微信扫一扫Workspace
~/follow/
├── sources/ # One file per followed entity
│ ├── people/ # @naval.md, @dhh.md
│ ├── topics/ # ai-safety.md, rust.md
│ └── feeds/ # techcrunch.md, hn-frontpage.md
├── archive/ # Captured content by date
│ └── YYYY-MM/
├── alerts.md # Alert configuration
└── index.md # Quick status: what's being followed
Quick Reference
| Task | Load |
|------|------|
| Add/configure sources | sources.md |
| Set up filtering rules | filtering.md |
| Configure alert tiers | alerts.md |
| Query archived content | querying.md |
| Platform-specific setup | platforms.md |
Core Loop
- Add source: User names person/topic/feed → create tracking file
- Monitor: Check sources on schedule (cron) or on-demand
- Filter: Apply relevance rules, skip noise
- Store: Archive what matters (summaries, not full dumps)
- Alert: Notify based on tier (immediate/daily/weekly/passive)
- Query: Answer "what did X say about Y?" from archive
Common Patterns
| User says | Agent does |
|-----------|------------|
| "Follow @naval on Twitter" | Create sources/people/naval.md, configure Twitter monitoring |
| "Track AI safety discussions" | Create topic tracker with keywords across multiple sources |
| "What has Competitor X posted this week?" | Query archive, synthesize summary |
| "Alert me immediately when Y happens" | Add to high-priority tier in alerts.md |
| "Give me a weekly digest of everything" | Configure weekly summary in alerts |
| "Stop following X" | Archive and mark inactive |
Capture Principles
- Summaries over full content — save space, stay legal
- Links + timestamps always — retrievable later
- Context for why it matters — not just "X posted"
- Deduplicate across sources — same news from 5 places = 1 entry