langchain

Name: langchain
Author: sordi-ai

17stars3forks

How to install langchain?

sordi-ai/skill-everything/langchain

$npx -y skills add sordi-ai/skill-everything --skill langchain

Installs into the current project.

Summary

Apply when building LangChain pipelines, LCEL chains, agents, or retrieval-augmented generation systems.

Files · 1

View on GitHub

SKILL.md

1	# Sub-Skill: LangChain / Agent Framework Conventions
2
3	Purpose: Prevent common LangChain mistakes — deprecated chain classes, missing retry/timeout guards, unsafe prompt handling, and unobservable pipelines.
4
5	---
6
7	## Rules
8
9	### Chain Construction
10
11	1. Use LCEL pipe syntax. Always use the LCEL pipe operator (`\|`) to compose runnables instead of deprecated constructor-based chain classes (`LLMChain`, `SequentialChain`, `TransformChain`). Reference: ERR-2026-026
12	2. Avoid legacy chain imports. Never import from `langchain.chains.llm` or `langchain.chains.sequential`; use `langchain_core.runnables` and `langchain_core.prompts` instead.
13	3. Prefer RunnablePassthrough for identity steps. Use `RunnablePassthrough` to thread context through a chain without mutation rather than writing a lambda that returns its input unchanged.
14	4. Use RunnableParallel for fan-out. Prefer `RunnableParallel` over manually calling multiple chains and merging dicts; it expresses intent and enables parallel execution.
15
16	### Prompts & Output Parsers
17
18	5. Use typed output parsers. Always attach an output parser (`PydanticOutputParser`, `JsonOutputParser`, `StrOutputParser`) to chains that produce structured data; never parse raw LLM strings manually downstream.
19	6. Inject format instructions via partial. Use `prompt.partial(format_instructions=parser.get_format_instructions())` to bind parser instructions into the prompt template rather than hard-coding them in the template string.
20	7. Separate system and human messages. Use `ChatPromptTemplate.from_messages([("system", ...), ("human", ...)])` instead of a single `PromptTemplate` for chat models; mixing roles in one string breaks structured output.
21
22	### Chat Models vs LLMs
23
24	8. Prefer ChatModel over LLM. Always use `ChatOpenAI`, `ChatAnthropic`, or equivalent chat-model classes for new code; the base `OpenAI` LLM class is deprecated for most use cases and lacks tool-calling support.
25	9. Pin model name explicitly. Never rely on the default model name in a chat model constructor; always pass `model="gpt-4o"` (or equivalent) so upgrades are intentional.
26
27	### Memory & State
28
29	10. Use RunnableWithMessageHistory for stateful chains. Prefer `RunnableWithMessageHistory` over manual history management or deprecated `ConversationChain`; it integrates cleanly with LCEL and supports async.
30	11. Scope memory by session ID. Always pass a `session_id` key when constructing `RunnableWithMessageHistory` to prevent cross-user memory leakage in multi-tenant services.
31
32	### Tools & Agents
33
34	12. Define tools with @tool decorator. Use the `@tool` decorator (or `StructuredTool.from_function`) with a typed signature and docstring; never pass raw callables to an agent without a schema.
35	13. Use create_tool_calling_agent for modern agents. Prefer `create_tool_calling_agent` + `AgentExecutor` over deprecated `initialize_agent`; it uses native tool-calling APIs and avoids ReAct string parsing.
36	14. Cap agent iterations. Always set `max_iterations` and `max_execution_time` on `AgentExecutor` to prevent runaway loops; default is unbounded.
37
38	### Retrieval & Vector Stores
39
40	15. Use retrieval chains via LCEL. Build RAG pipelines with `retriever \| format_docs \| prompt \| llm \| parser` rather than `RetrievalQA.from_chain_type`; the latter is deprecated and hides the retrieval step.
41	16. Ensure document loaders are lazy. Prefer `.lazy_load()` over `.load()` for large corpora to avoid loading all documents into memory at once.
42
43	### Reliability & Observability
44
45	17. Wrap LLM calls with retry. Use `.with_retry(stop_after_attempt=3, wait_exponential_jitter=True)` on any runnable that calls an external API; never let transient rate-limit errors propagate uncaught.
46	18. Set request timeout. Always pass `request_timeout` (or `timeout`) to chat model constructors; omitting it allows indefinitely hanging requests.
47	19. Attach callbacks for observability. Use `callbacks=[LangSmithTracer()]` or equivalent on chains in production; never ship a pipeline with no tracing so failures are diagnosable.
48	20. Count tokens before sending. Before sending large contexts, use `llm.get_num_tokens(text)` or a tiktoken counter to verify the payload fits within the model's context window.
49
50	### Security
51
52	21. Sanitize user input before prompt injection. Never interpolate raw user strings directly into system prompts; use a dedicated input variable in the prompt template and validate/strip control characters before binding.
53	22. Disable dangerous tools in untrusted contexts. Avoid giving agents tools with filesystem or shell access when processing untrusted input; scope tool permissions to the minimum requi

People who liked this also viewed

6 picks

Type

hyperframes-cliUse the HyperFrames CLI development loop: init, add, catalog, capture, lint, check, snapshot, compare, grade-compare, preview, play, present, beats, keyframes, single or batch render, publish, cloud, SkillsJul 2026262k37k
request-refactor-planSkills for Real Engineers. Straight from my .agents directory.SkillsJul 2026165k181k
analyze-projectRigor Analyze / Rigor Audit read-only skill for deep learning research repositories. Use when the user wants to read and understand a repository, inspect model structure and training or inference entrSkillsMay 2026161k507
explore-codeRigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated braSkillsMay 2026161k507
ubiquitous-languageSkills for Real Engineers. Straight from my .agents directory.SkillsJul 2026160k181k
scaffold-exercisesSkills for Real Engineers. Straight from my .agents directory.SkillsJul 2026149k181k

langchain

17stars3forks

View on GitHub

How to install langchain?

sordi-ai/skill-everything/langchain

$npx -y skills add sordi-ai/skill-everything --skill langchain

Installs into the current project.

