本页还没有译文
本页 117 段还没译成简体中文,下面显示的是英文原文。Contributing to code
还没翻译译自英文(docs.langchain.com)本页 117 段都还没译
Code contributions are welcome! Whether you're fixing bugs, adding features, or improving performance, your contributions help deliver a better developer experience for thousands of developers.
Getting started
If you are looking for something to work on, check out the issue labeled "help wanted" in our repos:
Note
Before submitting large new features or refactors, please first open an issue or post to the forum for discussion. This ensures alignment with project goals and prevents duplicate work.
Quick fix: submit a bugfix
For simple bugfixes, you can get started immediately:
-
Reproduce the issue
Before even cloning the repository, ensure you can reliably reproduce the bug. This helps confirm the issue and provides a starting point for your fix. Maintainers and other contributors should be able to reproduce the issue based on your description without additional setup or modifications.
-
Fork the repository
Fork either the LangChain, LangGraph, or Deep Agents repo to your personal GitHub account
-
Clone and setup
git clone https://github.com/your-username/name-of-forked-repo.git # For instance, for LangChain: git clone https://github.com/parrot123/langchainjs.git # For LangGraph: git clone https://github.com/parrot123/langgraphjs.git# Inside your repo, install dependencies pnpm install # Create a build for all packages to resolve workspace dependencies pnpm build -
Create a branch
Create a new branch for your fix. This helps keep your changes organized and makes it easier to submit a pull request later.
git checkout -b your-username/short-bugfix-name -
Write failing tests
Add unit tests that will fail without your fix. This allows us to verify the bug is resolved and prevents regressions
-
Make your changes
Fix the bug while following our code quality standards. Make the minimal change necessary to resolve the issue. We strongly encourage contributors to comment on the issue before they start coding. For example:
"I'd like to work on this. My intended approach would be to [...brief description...]. Does this align with maintainer expectations?"
A 30-second comment often prevents wasted effort if your initial approach is wrong.
-
Run build
Run the build command to ensure the package still builds properly
pnpm build # or build a specific workspace package pnpm --filter @langchain/core build -
Verify the fix
Ensure that tests pass and no regressions are introduced. Ensure all tests pass locally before submitting your PR
pnpm lint pnpm test # For bugfixes involving integrations, also run: pnpm test:int # Or run tests in a specific workspace package cd libs/langchain-core pnpm test pnpm lint # Or run tests for a specific package from the root of the repo pnpm --filter @langchain/core test pnpm --filter @langchain/core lint -
Document the change
Update docstrings and/or inline comments if behavior changes
-
Submit a pull request
Follow the PR template provided. If applicable, reference the issue you're fixing using a closing keyword (e.g.
Fixes #ISSUE_NUMBER) so that the issue is automatically closed when your PR is merged.
Full development setup
For ongoing development or larger contributions:
- Review our contribution guidelines for features, bugfixes, and integrations
- Set up your environment following our setup guide below
- Understand the repository structure and package organization
- Learn our development workflow including testing and linting
Contribution guidelines
Before you start contributing to LangChain projects, take a moment to think about why you want to. If your only goal is to add a "first contribution" to your resume (or if you're just looking for a quick win) you might be better off doing a boot-camp or an online tutorial.
Contributing to open source projects takes time and effort, but it can also help you become a better developer and learn new skills. However, it's important to know that it might be harder and slower than following a training course. That said, contributing to open source is worth it if you're willing to take the time to do things well!
Backwards compatibility
Warning
Breaking changes to public APIs are not allowed except for critical security fixes.
See our versioning policy for details on major version releases.
Maintain compatibility via:
Stable interfaces
Always preserve:
- Function signatures and parameter names
- Class interfaces and method names
- Return value structure and types
- Import paths for public APIs
Safe changes
Acceptable modifications:
-
Adding new optional parameters/type parameters
-
Adding new methods to classes
-
Improving performance without changing behavior
-
Adding new modules or functions
Before making changes
-
Would this break existing user code?
-
Check if your target is public
-
Are there existing usage patterns in tests?
New features
We aim to keep the bar high for new features. We generally don't accept new core abstractions from outside contributors without an existing issue that demonstrates an acute need for them. This also applies to changes to infrastructure and dependencies.
In general, feature contribution requirements include:
-
Design discussion
Open an issue describing:
- The problem you're solving
- Proposed API design
- Expected usage patterns
-
Implementation
- Follow existing code patterns
- Include comprehensive tests and documentation
- Consider security implications
-
Integration considerations
- How does this interact with existing features?
- Are there performance implications?
- Does this introduce new dependencies?
