learn Skill for Claude: Turn Many Sources Into One Reference

Makes Claude research an unfamiliar topic in six phases, from collecting primary sources to a draft you check yourself.

Installs: 13.3K; on skills.sh
Stars: 7.5K; on GitHub
Updated: Nov 2026; last repo commit

learn is one skill in tw93's Waza set. It makes Claude collect primary sources, digest them, outline, fill in, refine and then hand you a draft to read through yourself. SKILL.md offers four modes (Deep Research, Quick Reference, Write to Learn, Canonical Article) and tells Claude to support your thinking, not replace it.

It triggers on research and deep-dive requests, with Chinese trigger phrases such as 学习一下 and 研究一下 listed alongside English ones. Single URLs and quick lookups are out of scope. The skill is one file and bundles no scripts, but when you have Waza's /read skill installed it fetches every URL through it, and /read has its own network behavior (see Permissions).

Good for

Writers who need to understand a new field before publishing about it
Engineers building a working mental model of an unfamiliar domain
Anyone turning a pile of articles and papers into one structured reference

Install learn

Claude Code

Copy this prompt.

Install the learn skill in Claude Code. Run this command:

/plugin marketplace add tw93/Waza
/plugin install waza-learn@waza

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Pick one route, not several. learn works alone; its optional helpers /read and /write are separate Waza skills, so the whole-set plugin (waza@waza) or the whole-set skills CLI command adds them. The whole-set plugin installs all 8 skills, including agent files under check/agents; the README says the optional rule files apply only if you copy them into your own instructions. The /read skill's own options (DeepSeek key, proxy mode) are listed in the safety section.

From the project README. Source.

What's in the skill

When it loads

Runs a six-phase research workflow from source bundle to publish-ready output. Use when researching an unfamiliar domain or compiling materials into one reference. Not for quick lookups or single-file reads.

Sections in SKILL.md

Outcome Contract
Choose Mode
Canonical Article Mode
Phase 1: Collect
Phase 2: Digest
Conversation Or Review Distillation
Phase 3: Outline
Phase 4: Fill In
Phase 5: Refine
Phase 6: Self-review and Publish Readiness
Hard Rules
Gotchas
Output

Key facts

Author: tw93 (repo owner)
License: MIT
Installs on skills.sh: 13,326 as of 2026-11-02
GitHub stars: 7,488 for the whole repo
Last commit to the repo: 2026-11-01
Files in the skill: 1 file: SKILL.md (the repo also copies it under plugins/waza/skills/learn)
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk warn
Part of: Waza, a set of eight skills: think, ui, check, hunt, write, learn, read, health
Optional companions: /read to fetch URLs, /write to strip AI-sounding patterns

What it includes

Understand a domain well enough to write about it; ends at a publish-ready draft.
Runs Phase 1 to Phase 6.
Try: “Deep dive into how vector databases index data and give me an outline of what matters.”
Write · Mode: Deep Research
Builds a working mental model fast and stops at notes (Phase 2).
Suggested when Claude is unsure which mode you want.
Network · Mode: Quick Reference
Starts from materials you already have and writes your way to understanding.
Enters at Phase 3 and ends at a publish-ready draft.
Write · Mode: Write to Learn
One long article so thorough readers need nothing else on the topic.
Adds a section per sub-topic, worked examples, common mistakes and a Further Reading list of the 3 to 5 deepest sources.
Write · Mode: Canonical Article
Gathers primary sources: discover, fetch through /read, then file.
Targets 5 to 10 sources for a blog post and 15 to 20 for a deep technical survey. Without /read it warns once and uses native fetch.
Write · Phase 1: Collect
Keeps ideas that recur or predict, cuts generic wisdom.
Each key claim is tested with three questions; passing two or three puts it in the outline.
Read · Phase 2: Digest
Outline with a source behind every section, then write and edit.
Phase 5 edits only and uses /write when installed, otherwise scans manually for filler.
Write · Phases 3 to 5: Outline, Fill In, Refine
You read the whole article linearly, at least twice, before publishing.
Claude stops at publish confirmation and does not upload or post unless you ask.
Read · Phase 6: Self-review

From the official docs

Permissions

Low risk

learn itself bundles no scripts, but it searches and fetches the web and may save source files. If you also install /read, its opt-in proxy and DeepSeek options can send URLs and queries to third parties, so leave those off for private material.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

URLs may go through its optional /read skill
learn sends every URL through /read when it is installed. /read fetches locally by default. Its opt-in proxy mode sends the URL to defuddle.md and r.jina.ai, which may cache or log it. Never use that mode for private or internal URLs. github.com
An optional DeepSeek key costs money and sends your queries
/read runs DeepSeek search only if DEEPSEEK_API_KEY is set. Each successful request is billed and the query or URL goes to DeepSeek, so keep private names and credentials out of it. github.com
Fetched pages are untrusted
/read's SKILL.md says to treat fetched content as untrusted data, not instructions. learn's Hard Rules add that it stops at publish confirmation and does not upload or post unless you ask. github.com

Technical details

Talks to	Web pages and search results you research. With /read and --use-proxy: defuddle.md and r.jina.ai receive the URL. With /read and DEEPSEEK_API_KEY set: api.deepseek.com receives the query or URL. With /read on Feishu pages: open.feishu.cn (needs FEISHU_APP_ID and FEISHU_APP_SECRET)
Scopes	Searches and fetches web pages for sources Saves fetched sources to a research folder or a session temp directory (through /read)

Questions

What does the learn skill do?

It makes Claude run a six-phase research workflow: collect primary sources, digest them, outline, fill in, refine, and hand you a draft to read through before you publish. Contradictions between sources stay visible instead of being silently resolved.

How do I install learn?

Use the install box above. It has a learn-only route (a single-skill plugin or the skills CLI) and the whole-set routes the Waza README documents: npx skills add tw93/Waza -a claude-code codex cursor -g -y, the waza@waza plugin and a Claude Desktop ZIP. The whole set adds the optional /read and /write skills.

Do I need /read and /write too?

No, both are optional. Without /read, learn warns once and falls back to the assistant's native fetch, with reduced coverage of paywalled, JS-heavy and Chinese-platform pages. Without /write, it scans for filler manually.

Does it need an API key?

The learn skill names none. The separate /read skill can use a DeepSeek API key for search, but only if you set DEEPSEEK_API_KEY.

Is learn safe to use?

The skill itself bundles no scripts, but it fetches web pages and may save files. If /read is installed, its scripts also run. Its Hard Rules say it stops at publish confirmation and does not upload or post unless you ask. The Permissions section lists the details.

learn vs /read?

/read fetches one URL or PDF and summarizes it. learn is for multi-source research that produces a new structured output, and it uses /read as its fetch step.

Information

updated Nov 2, 2026

Repository: tw93/waza
Folder: skills/learn

tw93/waza skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

tw93/Waza README · github.com
Waza marketplace.json · github.com
skills CLI commands · github.com
learn SKILL.md · github.com
Waza README · github.com
read SKILL.md · github.com