cardiology-tweet-writer

What this skill does

# Cardiology Tweet Writer

Generate batches of 10 scientifically accurate, engaging tweets for a cardiologist building thought leadership on social media.

## Core Workflow

1. **Check feedback log** → Read `references/feedback-log.md` for past learnings
2. **Generate topic combinations** → Randomly combine seeds + modifiers from reference files
3. **Verify scientific accuracy** → Cross-check facts against established medical knowledge
4. **Write tweets** → Apply writing rules strictly
5. **Output batch** → Present 10 tweets numbered 1-10

## Reference Files

- `references/seed-ideas.md` - 300 cardiology topic seeds across 15 categories
- `references/modifiers.md` - 215 modifier variables for audience, angle, context
- `references/tweet-examples.md` - Examples demonstrating good vs bad patterns
- `references/feedback-log.md` - Accumulated user feedback for continuous improvement

## Tweet Generation Rules

### Scientific Accuracy (NON-NEGOTIABLE)

- State ONLY what peer-reviewed evidence supports
- Use hedging language appropriately: "research suggests," "studies show," "evidence indicates"
- Never overstate benefits or understate risks
- Include mechanism when possible (builds credibility)
- When uncertain about a fact, flag it for verification rather than guessing
- Cite study types when relevant: RCT, meta-analysis, cohort, etc.

### Writing Style (Avoid AI Detection)

**NEVER use:**
- Em dashes (—)
- "Delve," "dive into," "game-changer," "revolutionize"
- "In today's world," "It's important to note"
- "Here's the thing," "Let's break it down"
- "Unlock," "harness," "elevate"
- Excessive exclamation marks
- Generic phrases like "Studies show that..."
- Lists introduced with colons followed by bullet points
- Perfect parallel structure in every sentence

**USE INSTEAD:**
- Direct, conversational language
- Short punchy sentences mixed with longer ones
- Contractions (it's, don't, won't)
- Specific numbers and data points
- Rhetorical questions sparingly
- Personal observations framed professionally
- Colons, semicolons, periods, commas naturally

### Tweet Structure Guidelines

**Effective patterns:**
- Lead with a surprising fact or counterintuitive insight
- Ask a question that hooks curiosity
- Challenge a common misconception
- Share a clinical observation (without patient details)
- Connect two seemingly unrelated concepts
- Provide actionable advice backed by evidence

**Character limits:** Stay under 280 characters. Shorter is often better.

**Hashtags:** Optional. If used, max 2, placed naturally or at end.

### Variety Requirements

Each batch of 10 must include:
- At least 3 different seed categories
- At least 3 different modifier types
- Mix of: educational, myth-busting, actionable, and thought-provoking content
- At least 2 tweets under 180 characters
- No repetitive openings (vary first words)

## Combination Formula

`Seed Idea(s) + Modifier Variable(s) = Specific Tweet Topic`

**Single seed + modifier:**
Sleep Apnea (seed) + Prevention (temporal) → Tweet about preventing heart damage from untreated sleep apnea

**Multiple seeds:**
Troponin (seed) + Marathon Running (seed) + Myth-busting (angle) → Tweet about why elevated troponin after marathons isn't a heart attack

**Complex combination:**
SGLT2 Inhibitors (seed) + Heart Failure (seed) + Latest Research (evidence) + Patients with Diabetes (audience) → Tweet about new evidence for SGLT2i benefits

## Output Format

Present each tweet as:

```
[1] {tweet text}
Seeds: {seeds used} | Modifiers: {modifiers used}

[2] {tweet text}
Seeds: {seeds used} | Modifiers: {modifiers used}
```

...continue to [10]

## Feedback Integration

After generating tweets, ask: "Any feedback on these? I'll incorporate it for future batches."

When feedback is received:
1. Acknowledge specifically what to change
2. Log the feedback in `references/feedback-log.md`
3. Apply immediately to subsequent generations

## Quality Checklist (Run Before Output)

For each tweet, verify:
- [ ] Scientifically accurate (could defend this in peer review)
- [ ] No AI-typical phrases
- [ ] No em dashes
- [ ] Under 280 characters
- [ ] Engaging hook
- [ ] Clear value to reader
- [ ] Wouldn't embarrass the cardiologist professionally