| name | explain_ai_native_eli5 |
| description | Explain AI-native engineering, four-word naming, TDD-first workflows, context engineering, and agent design in plain English. Use when Codex needs to turn dense notes, architecture writeups, or prompt-engineering guidance into an ELI5 markdown explainer with simple analogies, low-drama tone, and crisp takeaways. |
Explain AI Native ELI5
Goal
Turn dense AI-native engineering notes into a short Markdown explainer that a smart beginner can understand in one pass.
Default Reader
Assume the reader is either:
- A curious 12-year-old
- A non-engineer teammate
- A new developer who wants the idea before the jargon
Writing Rules
- Start with the main idea in one sentence.
- Use plain words first, technical words second.
- Keep exact terms only when they matter, then define them once.
- Prefer short sections and short paragraphs.
- Use one everyday analogy per major concept.
- Keep the tone calm, factual, and low-drama.
- Output Markdown only unless the user explicitly asks for another format.
Workflow
- Find the core claim.
- Answer: "What is this really saying?"
- Translate each concept into a real-world analogy.
four-word naming -> putting clear labels on drawers
TDD-first -> writing the answer key before building
summary docs -> leaving bookmarks so you do not get lost
self-reasoning -> talking out loud to catch your own mistakes
anti-pattern lists -> a "do not step here again" map
context engineering -> packing the right tools into a backpack
sub-agents -> sending teammates into different rooms to check things in parallel
- Explain in three layers.
- Layer 1: one-sentence takeaway
- Layer 2: short story version
- Layer 3: concrete rules or examples
If the user wants a visual terminal-native companion, pair this skill with craft-ascii-diagram-layouts:
- this skill owns the plain-English teaching arc
craft-ascii-diagram-layouts owns the monospaced page composition
- Preserve what is source-specific.
- If the source claims numbers like
67% faster or 90% fewer bugs, present them as claims from the source notes unless independently verified.
- Keep facts separate from interpretation.
- End with one sticky sentence.
- The final line should be memorable enough that the reader can repeat it later.
AI-Native Cheat Sheet
Use these simple translations when the source material matches the meta-patterns in this repo.
| Source idea | ELI5 translation |
|---|
| LLMs are search and assimilation tools | The model is better at matching patterns than magically understanding everything |
| Four-word naming | Clear names help both people and models find the right drawer faster |
| Iteration matters | The first answer is a draft, not the finish line |
| Summary documents | Long chats forget, so write checkpoints |
| Self-reasoning | Asking the model to explain itself helps it catch weak logic |
| Anti-pattern references | Past mistakes are warning signs for future work |
| TDD-first | Tests tell the model what "done" looks like before coding starts |
| Context engineering | Good results depend on loading the right information, not all information |
| Give agents a computer | Agents work better when they can use files, shell tools, and scripts |
| Context offloading | Save old details to files instead of stuffing everything into the prompt |
| Context isolation | Split separate problems into separate workers so they do not trip over each other |
If the user wants the built-in plain-English version of this repo's meta-patterns note, load AI-native engineering ELI5.
Recommended Output Shape
Use this structure unless the user asks for something shorter:
# <Topic in simple words>
## Big Idea
## Why It Matters
## Core Ideas Made Simple
## Tiny Example
## What To Remember
Tiny Example
When explaining four-word naming, prefer examples like this:
Bad name: `filter()`
Better name: `filter_implementation_entities_only()`
Why? The better name tells you:
- what it does
- what it looks at
- what it keeps
Guardrails
- Do not turn ELI5 into "wrong but cute."
- Do not remove important nuance when safety, correctness, or measurement matters.
- Do not repeat hype claims as universal truth.
- Do not drown the reader in jargon.
- Do not explain every detail from the source; keep only the details that unlock understanding.
- Do not default to ASCII diagrams unless the user wants a visual explainer or the layout would clearly help.
Done Check
Before finishing, verify:
- The first paragraph gives the big idea fast.
- Each major concept has a simple analogy.
- There is at least one concrete example.
- The explanation is readable by a smart beginner.
- The ending compresses the whole idea into one sentence.