A tiny Skill called ELI5 has been circulating among developers for a surprisingly simple reason: it changes a model’s output without burying the model under a giant prompt.
Give it a topic such as agents, DNS, RAG, or model post-training, and it returns a visual, low-text HTML explainer instead of several pages of prose. It can also inspect an unfamiliar repository and draw the main modules and data flow.
The surprising part is the source. The current SKILL.md is only ten lines long. Once the metadata is removed, the core instruction is essentially one sentence:
Explain like I'm someone who knows nothing about this topic,
using a HTML artifact with big pictures and few words.
Topic: $ARGUMENTS
Explain like I'm someone who knows nothing about this topic,
using a HTML artifact with big pictures and few words.
Topic: $ARGUMENTS
There is no prescribed grid, color palette, component library, or animation system. The Skill fixes the audience, goal, and medium, then lets the model decide how to build the page.
It is a useful lesson for agent design: as models become more capable, a good Skill does not always need to become longer. It needs to become clearer.
First, a useful distinction: community is not official
ELI5 lives in anthropics/claude-plugins-community, a read-only mirror of Anthropic’s community plugin marketplace. The repository says listed plugins pass the marketplace submission, scanning, and distribution process.
That does not make every entry an Anthropic-maintained product.
- The repository is hosted under the Anthropic GitHub organization.
- ELI5 lists Thariq Shihipar as its author.
- The current plugin version is 1.0.0 and the license is MIT.
- It is a community plugin, separate from
anthropics/claude-plugins-official. - Claims that it is “internally viral” come from community reporting, not published adoption data from Anthropic.
This distinction does not reduce the Skill’s usefulness. It simply avoids turning community distribution into an official product endorsement.
Why one sentence changes the result
The prompt looks minimal, but it makes three important decisions.
1. It changes the assumed reader
“Someone who knows nothing about this topic” forces the model to remove unexplained abbreviations and hidden prerequisites.
A normal RAG explanation may jump immediately to embeddings, vector databases, and recall. An ELI5 explanation has to begin with the more basic problem: finding the most relevant pieces in a large collection before asking a model to answer.
2. It changes the medium
The Skill does not request an article. It requests an HTML artifact.
That encourages the model to ask different questions:
- Which relationship should become a flow diagram?
- Which alternatives belong side by side?
- Where can color and icons create hierarchy?
- Which idea benefits from a small interaction?
- How can a page reveal one concept at a time?
3. It constrains information density
“Big pictures and few words” is an editorial rule, not a decorative preference.
Large visuals prioritize relationships. Limited text forces the model to remove secondary detail. The model cannot merely paste a long answer into cards; it has to decide what the reader should remember.
How to install ELI5 in Claude Code
Option 1: Install from the community marketplace
Add the marketplace:
claude plugin marketplace add anthropics/claude-plugins-community
claude plugin marketplace add anthropics/claude-plugins-community
Install ELI5:
claude plugin install eli5@claude-community
claude plugin install eli5@claude-community
Then invoke it in Claude Code:
/eli5 how does DNS work?
/eli5 how does DNS work?
If the Skill is discovered correctly, Claude should create a visual HTML explainer for the topic.
Option 2: Ask an agent to install the Skill file
Because ELI5 is ultimately a SKILL.md, an agent with file access can install it into its own Skill directory:
Read and install this Skill:
https://github.com/anthropics/claude-plugins-community/blob/main/eli5/skills/eli5/SKILL.md
Before writing anything, tell me which directory you will use.
Do not overwrite an existing Skill with the same name.
Read and install this Skill:
https://github.com/anthropics/claude-plugins-community/blob/main/eli5/skills/eli5/SKILL.md
Before writing anything, tell me which directory you will use.
Do not overwrite an existing Skill with the same name.
Codex and other agents may use different directories and invocation rules. Do not assume /eli5 works everywhere. A portable request is more explicit:
Use the ELI5 Skill to turn “how DNS works” into a visual HTML explainer with large pictures and minimal text.
Use the ELI5 Skill to turn “how DNS works” into a visual HTML explainer with large pictures and minimal text.
