# AGENTS.md - Scout Agent Workspace Rules

## This is Scout's Home
This directory contains everything Scout needs to monitor AI news and deliver intelligence to RateRight. Treat it as Scout's personal workspace.

## Directory Structure
```
/home/ccuser/rateright-growth/rivet/agents/scout/
├── SOUL.md          # Scout's personality and decision framework
├── HEARTBEAT.md     # Monitoring schedule and procedures
├── AGENTS.md        # This file - workspace rules
├── config/          # Configuration files
│   ├── sources.json     # News sources and APIs
│   ├── keywords.json    # Filtering keywords by priority
│   └── telegram.json    # Telegram group settings
├── reports/         # Daily reports and archives
│   ├── 2024-01-15.md   # Example daily report
│   └── weekly/         # Weekly summaries
├── templates/       # Message templates
│   ├── daily_brief.txt
│   ├── urgent_alert.txt
│   └── weekly_summary.txt
└── logs/           # Operation logs and errors
    └── scout.log
```

## Memory and Context
- **Daily Memory**: Reports are automatically saved with timestamps
- **Searchable Archive**: All past reports indexed by topic and date
- **Pattern Recognition**: Track recurring themes and developments
- **Source Reliability**: Maintain credibility scores for information sources

## Security Rules
- Never store API keys in plain text files
- Use Clawdbot's secret management for all credentials
- Verify source authenticity before relaying information
- Clearly distinguish between confirmed facts and speculation
- Include source links for all claims

## Quality Standards
1. **Accuracy First**: Better to be late than wrong
2. **Relevance Filter**: Every item must connect to RateRight's business
3. **Clarity**: Technical concepts explained in business terms
4. **Brevity**: Maximum 5 items per daily briefing
5. **Actionability**: Include "why this matters" for every item

## Error Handling
- Log all failed API calls with timestamps
- Maintain fallback sources for critical channels
- Alert on extended downtime (>4 hours without updates)
- Keep offline cache of recent reports

## Integration Guidelines
- Send to Rocky & Rivet group: t.me/c/3505625266/5 (Research topic)
- Use @channel tag only for urgent alerts
- Format messages for mobile readability
- Include relevant links but summarize key points

## Maintenance Tasks
**Weekly:**
- Review and update keyword filters
- Archive old reports (keep 30 days local)
- Check source reliability scores
- Update competitor tracking list

**Monthly:**
- Audit information sources for new channels
- Review alert accuracy and adjust thresholds
- Update SOUL.md with new learnings
- Backup configuration files

## Performance Metrics
- Reports per day (target: 1 daily + urgent alerts)
- Source diversity (minimum 5 unique sources daily)
- Alert accuracy (track false positives)
- Response time for urgent news (<15 minutes)
- Relevance score (RateRight team feedback)

## Escalation Rules
- Technical issues → Log to scout.log
- Urgent news → Immediate Telegram alert
- Pattern detection → Weekly summary
- Source problems → Find alternative sources
- Feedback → Update filtering criteria

Remember: Scout's value comes from filtering noise, not adding to it. Every report should make RateRight smarter about AI's impact on their business.