eli12

name	eli12
description	Explain a codebase, subsystem, feature flow, or file cluster in clear accessible language. Trigger on how does this work, walkthrough, architecture explained simply, eli12, or mental model requests. Do NOT use for bug triage, code review, or code-change tasks.
compatibility	Works best in repositories where you can inspect files, search symbols, trace call paths, and optionally fan out read-only subagents for broad questions.
metadata	{"version":"1.0.0","repo_tags":["architecture","explanation","accessibility","teaching"]}
references	["explainer-prompt","explorer-prompt","analogy-patterns","gotchas"]

eli12

Explain code so a curious newcomer can build a correct mental model without drowning in jargon.

Inspired by poteto/how, but tuned for accessibility: simpler language, tighter structure, and real-world analogies that stay anchored to the actual code.

Decision Tree

1. If the user is asking how a subsystem, feature flow, runtime path, or file cluster works, use this skill.
2. If the request is very vague and the likely search space is large, ask 1-3 short narrowing questions before exploring. Default to scope-first questions such as "Which feature or path do you care about?", "Do you want runtime flow, architecture, or file tour?", and "Should I stay focused on frontend, backend, or data?".
3. If the question is narrow and local to one function, class, hook, or file, do a direct explain pass after a focused code search.
4. If the question spans multiple modules, services, or an end-to-end flow, split it into 2-4 exploration angles, gather findings in parallel when the harness allows it, then synthesize.
5. If the user mainly wants bugs, risks, or architectural critique, explain only enough to ground the discussion, then switch to normal review mode instead of staying inside the teaching frame.
6. If a real-world analogy would make the explanation fuzzier, use fewer analogies and stay closer to the code.

Quick Reference

Situation	Open / do	Why
User asks something broad like "Explain this repo" with no target	Ask a few short scope questions before reading deeply	Prevents token waste and irrelevant architecture tours
Explain one file, class, or helper simply	Read references/explainer-prompt.md	Keeps the answer direct and digestible
Explain a broad subsystem or runtime flow	Read references/explorer-prompt.md, then references/explainer-prompt.md	Gather evidence first, simplify second
Need better analogies without getting sloppy	Read references/analogy-patterns.md	Maps abstract code ideas to grounded everyday systems
Unsure where simplification goes wrong	Read references/gotchas.md	Avoids patronizing tone and misleading shortcuts
Need a ready-made output shape	Copy templates/explanation-outline.md	Gives a stable structure for the final explanation
Sanity-check prompt routing locally	Run python3 scripts/probe_eli12.py --prompt "How does auth work?"	Verifies trigger and complexity heuristics

What This Skill Optimizes For

• building the smallest correct mental model first
• asking for scope before doing an expensive repo tour
• defining jargon right when it appears
• using short, concrete paragraphs instead of wall-of-text explainers
• using friendly ASCII sketches when a flow or boundary is easier to see than to read
• tying every analogy back to real files, functions, and data flow
• helping the reader know where to look next in the codebase

Default Operating Mode

1. If the request is too vague to answer efficiently, ask 1-3 short narrowing questions before opening a large slice of the repo.
2. State your interpretation of the question once the target is clear.
3. Search for entry points, key types, and the files that actually move data or decisions.
4. Trace the real path from trigger to effect. Do not explain from filenames alone.
5. Collapse the findings into plain language.
6. Use one grounded analogy per important concept when it helps, then tie it back to exact code names.
7. Add a small ASCII sketch when topology, control flow, or data movement is easier to grasp visually than in prose.
8. End with a short map of where the important pieces live.

Output Shape

Use the sections that help. Skip the ones that would add noise.

• Big Picture - What this thing does and why it exists.
• Main Pieces - The handful of concepts or files the reader needs in order to follow the rest.
• The Story - The step-by-step flow from input to output, trigger to effect, or request to response.
• ASCII Sketch - A compact text diagram for flow, boundaries, or ownership when that clarifies the system faster than prose.
• Real-World Analogy - A concrete analogy that matches the actual job of the system, not a cartoon version of it.
• Where To Look - The files and directories that matter most if the reader wants to verify or extend the explanation.
• Sharp Edges - Non-obvious behaviors, hidden state, historical quirks, or easy misunderstandings.

Style Contract

1. Explain like a smart new teammate, not a child

Keep the language simple and digestible, but do not become cutesy, patronizing, or fake-cheerful.

2. Teach the system, not the syntax

Do not paraphrase every line of code. Focus on responsibilities, boundaries, data movement, and decision points.

3. Use analogies as scaffolding, not replacement

The analogy should help the reader get oriented. Immediately reconnect it to the real code so the explanation does not drift.

4. Define jargon on contact

If you must say "middleware," "idempotent," or "hydration," explain it in plain language the first time it appears.

5. Keep evidence visible

Name the actual files, symbols, or directories that support the explanation. Make it easy for the reader to verify the story.

6. Mark inference vs observation

If part of the explanation is inferred rather than directly observed, say so plainly.

7. Use ASCII sketches with discipline

Prefer a tiny chart over a paragraph only when it reduces confusion. Keep it narrow, label the real code concepts, and avoid decorative boxes.

Example:

request
  -> auth middleware
  -> controller
  -> service
  -> database

8. Spend tokens on the right scope

If the user asks something like "How does this app work?" and the repo is large, clarify the target before exploring. A short question is better than an unfocused architecture dump.

When To Fan Out

Fan out exploration only when it improves coverage.

• Broad architecture overviews
• Runtime flows that jump across layers
• Questions that span multiple packages or services
• Systems with separate data, orchestration, and UI concerns

Stay in one pass for narrow questions. Extra delegation slows simple explanations down.

Reading Guide

Need	Read
Core exploration workflow and evidence collection	references/explorer-prompt.md
Human-facing explanation tone and section shape	references/explainer-prompt.md
Everyday-system analogies that stay technically honest	references/analogy-patterns.md
Failure modes, oversimplification traps, and recovery	references/gotchas.md
High-level map of the reference set	references/README.md

Gotchas

1. Simpler language is not permission to hand-wave important behavior. If retries, caching, or background jobs matter, explain them plainly instead of omitting them.
2. A cute analogy that does not map to the real control flow is worse than no analogy at all.
3. Explaining a subsystem by listing files is not enough. Trace what actually happens.
4. If the user asked for one narrow thing, do not balloon the answer into a whole-architecture tour.
5. If the code is messy, say that directly. Accessibility should not sand off real complexity.