DreambigOu/ELI5

A Claude Code skill that explains anything to anyone — kids, managers, engineers, parents. It adapts tone, vocabulary, analogies, and framing to match the audience.

Ever needed to explain a technical concept to your manager? Or break down code for a 5th grader? ELI5 makes Claude automatically adjust its explanation style based on who's listening.

Read the full blog post on how this skill was built: Building an ELI5 Skill for Claude

ELI5 what a database index is
Explain this code to my manager
Break down how git merge conflicts work for a 5th grader
Explain this error to my mom
Simplify this for a designer

The skill detects the target audience from your prompt and calibrates:

• Vocabulary — no jargon for kids, proper terminology for engineers
• Analogies — toys and playground for age 5, business outcomes for managers
• Tone — playful for children, professional for directors, warm for family
• Depth — short and sweet for simple audiences, nuanced for grad students
• Framing — impact/risk for managers, UX for designers, architecture for engineers

Copy the skill into your Claude Code skills directory:

git clone https://github.com/DreambigOu/ELI5.git
cp -r ELI5/skills/eli5 ~/.claude/skills/eli5

Then use it in Claude Code by saying things like "ELI5 this" or "explain this to my manager."

Read the full guide on how the eval system works: How to Evaluate a Claude Code Skill

Test cases are defined in eli5-workspace/evals.json. Add a new entry to the evals array:

{
  "id": 3,
  "name": "explain-recursion-teenager",
  "prompt": "Explain recursion like I'm 15",
  "audience": "Age 15",
  "assertions": [
    "Uses social media, gaming, or phone references as analogies",
    "Tone is casual but not cringey — no 'fellow kids' energy",
    "Correctly explains the concept of a function calling itself",
    "Mentions a base case or stopping condition"
  ]
}

Each test case needs:

• A prompt — what the user would say to Claude
• A name — directory-friendly identifier for storing results
• Assertions — specific, verifiable criteria to grade against (4 per test works well)

Prerequisites: Claude Code CLI installed and the skill installed at ~/.claude/skills/eli5/.

# Run all tests + auto-grade with pass/fail
python eli5-workspace/run-evals.py

# Run a single test
python eli5-workspace/run-evals.py --test=1

# Skip baseline, only test the skill
python eli5-workspace/run-evals.py --with-skill-only

# Grade existing outputs without re-running tests
python eli5-workspace/run-evals.py --grade-only

The script does three things:

1. Runs each prompt twice — once with the skill, once without (baseline)
2. Auto-grades every output against its assertions using Claude
3. Prints a pass rate summary comparing skill vs baseline

Example output:

--- Test 1: explain-db-index-age5 ---
  [with skill]
    PASS  #1 — No technical jargon present
    PASS  #2 — Uses book/page analogy and toy/messy room analogy
    PASS  #3 — Sentences are short and conversational
    PASS  #4 — Warm, enthusiastic tone with "huuuge", "super duper fast"
  [baseline]
    PASS  #1 — No technical jargon found
    FAIL  #2 — Uses phone book analogy, not child-friendly
    PASS  #3 — Sentences are generally short
    FAIL  #4 — Tone is informative but encyclopedic

=========================================
  PASS RATE SUMMARY — Iteration 1
=========================================
  With Skill:    10/12 passed (83.3%)
  Without Skill: 5/12 passed (41.6%)
  Delta:         41.7%
=========================================

Results are saved to eli5-workspace/iteration-N/, auto-incrementing with each run. Each test case produces grading.txt files with detailed evidence.

See eli5-workspace/eval-results.md for the full evaluation strategy and detailed grading.

The biggest improvement is in audience-specific framing — especially for non-technical audiences like managers (0% baseline to 50% with skill).

PRs welcome! Ideas for improvement:

• Add more audience types (e.g., CEO, intern, journalist)
• Add non-English language support
• Improve evaluation coverage with more test cases

MIT