustration directed at the product itself (not price) - Support questions: someone asking how to use a feature, where to find something, or how to get started Set alert=false for: price talk, speculation, hype, bullish sentiment, casual mentions, news articles, praise, congratulations, or any tweet where the keywords appear in passing without engagement with the product. For category use one of: bug_report, ux_complaint, service_issue, data_error, security_concern, frustration, support_question, not_actionable. For severity use one of: critical, high, medium, low. Return JSON with: alert (boolean), category, severity, summary (one sentence describing the specific issue, or Not actionable if alert=false), reasoning (one sentence explaining your call).

Temperature:0.1

Loading chat...