caveman

name	caveman
description	Ultra-compressed content writing style reducing word count ~75% while keeping technical accuracy exact. Use when writing cavemanBody fields in Sanity blog posts and documentation pages. Drops articles, filler, hedging. Fragments OK. Pattern: [thing] [action] [reason]. [next step].
triggers	["caveman mode","caveman version","write caveman","cavemanBody","less words","/caveman","write blog post","write doc","write documentation"]

Caveman Skill

Write compressed content. Less word. More understand.

Core Rules

Drop all of these:

• Articles: a / an / the
• Filler: just, really, basically, actually, simply, essentially, generally
• Hedging: might, could, perhaps, it seems, it appears, arguably
• Pleasantries: note that, keep in mind, it's worth mentioning, feel free to
• Weak openers: "In this post we'll explore...", "Welcome to...", "This guide covers..."

Keep all of these:

• Technical terms (exact, unchanged)
• Code blocks (verbatim, never compress code)
• Numbers and data
• Product/feature names

Pattern

[thing] [action] [reason]. [next step].

Bad: "In order to get started with the screener, you'll first need to make sure that you have created an account and logged in." Good: "Sign in first. Screener needs auth."

Bad: "The news impact score is calculated based on several factors including sentiment, volume, and relevance to the asset." Good: "News impact score = sentiment + volume + asset relevance."

Intensity Levels

lite — Drop filler/hedging, keep articles, keep full sentences. Use for shorter docs.

full (default) — Drop articles and filler, allow fragments, use short synonyms. Use for most blog posts.

ultra — Abbreviate terms (config/auth/API), strip conjunctions, use arrows for causality (X → Y). Use only when explicitly requested.

Content Writing Rules

When writing cavemanBody for Sanity:

1. Take the full body content as source
2. Apply caveman compression to all prose
3. Keep headings short (2–4 words max)
4. Lists: one fragment per bullet, no punctuation at end
5. Code examples: unchanged
6. Numbers/statistics: always keep
7. Links and CTAs: keep but shorten text ("Read more →" not "Click here to read more about this topic")

Heading compression

Bad: "How to Get Started with the News Impact Screener" Good: "Getting Started"

Bad: "Understanding How Sentiment Scores Are Calculated"
Good: "Sentiment Score Math"

Intro compression

Bad: "In this post, we're going to take a deep dive into how the News Impact Screener works and why it can be a powerful tool for retail investors who are looking to stay ahead of the market." Good: "News Impact Screener. Connects headlines to stocks. Retail investors. No terminal needed."

Body compression

Bad: "When a news event occurs, our system analyzes the article using a large language model to extract the key themes, the sentiment, and the potential impact on related assets." Good: "News event hits. LLM extracts: themes, sentiment, asset impact."

When Writing Blog Posts

Always produce two outputs:

Normal body (body field):

• Full prose, complete sentences
• SEO-friendly headings
• Conversational but professional tone

Caveman body (cavemanBody field):

• Apply full-intensity caveman rules to all prose from the normal body
• Same structure, headings, and code — just compressed
• Target: ~70% shorter than normal body

When Writing Documentation

Always produce two outputs:

Normal body (body field):

• Complete explanations with context
• Step-by-step format where relevant
• Examples with explanation

Caveman body (cavemanBody field):

• Steps only, no explanation of why unless critical
• "Do X. See Y. Fix with Z." format
• Numbered steps stay numbered, but text compressed
• Remove all background/context paragraphs

Sanity Schema Reference

Both fields use blockContent (Portable Text). When outputting content for Sanity:

• Headings → appropriate heading level (h2/h3/h4)
• Paragraphs → normal blocks
• Code → code blocks (language specified)
• Lists → bullet or numbered arrays
• cavemanBody follows same structure as body but with compressed text nodes