Research topics online and create comprehensive learning guides with RAG-optimized indexes
npx claudepluginhub agent-sh/learnResearch any topic online and create comprehensive learning guides with RAG-optimized indexes
No description available.
Production-ready workflow orchestration with 79 focused plugins, 184 specialized agents, and 150 skills - optimized for granular installation and minimal token usage
Curated collection of 141 specialized Claude Code subagents organized into 10 focused categories
Research any topic online and create comprehensive learning guides with RAG-optimized indexes for your AI agents.
AI agents work better when they have curated, pre-researched knowledge to draw from instead of searching the web on every question. /learn builds that knowledge base systematically - gathering sources, scoring them for quality, and synthesizing structured guides that agents can reference instantly.
Use cases:
agentsys install learn
Requires agentsys to be set up in your project.
/learn react hooks
This searches the web for ~20 sources on React hooks, scores each source for authority and depth, fetches the top results, and writes a synthesized guide to agent-knowledge/react-hooks.md with a companion resources/react-hooks-sources.json containing full source metadata.
The learn skill follows a six-stage methodology:
Progressive discovery - Funnel approach: broad queries for landscape mapping, focused queries for core content, deep queries for advanced material. Avoids noise from dumping all queries at once.
Quality scoring - Each source is scored on a 100-point scale across five dimensions: authority (3x weight), recency (2x), depth (2x), examples (2x), and uniqueness (1x). Official docs score highest; undated blog posts score lowest.
Just-in-time extraction - Only high-scoring sources get fetched. Summaries and key insights are extracted - never full content. This keeps token usage predictable and respects copyright.
Synthesis - A structured learning guide is generated with prerequisites, core concepts, code examples, common pitfalls, best practices, and further reading. Content is cross-referenced across sources, not copied from any single one.
RAG index - The master index (agent-knowledge/CLAUDE.md and AGENTS.md) is updated with the new topic, trigger phrases, and keyword mappings so agents can find relevant guides automatically.
Enhancement - Runs enhance:enhance-docs and enhance:enhance-prompts on the output to improve RAG retrieval quality. Skip with --no-enhance.
# Default depth (20 sources)
/learn recursion
# Deep research (40 sources)
/learn kubernetes networking --depth=deep
# Quick overview (10 sources)
/learn python decorators --depth=brief
# Skip enhancement pass
/learn typescript generics --no-enhance
| Level | Sources | When to Use |
|---|---|---|
brief | 10 | Quick overview, time-sensitive topics |
medium | 20 | Balanced coverage (default) |
deep | 40 | Comprehensive research, complex topics |
Each run creates or updates:
agent-knowledge/
CLAUDE.md # Master index (updated)
AGENTS.md # Master index for OpenCode/Codex (updated)
<topic-slug>.md # Synthesized learning guide
resources/
<topic-slug>-sources.json # Source metadata with quality scores
If a guide already exists for the topic, you are prompted to either update the existing guide with new sources or start fresh.
| Component | Type | Model | Role |
|---|---|---|---|
learn | command | - | Entry point, argument parsing |
learn-agent | agent | sonnet | Research coordination, web search, synthesis |
learn | skill | - | Research methodology, scoring rubric, templates |
agent-knowledge/ directory in the workspace (created automatically)MIT