Langchain Py Pack

> Pain-first Claude Code skills for LangChain 1.0 and LangGraph 1.0 in Python. > Every skill opens with a specific production failure mode — not capability prose.

Why This Pack Exists

LangChain 1.0 and LangGraph 1.0 shipped October 2025 with a new middleware model, typed content blocks (text/tool_use/image), stable checkpointing, first-class human-in-the-loop interrupts, three streaming modes, and native OpenTelemetry export. Token accounting, structured output, agent control flow, and memory all behave differently than they did in the 0.2 / 0.3 era.

This pack replaces the legacy langchain-pack with pain-first, Python-native skills. Every skill opens with a specific failure mode — a real exception, a hardcoded threshold, a version-specific regression — not capability prose:

> ChatAnthropic.stream() blocks llmOutput token counts until the stream completes. > Live cost dashboards built on that field lag by stream_duration seconds — > sometimes by 20+ seconds on long responses. Use astream_events(version="v2") > or a callback handler to read tokens incrementally.

See docs/pain-catalog.md for the full 68-entry catalog of LangChain 1.0 pain points that every skill in this pack anchors to.

Installation

/plugin install langchain-py-pack@claude-code-plugins-plus

TypeScript Counterpart

For LangChain.js + @langchain/langgraph in Node 22+ / Vercel / Cloud Run: langchain-ts-pack (Epic B — not yet shipped). Same 34-skill taxonomy, JS-native.

What Is Shipped Today

Both skills are gold-standard quality (≥200 lines each, 3-4 references per skill, every example pinned to langchain-core 1.0.x, A-grade at the 100-point rubric).

Roadmap — 32 More Skills Landing Across Four Epics

Each epic is one PR. Every skill follows the same gold-standard quality bar as the two shipped above: concrete pain in the Overview, ≥2 error codes named, 2-4 references, decision trees / comparison tables where applicable, and every code block pinned to LangChain 1.0.x.

Epic A2 — Getting Started + Core Workflows (S01-S08 minus the two shipped)

Epic A3 — Operations + Pro (S09-P20)

Epic A4 — Flagship + LangGraph v1.0 (F21-L34)

Quick Start

1. Install the pack

/plugin install langchain-py-pack@claude-code-plugins-plus

2. Install LangChain 1.0 + LangGraph 1.0

python -m venv .venv && source .venv/bin/activate

pip install "langchain>=1.0,<2.0" "langchain-core>=1.0,<2.0" \
            "langchain-anthropic>=1.0,<2.0" \
            "langgraph>=1.0,<2.0"

3. A minimal agent with memory

from langchain_anthropic import ChatAnthropic
from langgraph.prebuilt import create_react_agent
from langgraph.checkpoint.memory import MemorySaver

llm = ChatAnthropic(model="claude-sonnet-4-6", temperature=0)

def add(a: int, b: int) -> int:
    """Add two integers."""
    return a + b

agent = create_react_agent(
    model=llm,
    tools=[add],
    checkpointer=MemorySaver(),
)

config = {"configurable": {"thread_id": "demo-1"}}
result = agent.invoke(
…

Key LangChain 1.0 / LangGraph 1.0 Links

• LangChain 1.0 release notes
• LangChain Python docs
• LangGraph overview
• LangGraph streaming modes
• astream_events v2
• Checkpointing and persistence
• Human-in-the-loop patterns
• LangSmith
• State of Agent Engineering 2026

Version Baseline

Every skill in this pack is pinned to langchain-core 1.0.x / langgraph 1.0.x. If you are on 0.2.x or 0.3.x, wait for langchain-upgrade-migration (Epic A4) or consult the legacy langchain-pack in the meantime.

License

MIT