eli5-用5岁小孩能听懂的方式解释一个话题

概述

ELI5 — Explain Like I'm 5

Turn any complex topic into something a child could understand.

Why it exists

Most explanations are written for people who already halfway know the answer. They hand-wave the hard parts, use words that need the explanation itself to make sense, and leave you more lost than before.

ELI5 is different. It forces a clean picture — one familiar analogy, a few short steps, and one concrete example.

How ELI5 works (technical)

ELI5 works by loading SKILL.md into the conversation context. When you type /eli5 <concept>, OpenClaw matches the description and injects the skill's rules into the model.

The model reads the rules and examples, then generates an explanation following the same pattern.

Switching models: Works as long as the model can follow contextual instructions. If a model ignores rules in context, results may vary.

Quick start

/eli5 schrödinger's cat   # explain any concept
/eli5 help                 # show all commands
/eli5 lang <lang>          # switch language (en/zh/es/kr/...)
/eli5 bonus on             # enable bonus explanations
/eli5 steps 5              # adjust max steps (default: 3, max: 5)
/eli5 fetch on             # enable auto-fetch from web

Default language: Controlled by ELI5_DEFAULT_LANG env var (read-only). If not set, fallback to English.

The Rules

1. Language priority:
   • ELI5_DEFAULT_LANG env var (read-only) — set once, use forever
   • /eli5 lang <lang> — switch and hold until next switch
   • Fallback: English
2. Be conversational, not formulaic — sound like a smart friend explaining, not a textbook. Skip the formula if it feels stiff.
3. Assume nothing — the user knows zero technical terms
4. Bridge: unknown → known — pick one familiar thing (toy, friend, magic box) and stick with it
5. Max n steps — short sentences, one action each. Default is 3, configurable via /eli5 steps <n>
6. Concrete example — one real thing the reader can picture. REQUIRED unless already perfectly clear.
7. Have personality — vary your explanations. A good comparison, a touch of humor, or a memorable contrast beats dry lists every time.
8. Bonus (default: off) — only show when bonus on or genuinely needed
9. Freshness indicator (always shown, in current language):
   • After the explanation, add a brief note:
     • Format: [Data: ~2024] [Freshness: ██████░░░░ 65%] [→ --fetch]
     • Visual bar: filled = fresh, empty = outdated
     • Scoring guidelines (subjective, use as reference):
       • Stable fields (philosophy, math, proven theories): 85-100
       • Technology that evolves slowly (OS, hardware): 70-85
       • Active tech fields (AI, frameworks, libraries): 50-70
       • Fast-moving topics (startups, trends, new releases): 30-50
     • Score 80-100: "Stable" → no action needed
     • Score 50-79: "Evolving" → consider –fetch
     • Score 0-49: "Outdated" → recommend –fetch strongly
     • If –fetch succeeded but data itself is old: use actual date + score based on how old the data is
     • If –fetch succeeded and data is latest: show actual date + high score
   • Example: [Data: 2024.03] [Freshness: ██████░░░░ 60%] [v1.2.0 — older]
10. Fresh fetch (default: off):
    • /eli5 fetch on → enable auto-fetch from web (GitHub, official docs, etc.)
    • /eli5 fetch off → disable
    • /eli5 <concept> --fetch → one-time fetch for this concept
    • If fetch succeeds → use latest content from web
    • If fetch fails → fall back to training data knowledge
    • Note: Fetch relies on OpenClaw's web search capability. If unavailable, falls back to training data.

Word Rules

Banned Words — Core Principle

Grandmother Test: Would my 80-year-old grandmother know this word? If NO → replace it or explain it immediately.

Happy Openings — Core Principle

Paint a picture. Make them visualize. Start with: "Imagine…", "Think of it as…", "Picture this…"

Forbidden Phrases — Core Principle

Never sound condescending, technical, or dismissive. Avoid: "Obviously…", "As you already know…", "Technically…", "In simple terms…"

Commands

/eli5 <concept>              # explain anything
/eli5 help                 # show this help (in current language)
/eli5 lang <lang>        # switch and hold language (en/zh/es/kr/...)
/eli5 bonus on|off          # toggle bonus (default: off)
/eli5 steps <n>             # set max steps (default: 3, max: 5)
/eli5 fetch on|off          # toggle auto-fetch from web (default: off)
/eli5 <concept> steps 5    # override steps for one answer
/eli5 <concept> in ZH      # override language for one answer
/eli5 <concept> --fetch   # fetch latest content for one answer

Env var (read-only): Set ELI5_DEFAULT_LANG in your environment. Skill reads it, does not write it.

Examples

** Schrödinger's cat** A cat that is both alive AND dead — until you open the box. Try it: /eli5 what is schrödinger's cat

** The Chinese Room** A person who pretends to understand Chinese but follows a rulebook instead. Try it: /eli5 what is the chinese room

** The Ship of Theseus** If you replace every plank of a ship, is it still the same ship? Try it: /eli5 what is the ship of theseus