LangChain

Avoid common LangChain mistakes — LCEL gotchas, memory persistence, RAG chunking, and output parser traps.

Security Scan

This is an instruction-only LangChain best-practices skill that asks for nothing sensitive and whose guidance is consistent with its stated purpose.

Purpose & Capability

Name/description (LangChain gotchas and best practices) matches the content of SKILL.md. The only declared runtime requirement is python3, which is reasonable for LangChain-related advice; no unexpected credentials, binaries, or config paths are requested.

Instruction Scope

SKILL.md contains high-level usage guidance and warnings only — it does not instruct the agent to read files, call external endpoints, or access credentials. Note: instructions are advisory; they could be used to guide actions if an agent is later asked to execute code, but the skill itself does not command execution or data collection.

Install Mechanism

No install spec and no code files are present, so nothing is written to disk or downloaded. This is the lowest-risk install profile (instruction-only).

Credentials

The skill declares no required environment variables, credentials, or config paths. There is no disproportionate request for secrets or unrelated service tokens.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request permanent presence or elevated platform privileges. Autonomous model invocation remains enabled by platform default but is not a special property of this skill.

Assessment

This skill is a read-only guide for LangChain best practices and appears coherent and low-risk: it asks for nothing sensitive and contains only advisory text. Before installing, confirm you trust the skill source (source/homepage are unknown). Because it's instruction-only, there is no code to execute now, but if the skill is later updated to include install steps or code files, re-check those for downloads, required credentials, or instructions that run arbitrary commands. If you plan to have an agent execute LangChain code using these tips, ensure the agent's execution environment (python3, installed packages) and access to data/credentials are controlled and limited to what you expect.

LCEL Basics

• | pipes output to next — prompt | llm | parser
• RunnablePassthrough() forwards input unchanged — use in parallel branches
• RunnableParallel runs branches concurrently — {"a": chain1, "b": chain2}
• .invoke() for single, .batch() for multiple, .stream() for tokens
• Input must match expected keys — {"question": x} not just x if prompt expects {question}

Memory Gotchas

• Memory doesn't auto-persist between sessions — save/load explicitly
• ConversationBufferMemory grows unbounded — use ConversationSummaryMemory for long chats
• Memory key must match prompt variable — memory_key="chat_history" needs {chat_history} in prompt
• return_messages=True for chat models — False returns string for completion models

RAG Chunking

• Chunk size affects retrieval quality — too small loses context, too large dilutes relevance
• Chunk overlap prevents cutting mid-sentence — 10-20% overlap typical
• RecursiveCharacterTextSplitter preserves structure — splits on paragraphs, then sentences
• Embedding dimension must match vector store — mixing models causes silent failures

Output Parsers

• PydanticOutputParser needs format instructions in prompt — call .get_format_instructions()
• Parser failures aren't always loud — malformed JSON may partially parse
• OutputFixingParser retries with LLM — wraps another parser, fixes errors
• with_structured_output() on chat models — cleaner than manual parsing for supported models

Retrieval

• similarity_search returns documents — .page_content for text
• k parameter controls results count — more isn't always better, noise increases
• Metadata filtering before similarity — filter={"source": "docs"} in most vector stores
• max_marginal_relevance_search for diversity — avoids redundant similar chunks

Agents

• Agents decide tool order dynamically — chains are fixed sequence
• Tool descriptions matter — agent uses them to decide when to call
• handle_parsing_errors=True — prevents crash on malformed agent output
• Max iterations prevents infinite loops — max_iterations=10 default may be too low

Common Mistakes

• Prompt template variables case-sensitive — {Question} ≠ {question}
• Chat models need message format — ChatPromptTemplate, not PromptTemplate
• Callbacks not propagating — pass config={"callbacks": [...]} through chain
• Rate limits crash silently sometimes — wrap in retry logic
• Token count exceeds context — use trim_messages or summarization for long histories