Six practical uses
1. Learn an unfamiliar concept
/eli5 what is model post-training, and why does it improve a model?
/eli5 what is model post-training, and why does it improve a model?
Instead of opening with supervised fine-tuning, preference optimization, and reinforcement learning, the page can establish one useful idea first: pre-training helps a model know things; post-training helps it apply that knowledge in a task.
2. Onboard to an unfamiliar repository
Use the ELI5 Skill to inspect this project and create a visual HTML guide showing:
1. where a user request enters;
2. how data moves through the main modules;
3. which databases and external services are called;
4. the first five files a new engineer should read.
Use the ELI5 Skill to inspect this project and create a visual HTML guide showing:
1. where a user request enters;
2. how data moves through the main modules;
3. which databases and external services are called;
4. the first five files a new engineer should read.
The reading order makes this much more useful than “explain this repository.”
3. Understand an architecture decision
/eli5 why does this system use a message queue instead of direct service-to-service calls?
/eli5 why does this system use a message queue instead of direct service-to-service calls?
A side-by-side page can make the tradeoffs in throughput, coupling, retries, and operational complexity visible.
4. Explain an incident
Use the ELI5 Skill to turn the current incident report into one visual page.
Show the normal path, failure path, first anomaly, amplifying factors, and impact.
Do not invent causes that are absent from the report.
Use the ELI5 Skill to turn the current incident report into one visual page.
Show the normal path, failure path, first anomaly, amplifying factors, and impact.
Do not invent causes that are absent from the report.
That final constraint matters. ELI5 improves communication; it does not have permission to manufacture incident facts.
5. Review for an interview
/eli5 how does RAG work, and how can retrieval recall be improved?
/eli5 how does RAG work, and how can retrieval recall be improved?
Start with query, retrieval, reranking, and generation. Then ask for a second screen comparing chunking, hybrid search, and rerankers.
6. Build internal training material
Use the ELI5 Skill to explain our support agent to non-technical colleagues.
Cover only four stages: customer question, knowledge retrieval, model response, and human handoff.
Put one conclusion on each screen and end with three comprehension questions.
Use the ELI5 Skill to explain our support agent to non-technical colleagues.
Cover only four stages: customer question, knowledge retrieval, model response, and human handoff.
Put one conclusion on each screen and end with three comprehension questions.
The result can open a training session, while policies and risk controls should still link back to the canonical documentation.
A stronger ELI5 template for real projects
The original is excellent for exploration. A team that needs repeatable output can add a few constraints without turning it into a thousand-line design system:
---
name: eli5-project
description: Explain an unfamiliar technical topic or project with a visual HTML artifact.
---
Explain the topic to a reader who has no prior knowledge.
Create a self-contained HTML artifact with large visuals and few words.
Requirements:
1. Start with a one-sentence explanation of what it is and why it matters.
2. Show the main flow before introducing details.
3. Use only facts from the provided files and cited sources.
4. Label uncertain or inferred information explicitly.
5. Keep each screen focused on one idea.
6. End with three common misunderstandings and three next steps.
7. Save the result as `eli5-explainer.html`.
Audience: $AUDIENCE
Topic: $ARGUMENTS
---
name: eli5-project
description: Explain an unfamiliar technical topic or project with a visual HTML artifact.
---
Explain the topic to a reader who has no prior knowledge.
Create a self-contained HTML artifact with large visuals and few words.
Requirements:
1. Start with a one-sentence explanation of what it is and why it matters.
2. Show the main flow before introducing details.
3. Use only facts from the provided files and cited sources.
4. Label uncertain or inferred information explicitly.
5. Keep each screen focused on one idea.
6. End with three common misunderstandings and three next steps.
7. Save the result as `eli5-explainer.html`.
Audience: $AUDIENCE
Topic: $ARGUMENTS
This version adds source discipline, inference labels, a deterministic output file, and clear next steps. Colors, card styles, and animation remain content-dependent choices for the model.
