# Punchy Insight Refiner

> Use when prose is correct but dull, long, or AI-flat and you need short human lines that keep the claim. Rewrites for punch without hype, without inventing facts.

- Skill: `cassidyrice/punchy-insight-refiner` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add cassidyrice/punchy-insight-refiner`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cassidyrice/punchy-insight-refiner/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: cassidyrice (https://skillmd.com/u/cassidyrice)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/cassidyrice/punchy-insight-refiner

---


# Punchy Insight Refiner

Turn correct-but-dead prose into lines a human will finish.

## When to Use

- Newsletter, Substack, or skill intro drafts sound like a model wrote them
- The claim is right; the sentences are long, soft, or hedged to death
- You need a hook title + tight body without adding new facts
- An agent produced a summary and you want the publishable cut

**Do not use** to invent evidence, soften a hard claim into mush, or expand short text into essays.

**Not for LinkedIn algorithm / forensic de-AI audit** → `linkedin-humanizer` (tiered scrub, detector spread, profile voice).  
**Not for full LinkedIn post drafting from a topic** → `linkedin-post-writer` / `linkedin-marketing`.  
This skill = **claim-preserving short-form** for any channel (newsletter, skill intro, CTA).

## Contract

| In | Out |
|----|-----|
| Draft prose + must-keep claims | Punchy rewrite |
| Optional audience + max words | Title options (2–3) |
| | Claim checklist (what stayed true) |

## Inputs

1. **Draft** — the raw text  
2. **Must-keep claims** — bullets that cannot change meaning  
3. **Constraints** — max words, tone (direct / warm / dry), forbidden words  

## Process

1. **Extract claims**  
   List every factual or instructional claim. If the user supplied must-keeps, those win.

2. **Kill the sludge**  
   Cut: “In today’s world”, “It’s important to note”, “delve”, stacked adverbs, fake wonder, throat-clearing intros.

3. **One idea per sentence**  
   Split compound sentences. Prefer subject–verb–object. Put the cost or surprise early.

4. **Hook first**  
   Open with tension, a concrete failure, or a sharp rule—not with the topic name.

5. **Rewrite short**  
   Target 30–40% fewer words unless constrained otherwise. Keep verbs active.

6. **Claim check**  
   For each must-keep claim: still true? still present? If a line sounds better but changes meaning, revert that line.

7. **Title pass**  
   Offer 2–3 titles: specific > clever. No clickbait that the body doesn’t pay off.

8. **Return package**
   ```text
   TITLE OPTIONS
   ...
   REFINED
   ...
   CLAIMS PRESERVED
   - ...
   CUTS (optional)
   - removed X because ...
   ```

## Invocation (agent)

```text
load_skill("punchy-insight-refiner")
# draft: <text>
# must_keep: [claim1, claim2]
# max_words: 400
```

## Verification

- Every must-keep claim appears in refined output (paraphrase OK, meaning not OK to drop)  
- No new numbers, names, or guarantees introduced  
- First two sentences work as a standalone hook  
- Word count under constraint  

## NEVER

- Add facts “for color”  
- Replace a precise claim with a vague vibe  
- Use hype stacks (“revolutionary”, “game-changing”, “unleash”)  
- Lengthen the draft  
- Open with “In this article/post/skill”  
- Run LinkedIn 2026 algorithm audits or multi-detector AI-score loops (use `linkedin-humanizer`)  
- Build a voice/brand profile from past posts (use `linkedin-humanizer --mode profile`)  

## Related skills

| Need | Skill |
|------|--------|
| LinkedIn de-AI + audit + voice profile | `linkedin-humanizer` |
| Native LinkedIn post from a theme | `linkedin-post-writer` |
| Full series episode (skill+video+newsletter) | `ai-skills-episode-factory` | 

## Minimal Example

**Draft:**  
It is important to note that agents often lose valuable work when a session ends, and packaging that work into a skill can help future runs.

**Must-keep:** Session-end loses work; packaging into a skill helps future runs.

**Refined:**  
Your agent solved it. Then the session ended—and the solution died with it.  
Package the path into a skill, or solve it from zero again tomorrow.

## Why it compounds

Publishable voice is a skill. Agents that refine before they ship stop flooding you with correct sludge.