new·The score now tells you which way it movedA brain's exam only ever grows: its own material writes questions, and so does every question a real caller asked and did not get answered. The score is a percentage over that growing set, so a brain that learned more could post a smaller number — and this week three did. One of them answered two MORE questions than the week before and showed eighteen points less. Printed as a single percentage, that reads as decline to a reader and as punishment to anyone who contributes material.all news →
mozg.beta
Sign in

explore / AI & agents / mozg / LangChain & LangGraph

LangChain · Agents

Answer questions about LangChain as documented today: building agents, model initialisation and message types, tool calling, structured output, streaming, context engineering and the middleware and retrieval patterns.

1,355 notes · updated 2026-08-14 · trained 80%

since last sitting: +41 newly passed — this brain is learning · the verified changelog →

use this brain
# sign in to get a token
$ claude mcp add --transport http mozg \
https://mozg.sh/mcp --header "Authorization: Bearer …"
> use mozg/langchain-core — …
Sign in — get this command with your token in it

New here? Sign in and this exact command appears with your token already inside — copy, paste into your terminal, done.

Or just add it

Sign in and add it, and every agent you have connected can read it — nothing to download, and it stays current as the author updates it.

Sign in to add it

Ask it things like

answers 41/52 on its latest exam
What is the minimum code needed to create a basic LangChain agent with OpenAI in Python?Agent Building Fundamentals
When should you use the subagents pattern vs a router pattern?Advanced Multi-Agent Patterns
What is the handoffs architecture and when do you use it?Advanced Multi-Agent Patterns
What is the difference between Deep Agents and create_agent?Agent Building Fundamentals
What is the difference between thread_id and context?Context Engineering
What object types does stream.messages yield?Streaming
How do you implement a subgraph handoff with context engineering in Python?Advanced Multi-Agent Patterns
What does ToolRuntime provide access to?Context Engineering

What it can answer

52 coverage checks
80%
Answers 41 of the exam's 52 questionsIt answers 41 questions, 41 more than last time, out of an exam that grew by 52.Nothing it used to answer has stopped working.
Tool Calling & Structured Output6 / 7
Advanced Multi-Agent Patterns5 / 5
Context Engineering4 / 5
Error Handling & Troubleshooting4 / 4
Streaming4 / 5
Structured Output & Responses4 / 4
Agent Building Fundamentals3 / 4
Middleware & Retrieval3 / 5
Model Initialization & Messages3 / 5
Testing & Deployment3 / 5
Gap: Reasoning Tokens1 / 1
LangSmith Integration1 / 2
exam badge — share this score →

Attacks survived

re-run weekly · 2026-08-10
gate-battery12/12 known hostile payloads caught by the ingest gate
injection-corpusevery active note scanned — no steering language
secret-corpusevery active note scanned — no keys or credentials

heuristic scans against known attack classes — measured and dated, not a promise of unpoisonability

Inside

  • Documentation repository structure · common patterns
  • Product and feature name capitalization rules · common patterns
  • Model references should use latest GA models · model initialization & messages
  • Integration page description format · common patterns
  • Frontmatter requirements for MDX files · common patterns
  • Mermaid diagram color palette for documentation · common patterns
  • Content block types use .content_blocks not .content · model initialization & messages
  • Code examples must be tested before publishing · testing & deployment
  • Release stage names capitalization rules · common patterns
  • Language-specific content uses :::python or :::js fences · common patterns
  • API reference links use @[ClassName] syntax · common patterns
  • Reusable snippets import form required · common patterns
  • Tabler icons are the only approved icon library · common patterns
  • Deep Agents eval matrix script and maintenance · testing & deployment

Licence

CC BY-NC-SA 4.0

Use it, copy it, build on it, with credit. Selling it is not allowed.