---
name: baoyu-x-post-writer
description: Write high-performing X (Twitter) posts and comments for Bo Wang (@bowang87). Covers technical AI/biology posts, viral comments, thread writing, and post rewrites. Use when Bo asks to "write an X post", "write a cool post", "write a viral comment", "rewrite this", or shares a paper/tweet URL and wants content drafted for X.
---

# X Post Writer — Bo Wang (@bowang87)

Bo is a computational biologist at Xaira Therapeutics, creator of scGPT. His X presence is technical, credible, and opinionated. Posts should read like a smart scientist with something to say — not a content marketer.

## Bo's Voice

- Direct, confident, occasionally dry
- Leads with evidence, lands on a sharp take
- Never hype-y; skepticism is a feature not a bug
- Strong POV on: Virtual Cell, AI drug discovery, perturbation biology, lab automation, AI safety
- Audience: AI researchers, computational biologists, biotech founders, ML engineers

## Core Algorithm Principles (from @wizofecom's guide)

X runs on an **interest graph**, not a social graph. Distribution is driven by keywords/topics, not follower count. This means:
- Use domain-specific terms (RLHF, MoE, cell-free synthesis, Tanimoto, perturbation) — they signal the right audience
- Profile clicks = strongest signal. Posts must make people want to see more from Bo
- Dwell time matters. Posts that reward reading get better distribution
- Self-RT at hour 6-8 triggers second distribution wave

**Positive signals:** profile clicks > RT/quotes/shares > dwell time > likes
**Negative signals:** mute, block, "not interested", low early engagement

## Post Structure (Snipe Framework)

```
[Hook] — tension, counterintuitive finding, or specific result
[Evidence] — data, quotes, specific numbers  
[Insight] — the non-obvious explanation
[Bo's take] — unique perspective only he can give → drives profile clicks
[Optional CTA or paper link]
```

## Hook Patterns (pick the sharpest one)

| Type | Example |
|------|---------|
| Specific result | "GPT-5 cut protein production costs by 40%." |
| Counterintuitive | "AI drug patents look identical to conventional ones." |
| Stakes/surprise | "The model cheated." |
| Volume of evidence | "36,000 experiments. 2 mistakes. Here's what went wrong." |
| The reveal | "OpenAI ran a secret AI scientist for 6 months in a real lab." |

**Rule:** First line = the most interesting fact or biggest tension. Never setup. Never context.

## Comment Structure

For commenting on someone else's tweet:
1. Add something the original didn't say (mechanism, counterexample, deeper implication)
2. Connect to Bo's domain when natural (data quality, causal biology, perturbation diversity)
3. End with a sharp, memorable line

## Formats

**Single post** — default. Max ~280 chars per tweet but X allows long-form. Use short punchy paragraphs.

**Thread** — when content has 3+ distinct insights. Number tweets (1/, 2/, etc.). Each tweet must stand alone.

**Comment** — 2-4 short paragraphs. Adds dimension, doesn't just agree.

## What to Avoid

- Opening with "Researchers found that..." — buried lede
- Ending with generic conclusions ("The future of X is here")
- Bullet lists as the main format — prose with line breaks reads better
- Hedging stacks ("could potentially possibly")
- Hollow amplifiers ("groundbreaking", "pivotal", "fascinating")
- Bullet headers ("Performance: Performance improved...") — use prose

## Workflow

1. **Fetch the source** — tweet via FxTwitter API (`api.fxtwitter.com/<user>/status/<id>`), paper via arXiv PDF or Camofox
2. **Extract the key finding** — what's the single most surprising or important result?
3. **Find Bo's angle** — how does this connect to his work/worldview?
4. **Draft with hook first** — write the opening line last (after you know the whole story)
5. **Offer 2-3 variants** when tone is ambiguous (sharp vs. softer, thread vs. single)

See `references/examples.md` for annotated good/bad examples and past posts that performed well.