Summary

Apply when building LangChain pipelines, LCEL chains, agents, or retrieval-augmented generation systems.

Files · 1

View on GitHub

SKILL.md

1	# Sub-Skill: LangChain / Agent Framework Conventions
2
3	Purpose: Prevent common LangChain mistakes — deprecated chain classes, missing retry/timeout guards, unsafe prompt handling, and unobservable pipelines.
4
5	---
6
7	## Rules
8
9	### Chain Construction
10
11	1. Use LCEL pipe syntax. Always use the LCEL pipe operator (`\|`) to compose runnables instead of deprecated constructor-based chain classes (`LLMChain`, `SequentialChain`, `TransformChain`). Reference: ERR-2026-026
12	2. Avoid legacy chain imports. Never import from `langchain.chains.llm` or `langchain.chains.sequential`; use `langchain_core.runnables` and `langchain_core.prompts` instead.
13	3. Prefer RunnablePassthrough for identity steps. Use `RunnablePassthrough` to thread context through a chain without mutation rather than writing a lambda that returns its input unchanged.
14	4. Use RunnableParallel for fan-out. Prefer `RunnableParallel` over manually calling multiple chains and merging dicts; it expresses intent and enables parallel execution.
15
16	### Prompts & Output Parsers
17
18	5. Use typed output parsers. Always attach an output parser (`PydanticOutputParser`, `JsonOutputParser`, `StrOutputParser`) to chains that produce structured data; never parse raw LLM strings manually downstream.
19	6. Inject format instructions via partial. Use `prompt.partial(format_instructions=parser.get_format_instructions())` to bind parser instructions into the prompt template rather than hard-coding them in the template string.
20	7. Separate system and human messages. Use `ChatPromptTemplate.from_messages([("system", ...), ("human", ...)])` instead of a single `PromptTemplate` for chat models; mixing roles in one string breaks structured output.
21
22	### Chat Models vs LLMs
23
24	8. Prefer ChatModel over LLM. Always use `ChatOpenAI`, `ChatAnthropic`, or equivalent chat-model classes for new code; the base `OpenAI` LLM class is deprecated for most use cases and lacks tool-calling support.
25	9. Pin model name explicitly. Never rely on the default model name in a chat model constructor; always pass `model="gpt-4o"` (or equivalent) so upgrades are intentional.
26
27	### Memory & State
28
29	10. Use RunnableWithMessageHistory for stateful chains. Prefer `RunnableWithMessageHistory` over manual history management or deprecated `ConversationChain`; it integrates cleanly with LCEL and supports async.
30	11. Scope memory by session ID. Always pass a `session_id` key when constructing `RunnableWithMessageHistory` to prevent cross-user memory leakage in multi-tenant services.
31
32	### Tools & Agents
33
34	12. Define tools with @tool decorator. Use the `@tool` decorator (or `StructuredTool.from_function`) with a typed signature and docstring; never pass raw callables to an agent without a schema.
35	13. Use create_tool_calling_agent for modern agents. Prefer `create_tool_calling_agent` + `AgentExecutor` over deprecated `initialize_agent`; it uses native tool-calling APIs and avoids ReAct string parsing.
36	14. Cap agent iterations. Always set `max_iterations` and `max_execution_time` on `AgentExecutor` to prevent runaway loops; default is unbounded.
37
38	### Retrieval & Vector Stores
39
40	15. Use retrieval chains via LCEL. Build RAG pipelines with `retriever \| format_docs \| prompt \| llm \| parser` rather than `RetrievalQA.from_chain_type`; the latter is deprecated and hides the retrieval step.
41	16. Ensure document loaders are lazy. Prefer `.lazy_load()` over `.load()` for large corpora to avoid loading all documents into memory at once.
42
43	### Reliability & Observability
44
45	17. Wrap LLM calls with retry. Use `.with_retry(stop_after_attempt=3, wait_exponential_jitter=True)` on any runnable that calls an external API; never let transient rate-limit errors propagate uncaught.
46	18. Set request timeout. Always pass `request_timeout` (or `timeout`) to chat model constructors; omitting it allows indefinitely hanging requests.
47	19. Attach callbacks for observability. Use `callbacks=[LangSmithTracer()]` or equivalent on chains in production; never ship a pipeline with no tracing so failures are diagnosable.
48	20. Count tokens before sending. Before sending large contexts, use `llm.get_num_tokens(text)` or a tiktoken counter to verify the payload fits within the model's context window.
49
50	### Security
51
52	21. Sanitize user input before prompt injection. Never interpolate raw user strings directly into system prompts; use a dedicated input variable in the prompt template and validate/strip control characters before binding.
53	22. Disable dangerous tools in untrusted contexts. Avoid giving agents tools with filesystem or shell access when processing untrusted input; scope tool permissions to the minimum requi

People who liked this also viewed

6 picks

Type