We will reject features that are likely to lead to security vulnerabilities or reports.
Security guidelines
Warning
Security is paramount. Never introduce vulnerabilities or unsafe patterns.
Security checklist:
Input validation
-
Validate and sanitize all user inputs
-
Properly escape data in templates and queries
-
Never use
eval(), as this can lead to arbitrary code execution vulnerabilities
Error handling
- Use specific exception types
- Don't expose sensitive information in error messages
- Implement proper resource cleanup
Dependencies
- Avoid adding hard dependencies
- Keep optional dependencies minimal
- Review third-party packages for security issues
Development environment
Tip
Using an AI coding agent?
-
Install the LangChain Docs MCP servers to give your agent access to up-to-date LangChain documentation and examples.
Connect LangChain docs MCP servers
Connect both LangChain documentation MCP servers to my coding agent so it can look up current LangChain, LangGraph, and LangSmith docs and API reference.
Servers to add:
docs-langchain: https://docs.langchain.com/mcpreference-langchain: https://reference.langchain.com/mcp
Detect which agent or editor I am using (Claude Code, Cursor, Codex CLI, Claude Desktop, Deep Agents Code, VS Code, Antigravity, or another MCP-compatible client). Use the matching setup from https://docs.langchain.com/use-these-docs.md:
- Claude Code:
claude mcp add --transport httpfor each server (project scope by default; use--scope useronly if I ask for global access). - Codex CLI:
codex mcp addwith each server URL. - Cursor, Deep Agents Code, VS Code, or Antigravity: merge both entries into the MCP settings JSON using the field names shown on that page for my client.
- Claude Desktop: add both URLs under Settings > Connectors.
Do not invent alternate MCP URLs. After configuring, confirm both servers are listed and reachable.
-
Install LangChain Skills to improve your agent's performance on LangChain ecosystem tasks, then click the Copy page button on the top right of this page and paste the raw content into your agent to have it set up your environment automatically.
Install LangChain Skills
Install LangChain Skills for my coding agent so it can perform better on LangChain, LangGraph, and Deep Agents tasks.
Use the Agent Skills installer from https://github.com/langchain-ai/langchain-skills:
npx skills add langchain-ai/langchain-skills --skill '*' --yesIf I ask for a global install instead, use:
npx skills add langchain-ai/langchain-skills --skill '*' --yes --globalDetect which agent or editor I am using. If I use Claude Code and prefer the plugin path, follow the marketplace install from that repository README (
/plugin marketplace addthen/plugin install). Do not invent alternate skill package names or install URLs. After installing, confirm the skills are available to the agent.
Warning
Our JS/TS projects uses pnpm for dependency management. Make sure you have the latest version installed, or run corepack enable (on Node 24+) to setup the required pnpm version.
Info
We strive to keep setup consistent across all JS/TS packages. From the repo root, run:
pnpm install
pnpm --filter {package-name} test # Verify tests pass before starting developmentOnce you've reviewed the contribution guidelines, find the package directory for the component you're working on in the repository structure section below.
Repository structure
LangChainLangGraphDeep Agents
LangChain is organized as a monorepo with multiple packages:
Core packages
langchain(located inlibs/langchain/): Main package with chains, agents, and retrieval logic@langchain/core(located inlibs/langchain-core/): Base interfaces and core abstractions
Partner packages
Located in libs/providers/, these are independently versioned packages for specific integrations. For example:
@langchain/openai: OpenAI integrations@langchain/anthropic: Anthropic integrations@langchain/google: Google integrations
Supporting packages
@langchain/textsplitters: Text splitting utilities@langchain/standard-tests: Standard test suites for integrations@langchain/community: Community maintained integrations (separate repo)
LangGraph is organized as a monorepo with multiple Python packages:
Core packages
langgraph(located inlibs/langgraph/): Core framework for building stateful, multi-actor agentslanggraph-prebuilt(located inlibs/prebuilt/): High-level APIs for creating and running agents and tools
Checkpoint packages
langgraph-checkpoint(located inlibs/checkpoint/): Base interfaces for checkpoint saverslanggraph-checkpoint-postgres(located inlibs/checkpoint-postgres/): Postgres implementationlanggraph-checkpoint-sqlite(located inlibs/checkpoint-sqlite/): SQLite implementation
SDK and CLI
langgraph-sdk(located inlibs/sdk-py/): Python SDK for the Agent Server APIlanggraph-cli(located inlibs/cli/): Official command-line interface
Deep Agents is organized as a monorepo with multiple Python packages:
Core packages
deepagents(located inlibs/deepagents/): Core framework for building deep agents with planning, filesystem, and subagent capabilitiesdeepagents-code(located inlibs/code/): Deep Agents Code — interactive terminal interface with conversation resume, web search, and sandboxesdeepagents-cli(located inlibs/cli/): Deploy tooling (deepagents deploy,deepagents init,deepagents dev) for shipping agents to LangSmith Deployments
Integration packages
deepagents-evals(located inlibs/evals/): Evaluation suite and Harbor integration with LangSmith tracingdeepagents-acp(located inlibs/acp/): Agent Client Protocol integration
Development workflow
Pre-commit hooks
Running tests
Info
Directories are relative to the package you're working in.