ELI5, normal chat, and grill-me
| Method | Best for | Strength | Limitation |
|---|---|---|---|
| Normal chat | Fast facts and explanations | Flexible and quick | Often becomes dense prose |
| ELI5 | Building a first mental model | Visual and approachable | Can oversimplify |
| grill-me | Clarifying an incomplete plan | Finds missing decisions and dependencies | Does not focus on visual explanation |
| Canonical documentation | Recording exact rules and boundaries | Citable and auditable | Higher learning cost |
ELI5 and grill-me are both short, but they constrain different things. ELI5 defines the final communication format. grill-me defines how the conversation should progress. Both show that a Skill’s value comes from identifying the constraints that determine the outcome—not from maximizing prompt length.
Limits to understand before using it
A simple explanation can still be wrong
A polished visual can feel authoritative. Versions, prices, legal rules, medical guidance, security claims, and incident details still require primary-source verification.
Simplification removes edge cases
ELI5 is a first map, not the complete territory. Use it to understand the main path, then read the canonical documentation for exceptions and operational detail.
Repositories may contain sensitive information
Exclude secrets, customer data, production configuration, and security-sensitive files. Give the agent the smallest directory scope necessary.
Generated HTML needs review
Before publishing, inspect scripts, outbound requests, third-party assets, mobile behavior, and accessibility. Model-generated HTML should not move directly into production.
Visual output may cost more than plain text
The model generates explanation, layout, CSS, and sometimes interaction code. Use ELI5 when a visual model will be reused or when the relationship is genuinely hard to explain in prose.
Where Apito fits
ELI5 defines how the agent should explain a topic. Apito belongs in the model-access layer, where a team manages API keys, endpoints, available models, request records, and cost visibility.
If Claude Code or another agent is already configured through Apito, the Skill generally does not need a separate key. It uses the model connection configured for the current agent.
User question
↓
ELI5 Skill: audience, objective, and output format
↓
Claude Code / Codex / another agent: read sources and execute
↓
Apito: model API access, model selection, logs, and cost visibility
↓
HTML file → human review → publish or archive
User question
↓
ELI5 Skill: audience, objective, and output format
↓
Claude Code / Codex / another agent: read sources and execute
↓
Apito: model API access, model selection, logs, and cost visibility
↓
HTML file → human review → publish or archive
Apito does not install the Skill or guarantee that an explainer is factually correct. Teams remain responsible for source quality, agent permissions, and final review.
Apito is an independent third-party model API access service and does not represent Anthropic. ELI5 is a community plugin, not an Apito-developed Skill.
The larger lesson
The most useful part of ELI5 is not that every answer should become a webpage. It is that the Skill preserves only three constraints that materially shape the result: assume no prior knowledge, explain visually, and reduce the amount of text.
Prompt writers often respond to uncertainty by specifying every step. Stronger models make another approach possible: define the outcome and non-negotiable boundaries, then let the model supply implementation details.
Short is not automatically better. High-risk and highly repeatable work needs sources, permissions, output contracts, and acceptance criteria. A sensible process is to begin with the smallest useful Skill, observe real failures, and add constraints only when those failures justify them.
FAQ
Is ELI5 an official Anthropic plugin?
No. It is a community plugin distributed through a marketplace mirror hosted by the Anthropic GitHub organization. The plugin lists Thariq Shihipar as its author.
Why is /eli5 missing after installation?
Confirm that the community marketplace was added successfully, the install command completed without errors, and a new Claude Code session has been started. Other agent products use their own Skill discovery and invocation rules.
Can ELI5 read an entire repository?
Only within the file permissions granted to the agent. Limit scope to relevant directories and exclude secrets, customer data, and production configuration.
Can the generated HTML be published directly?
It should be reviewed first for factual accuracy, scripts, external requests, asset rights, responsive layout, and accessibility.
Does ELI5 require a separate API key?
No. The Skill is an instruction file. Running it consumes tokens and quota from the model connection already configured in the agent.
Should every topic use ELI5?
No. Plain chat is faster for a simple fact. Canonical documentation and professional review remain necessary for exact specifications, complete arguments, and high-risk decisions.
Sources
To put Skills into a repeatable agent workflow, use Apito as a unified model API access layer and compare model quality, latency, and cost on your own tasks.
