/hook-lab

---
name: hook-lab
description: >-
  Write and test the first lines of a LinkedIn post (the part people see
  before "...see more"): generates ten hooks across proven shapes, scores
  each against a checklist, keeps the three strongest, and says which one
  to post and how to test the others. Use this skill whenever the user asks
  for "a hook", "first line", "opening line", "headline for my post", "why
  is nobody reading my post", "make this grab attention", or shares a post
  draft and wants more reach. ALSO use it when drafting any LinkedIn post
  from scratch. Do NOT use for clickbait that the post does not pay off,
  for fake statistics or invented stories, or for paid ad copy with
  platform-specific rules.
---

# Hook lab

On LinkedIn, the first two or three lines decide whether anyone reads the
rest. On a phone that is about 140 characters before "...see more". A post
with a brilliant middle and a vague opening gets the same reach as a bad
post, because almost nobody reaches the middle.

Strong hooks are not tricks. They are the most interesting true thing in
the post, moved to the top and said plainly. Most weak hooks fail the same
way: they announce a topic ("Let's talk about AI in finance") instead of
stating a fact, a tension or a result. This skill writes many options fast,
then judges them coldly, because the writer is the worst judge of their own
favourite line.

## Workflow

### 1. Find the raw material

From the draft or the user's idea, list:
- the single most surprising fact or number,
- the concrete actor (a person, company, role) in the story,
- the tension (belief vs reality, before vs after, effort vs result),
- the payoff the reader gets by the end.

If there is no surprising fact, say so: the post needs one before it needs
a hook.

### 2. Write ten hooks across these shapes

The examples use made-up numbers to show the shape. Yours must be real.

1. **Actor did X:** "A CFO let AI write the board pack. Then the auditor called."
2. **Number first:** "37 hours. That's what one report cost our team last quarter."
3. **Myth / fact:** "Myth: AI will replace analysts. What I see: analysts who check AI replace those who don't."
4. **Two numbers that clash:** "71% feel confident with AI. 33% passed the test."
5. **Before / after:** "Monday: 3 hours on the sales report. Friday: 20 minutes."
6. **Confession:** "I made a €4,000 mistake with an AI spreadsheet."
7. **Plain explainer:** "A skill is a set of instructions Claude reads before it starts."
8. **This week:** "Three banks changed their AI rules this week."
9. **List promise with a number:** "5 checks I run before any AI number leaves my desk:"
10. **Question with stakes** (use sparingly): "Would you sign a report you didn't check?"

Each hook: at most three short lines, no blank lines between them, under
about 140 characters before the cut-off.

### 3. Score each one (0-2 per item, max 10)

- **Specific:** names a number, actor or moment.
- **True and paid off:** the post delivers exactly what the hook promises.
- **Tension:** creates a gap the reader wants closed.
- **Plain:** a 12-year-old understands it on the first read.
- **Fits the reader:** speaks to the people the user wants to reach.

Cut anything under 6. Never keep a hook that fails "true and paid off".

### 4. Deliver

```
## Best hook (score 9/10)
[hook]
Why: [one line]

## Two to test next
[hook] (8/10), [hook] (7/10)

## Test plan
Post the best one now. Use the others on the next two posts of the same
type, then compare impressions after 72 hours against your recent average.
```

## Rules

**Never invent the hook's fact.** If the best hook needs a number the user
does not have, ask for it or drop the hook.

**No bait.** "You won't believe..." and "Nobody talks about this" are
banned. They trade one click for long-term trust.

**One idea per hook.** If a hook needs a comma and an "and", it is two hooks.

**Data decides.** Your score is a guess. The user's own results after 72
hours beat any checklist. Ask them to log which hook shapes win.

## Example

Draft opens: "Today I want to share some thoughts about using AI for
financial reporting." Score 2/10: no fact, no actor, no tension. The post
mentions that an AI summary got Q3 revenue wrong by €40k and the user
caught it with one formula. Best hook: "AI got our Q3 revenue wrong by
€40,000. One formula caught it." (9/10). Same post, new first line.