Writing content across different platforms and styles. This skill should be used when creating blog posts, LinkedIn posts, X/Twitter threads, technical documentation, or other written content. It intelligently selects the appropriate writing style based on the platform, audience, and content type, then applies that style consistently throughout the piece.
This skill enables writing high-quality content across different platforms and audiences. It dynamically selects the appropriate writing style based on the task, then applies that style consistently. Available styles include deep technical blog posts, social media content (LinkedIn, X/Twitter), tutorials, and more.
Before writing, determine the appropriate style based on these factors:
| Factor | Consider | |--------|----------| | Platform | Blog, LinkedIn, X/Twitter, documentation, newsletter | | Audience | Engineers, executives, general tech audience, developers | | Depth | Deep-dive technical, lightweight explainer, announcement | | Tone | Authoritative, conversational, promotional, educational | | Length | Long-form (2000+ words), medium (500-1500), short (<500), thread |
Is this for social media?
├─ Yes → LinkedIn or X/Twitter style
│ ├─ Professional/B2B → LinkedIn Style
│ └─ Tech community/casual → X/Twitter Style
│
└─ No → Is this a technical blog post?
├─ Yes → What voice/style?
│ ├─ Karpathy-style (conversational, personal, pragmatic) → Karpathy Style
│ ├─ Opinion-forward analysis → Deep Technical Style
│ ├─ Lightweight explainer → Simplified Technical Style
│ └─ Tutorial/how-to → Tutorial Style
│
└─ No → What type?
├─ Announcement → Announcement Style
├─ Newsletter → Newsletter Style
└─ Documentation → Documentation Style
Use when: Writing technical content that feels personal and accessible. Best for blog posts about AI/ML, tutorials, or technical essays where you want the voice of an expert friend explaining things over coffee.
Characteristics:
Load: references/karpathy-style.md for comprehensive guidance
Core Philosophy: "Don't be a hero. Start simple. Be patient."
Opening Hooks (choose one):
| Pattern | Example | |---------|---------| | Personal experience | "I still remember when I trained my first recurrent network..." | | Intriguing premise | "We're going to feed 2 million selfies to a neural network and have it tell us what makes a selfie good." | | Leaky abstraction | "Neural nets are not 'off-the-shelf' technology the second you deviate slightly from training an ImageNet classifier." | | Strong observation | "Trees are solidified air." |
Example opening:
"I've trained a lot of neural networks over the years and I've come to find that the process isn't nearly as simple as it seems. They are a leaky abstraction. Here's what I've learned."
Tweet/X style (short-form):
Use when: Writing opinion-forward technical analysis about AI, LLMs, agents, or engineering topics. Best for content targeting senior engineers who want bold takes backed by evidence.
Characteristics:
Load: references/deep-technical-style.md for comprehensive guidance
Core Philosophy: "Do the simple thing first."
Opening Hooks (choose one):
| Pattern | Example | |---------|---------| | Contrarian position | "The death of prompt engineering has been greatly exaggerated." | | Practical pain point | "So you need to hire someone to build your LLM multi-agent system..." | | Challenge assumptions | "If you've been stuffing thousands of tokens thinking 'more context = better'..." | | Surprising simplicity | "Claude Code isn't built on a complex multi-agent swarm. It's a single-threaded loop." |
Example opening:
"Most AI architectures are overengineered. A single-threaded loop often outperforms elaborate multi-agent orchestrations."
Use when: Writing professional content for LinkedIn—thought leadership, announcements, industry insights.
Characteristics:
Structure:
[Hook line that creates curiosity]
[Personal context or story - 2-3 short paragraphs]
[Key insight or lesson - bulleted if multiple points]
[Call-to-action or engaging question]
#relevanthashtags (3-5 max)
Example:
Most AI features never make it to production.
Not because the models aren't good enough.
Because teams build for demos, not for users.
After shipping 50+ AI features, here's what actually matters:
→ Start with the simplest possible implementation → Measure what users do, not what they say → Treat AI like any other code: test it, monitor it, iterate
What's the biggest lesson you've learned shipping AI?
