Sentiment Analysis

Track what people say about anything — brands, crypto, products, competitors — across Twitter/X, Reddit, YouTube, Hacker News, and news sites.

One-shot analysis for quick checks. Scheduled monitoring for ongoing tracking. Multi-entity dashboards to compare multiple things at once.

Setup

On first use, read setup.md and follow its guidelines. Data is stored locally in ~/sentiment-analysis/.

When to Use

User wants to know public opinion about something. Could be:

• "What are people saying about [brand]?"
• "How's sentiment on [crypto] right now?"
• "Monitor [product] mentions and alert me on negative spikes"
• "Compare sentiment: [brand A] vs [brand B]"

Architecture

Data lives in ~/sentiment-analysis/. See memory-template.md for setup.

~/sentiment-analysis/
├── memory.md           # Config, entities, preferences
├── entities/           # One file per tracked entity
│   ├── brand-name.md
│   └── crypto-xyz.md
├── reports/            # Generated analysis reports
│   └── YYYY-MM-DD-entity.md
└── alerts.md           # Alert history

Quick Reference

| Topic | File |

|-------|------|

| Setup process | setup.md |

| Memory template | memory-template.md |

Core Rules

1. Source Diversity Matters

Never rely on a single platform. Each source has bias:

• Twitter/X: Real-time, emotional, viral content
• Reddit: Longer discussions, honest opinions, niche communities
• YouTube: Comments show product experiences
• Hacker News: Tech-focused, skeptical, early adopter views
• News sites: Official narratives, PR-filtered

Use at least 2-3 sources per analysis. Note source distribution in reports.

2. Time Windows Change Everything

Sentiment shifts fast. Always specify and report time window:

• Last 24h: Breaking news, viral events
• Last 7d: Weekly trends, sustained campaigns
• Last 30d: Product launches, seasonal patterns

Default: Last 7 days unless user specifies otherwise.

3. Quantify, Don't Guess

Every report includes concrete metrics:

📊 Entity: [Name]
🕐 Period: [Date range]
📈 Volume: [X mentions found]
😊 Positive: XX% | 😠 Negative: XX% | 😐 Neutral: XX%

Top Themes:
1. [Theme] — XX mentions, XX% negative
2. [Theme] — XX mentions, XX% positive

Notable Posts:
- [Quote] — [Platform, engagement]

4. Alerts Are Specific

Don't alert on every change. Track baselines and alert on:

• Negative spike >20% above baseline
• Viral negative post (>10x normal engagement)
• New negative theme appearing
• Competitor positive spike

5. Multi-Entity Comparison

When tracking multiple entities, always show relative performance:

📊 Sentiment Comparison (Last 7d)

| Entity | Volume | Positive | Negative | Trend |
|--------|--------|----------|----------|-------|
| Brand A | 1,240 | 62% | 18% | ↗️ +5% |
| Brand B | 890 | 45% | 32% | ↘️ -8% |

6. Scheduled Monitoring

For ongoing tracking, use cron. Default schedules:

• Critical entities: Daily at 09:00
• Regular entities: Every 3 days
• Background entities: Weekly

Store schedule in memory.md. Deliver reports to user's preferred channel.

7. Save Everything

After each analysis:

1. Update entity file with new data
2. Compare to previous analysis
3. Note trend changes
4. Archive raw findings

Common Traps

• Single-source analysis → Completely skewed view. Reddit hates everything, Twitter loves drama. Always cross-reference.
• No time window → "Sentiment is positive" means nothing without dates. A product can be loved one week, hated the next.
• Vanity metrics → High volume ≠ positive sentiment. 1000 mentions with 80% negative is worse than 100 mentions with 60% positive.
• Ignoring context → A spike in "crypto X is dead" might be sarcasm or memes. Read actual posts, not just keyword counts.
• Alert fatigue → Alerting on every fluctuation makes users ignore alerts. Only signal meaningful changes.

External Endpoints

| Endpoint | Data Sent | Purpose |

|----------|-----------|---------|

| Search engines (via web_search) | Query text | Find mentions |

| Social platforms (via web_fetch) | URL requests | Read content |

No API keys required. No data stored externally. All analysis happens locally.

Security & Privacy

Data that leaves your machine:

• Search queries sent to web search (query text only)
• URL requests to public posts (reading only)

Data that stays local:

• All entity tracking in ~/sentiment-analysis/
• Historical sentiment data
• Alert configurations

This skill does NOT:

• Require accounts on any platform
• Store data on external servers
• Send personal information anywhere
• Access private/protected content

Related Skills

Install with clawhub install if user confirms:

• analytics — web traffic and conversion data
• branding — brand strategy and guidelines
• monitor — system and service monitoring

Feedback

• If useful: clawhub star sentiment-tracker
• Stay updated: clawhub sync