We favor unit tests over integration tests when possible. Unit tests run on every pull request, so they should be fast and reliable. Integration tests run on a schedule and require more setup, so they should be reserved for confirming interface points with external services.
Unit tests
Location: src/tests/FILENAME_BEING_TESTED.test.ts
Unit tests cover modular logic that does not require calls to outside APIs. If you add new logic, you should add a unit test. In unit tests, check pre/post processing and mock external dependencies.
Requirements:
- No network calls allowed
- Test all code paths including edge cases
- Use mocks for external dependencies
To run unit tests:
# Run the entire test suite
pnpm test
# Or run a specific test file
pnpm test src/tests/FILENAME_BEING_TESTED.test.ts
# Or run a specific test function
pnpm test -t "the test that should be run"Integration tests
Location: src/tests/FILENAME_BEING_TESTED.int.test.ts
Integration tests cover logic that requires making calls to outside APIs (often integration with other services).
Integration tests require access to external services/provider APIs (which can cost money) and therefore are not run by default.
Not every code change will require an integration test, but keep in mind that we'll require/run integration tests separately as part of our review process.
Requirements:
- Test real integrations with external services
- Use environment variables for API keys
- Skip gracefully if credentials unavailable
To run integration tests:
pnpm test:intCode quality standards
Contributions must adhere to the following quality requirements:
Type hintsDocumentationCode style
Required: Complete types for all functions
function processDocuments(
docs: Document[],
processor: DocumentProcessor,
batchSize: number = 100
): ProcessingResult {
// ...
}Required: JSDocs for all exported functions and interfaces
/**
* Document processing instance.
*/
interface FooDocumentProcessor {
/**
* Process documents in batches.
*
* @param docs - List of documents to process.
* @returns Processing results with success/failure counts.
*/
process(docs: Document[]): ProcessingResult;
}
/**
* Process documents in batches.
*
* @param docs - List of documents to process.
* @param processor - Document processing instance.
* @param batchSize - Number of documents per batch.
* @returns Processing results with success/failure counts.
*/
export function processDocuments(
docs: Document[],
processor: DocumentProcessor,
batchSize: number = 100
): ProcessingResult {
// ...
}Automated: Formatting and linting:
pnpm lint # Check style and types
pnpm format # Apply formattingStandards:
- Descriptive variable names
- Break up complex functions (aim for fewer than 20 lines)
- Follow existing patterns in the codebase
Test writing guidelines
In order to write effective tests, there's a few good practices to follow:
- Encapsulate the test in a
describeblock that describes the component being tested - Use natural language to describe the test name
- Be exhaustive with assertions
- Only use snapshots for reasonably sized data objects
Unit testsIntegration testsMock usage
describe("DocumentProcessor", () => {
it("Should handle empty document list", () => {
const processor = new DocumentProcessor();
const result = processor.process([]);
expect(result.success).toBe(true);
expect(result.processedCount).toBe(0);
expect(result.errors).toHaveLength(0);
});
});describe("ChatOpenAI", () => {
it("Should test with real API", () => {
const chat = new ChatOpenAI();
const response = chat.invoke("Hello");
});
});describe("APIService", () => {
it("Should call with retry", () => {
const mockClient = new MockClient();
const service = new APIService(client: mockClient);
const result = service.callWithRetry();
});
});Submitting your PR
Once your tests pass and code meets quality standards:
- Push your branch and open a pull request
- Follow the provided PR template
- Reference related issues using a closing keyword (e.g.,
Fixes #123) - Wait for CI checks to complete
Note
If your PR includes AI-generated content, you must follow our acceptable uses of LLMs policy. PRs that appear to be low-effort, AI-generated spam will be closed without comment.
Warning
Address CI failures promptly. Maintainers may close PRs that do not pass CI within a reasonable timeframe.
Getting help
Our goal is to have the most accessible developer setup possible. Should you experience any difficulty getting setup, please ask in the community slack or open a forum post.
Check
You're now ready to contribute high-quality code to LangChain!