Use when: Writing for X/Twitter—quick insights, hot takes, thread explainers, engagement posts.
Characteristics:
Thread structure:
1/ [Bold claim or hook]
2/ [Context or "here's why"]
3-N/ [Key points, one per tweet]
N+1/ [Takeaway or call-to-action]
[Optional: link to longer content]
Example:
Unpopular opinion: RAG is overengineered for most use cases.
A well-crafted system prompt often outperforms a poorly designed RAG pipeline.
Thread on when to use (and skip) RAG 🧵
Use when: Writing accessible technical content for broader audiences—explainers, introductions, "what is X" posts.
Characteristics:
Example opening:
Every time you ask ChatGPT a question, something interesting happens behind the scenes. The AI doesn't just "know" the answer—it builds one, word by word, based on patterns it learned from millions of conversations.
Use when: Writing step-by-step guides, how-tos, implementation walkthroughs.
Characteristics:
Structure:
## What You'll Build
[One sentence + screenshot/diagram]
## Prerequisites
[Bulleted list]
## Steps
### Step 1: [Action verb]
[Explanation + code]
### Step 2: ...
## Troubleshooting
[Common issues + solutions]
## Next Steps
[Where to go from here]
Use when: Announcing features, releases, milestones, company news.
Characteristics:
Example:
We just shipped streaming support for Deep Agent.
Now you can see your agent's thinking in real-time—tool calls, responses, everything—as it happens.
Why it matters: Debugging agent loops is 10x easier when you can watch them run.
Try it:
createDeepAgent({ model, onEvent: console.log })
Ask these questions:
For Deep Technical style, load references/deep-technical-style.md for detailed guidance including:
For other styles, refer to the characteristics above.
Apply the chosen style's characteristics from the start. Don't write generically then "style" it later—the style should inform structure, not just tone.
Each style has implicit requirements. Before finalizing:
Karpathy Style:
Deep Technical Style:
LinkedIn Style:
X/Twitter Style:
Critical: All content must avoid patterns that signal AI-generated text. Load references/ai-slop-avoidance.md for comprehensive guidance.
Never use these overused AI words:
| Category | Banned Words | |----------|--------------| | Verbs | delve, embark, unleash, harness, unlock, navigate, revolutionize, foster, elevate | | Adjectives | vibrant, bustling, intricate, pivotal, crucial, cutting-edge, robust, seamless, meticulous | | Nouns | tapestry, realm, landscape, paradigm, synergy, beacon, testament, game-changer | | Phrases | "it's important to note", "in today's digital age", "dive into", "at its core", "a myriad of" |
| Pattern | Problem | Fix | |---------|---------|-----| | Em-dashes (—) | Strong AI signal, ban entirely | Use commas, parentheses, periods (hyphens/en-dashes OK) | | Uniform sentences | All ~25-30 words | Mix 5-word punches with 40-word explanations | | Excessive hedging | "might", "arguably", "perhaps" | Take clear positions | | Immediate lists | Jump to bullets without narrative | Build tension first | | Perfect grammar | Zero typos feels robotic | Occasional imperfection is human | | Balanced both-sides | Artificial neutrality | Take a stance |
karpathy-style.md - Guide for writing in Andrej Karpathy's conversational, pragmatic technical style with blog post patterns, tweet patterns, and recurring themesdeep-technical-style.md - Comprehensive guide for opinion-forward technical blog posts with detailed examples, rhetorical techniques, and pre-publication checklistai-slop-avoidance.md - Detailed guide to avoiding AI-generated content patternsTo add new writing styles:
references/[style-name]-style.mdGenerate or edit images via Gemini 3 Pro Image (Nano Banana Pro).
Batch-generate images via OpenAI Images API. Random prompt sampler + `index.html` gallery.
Generate spectrograms and feature-panel visualizations from audio with the songsee CLI.
Extract frames or short clips from videos using ffmpeg.
Search GIF providers with CLI/TUI, download results, and extract stills/sheets.
Category:media-generate