How to use the Learn: Multi-Source Research agent skill

What the learn skill makes the agent do

Without the skill, an agent answers from what it already knows or runs a quick search. With the learn skill installed, the agent follows a six-phase process: it collects primary sources, takes notes from them, builds an outline, drafts sections, edits, and then reviews the result. The source describes the goal as turning unfamiliar material into a reliable mental model, reference, article or notes set you can use.

The skill checks for contradictions and flags uncertainty instead of silently choosing one answer. It is designed to support your thinking rather than replace it.

• Deep Research: builds understanding of a domain and ends with a publish-ready draft.
• Quick Reference: builds a working mental model quickly and ends with notes only.
• Write to Learn: you already have materials, and writing them up forces understanding; ends with a draft.
• Canonical Article: one article meant to cover a topic so thoroughly that readers need nothing else; ends with a single authoritative reference.

Install the learn skill (ZIP or Claude apps)

Download the ZIP of the learn skill from this page. The skill's files include SKILL.md, which holds all of its instructions.

Option one: unzip it into your personal skills folder at ~/.claude/skills/. Option two: unzip it into a project's .claude/skills/ folder so it applies only to that project.

Option three: upload it in the Claude apps under Settings → Capabilities → Skills.

• Keep the folder structure from the ZIP intact so SKILL.md sits where the agent expects it.
• Restart or start a new session if the agent does not pick up the skill straight away.

How to trigger the skill with natural requests

You do not need a special command. Describe what you want in plain language, and the skill's description tells the agent when to use it: researching an unfamiliar domain or compiling materials into one reference. The skill also lists trigger phrases, both English and Chinese, such as research, deep dive, help me understand, compile sources and unfamiliar domain.

If you want a single comprehensive piece, say so. Phrases like one article is enough or make this a one-stop reference point the agent toward Canonical Article mode. If the mode is unclear, the skill suggests Quick Reference.

• Help me understand how retrieval-augmented generation works, using these links I have collected.
• Compile these five articles into one structured reference I can use without searching further.
• Deep dive into this unfamiliar area and draft an article I can publish after I review it.
• Give me quick research notes on this topic, no article needed.

What you need before you start

The skill itself is text instructions. It works with whatever sources you provide or whatever web search your environment offers. Two companion skills improve results but are optional. The /read skill fetches URLs, and the /write skill removes AI-style patterns from text. If /read is missing, the agent warns once and falls back to its native fetch tools. If /write is missing, the agent warns once and scans the text manually for filler and binary contrasts.

A search plugin is also optional. Without one, the environment's native web search is used.

• Optional: /read for fetching URLs, including paywalled, JavaScript-heavy and Chinese-platform pages.
• Optional: /write for editing out AI-style patterns.
• Optional: a search plugin for discovering sources.
• Runs scripts: no. Uses network: yes.

Safety notes and limits

The skill does not run scripts, but it does use the network, because its job is to find and fetch sources. Review the sources it collects, especially for sensitive topics. The source says that when /read is not installed, coverage drops on paywalled, JavaScript-heavy and Chinese-platform pages, and the output may be incomplete.

The skill also says not to use it for quick factual lookups or for reading a single file, because the process is heavy. Its own rules say it stops at publish confirmation: it does not upload, post or distribute anything unless you explicitly ask.

• Expect slower results: collecting and cross-checking many sources takes time.
• The skill targets 5 to 10 sources for a blog post and 15 to 20 for a deep technical survey.
• Contradictions between sources are kept visible, with the evidence each side gives.
• Your own reading is still the final check: the skill's self-review asks you to read the whole piece before publishing.