"Work on LangChain partner integration packages under libs/partners
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill integrations --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Integrations?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-integrations-12ffef88)More formats (shields.io, HTML) on the badges page.
---
name: integrations
description: "Work on LangChain partner integration packages under libs/partners
and provider model-profile data: package metadata, provider clients,
chat/LLM/embedding/vector store classes, credentials, tests, cassettes,
optional dependencies, and profile refresh workflows."
disable-model-invocation: true
metadata:
disco-role: operating
license: MIT
---
# Integrations
Use this sub-skill for practical work on LangChain-maintained partner packages and provider model-profile data. It covers provider-specific packages such as OpenAI, Anthropic, Ollama, Qdrant, Chroma, DeepSeek, Fireworks, Groq, Hugging Face, MistralAI, Nomic, Perplexity, OpenRouter, xAI, and Exa.
## When to Use
- The user asks to add, change, or debug a provider package under `libs/partners/<provider>`.
- The work touches provider chat models, LLMs, embeddings, vector stores, retrievers, tools, provider clients, middleware, output parsers, optional dependencies, credentials, or environment-variable handling.
- The user asks about package-local `pyproject.toml`, extras/dependency groups, import/version scripts, provider-specific unit or integration tests, VCR/cassette isolation, or model-profile refresh for provider data directories.
- The user needs safe validation that avoids network calls by default and escalates only when credentials, services, and permission are available.
## Route Elsewhere
- For package-agnostic `langchain-core` primitives, use `../core-primitives/SKILL.md`.
- For high-level `langchain` v1 agent construction, middleware routing, or `init_chat_model` behavior outside a provider package, use `../agents-and-middleware/SKILL.md`.
- For monorepo-wide package metadata, CI labels, release wiring, contributor conventions, or root development workflow, use `../monorepo-development/SKILL.md`.
- For implementing shared standard-test suites or model-profile CLI internals, route to `../testing-and-profiles/SKILL.md` when that sibling is present; this sub-skill only covers partner-side usage and provider data refresh.
## Reference Map
- Start with [references/provider-patterns.md](references/provider-patterns.md) for partner package layout, representative public classes, dependency patterns, source/test locations, edit workflows, and validation signals.
- Use [references/credentials-and-testing.md](references/credentials-and-testing.md) for API keys, local service checks, no-network unit tests, integration-test skip conditions, VCR cassette handling, and OpenAI Codex cassette safety.
- Use [references/model-profiles.md](references/model-profiles.md) for provider profile data directories, `langchain-profiles refresh`, augmentation files, generated profile diffs, and profile drift review.
- Use [references/troubleshooting.md](references/troubleshooting.md) for missing optional SDKs, credential failures, cassette isolation, model-profile drift, vector store service availability, and provider response/tool-call shape mismatches.
- Run [scripts/partner_package_check.py](scripts/partner_package_check.py) for a read-only metadata/layout check of a partner package when a Python interpreter is available.
## Fast Workflow
1. Identify the provider package, then work from its package directory: `libs/partners/<provider>`.
2. Inspect `pyproject.toml`, the `langchain_<provider>/` package root, `tests/unit_tests/`, `tests/integration_tests/`, and package-local `scripts/` before editing.
3. Keep provider-specific behavior in the partner package. Shared interfaces, message schemas, callbacks, base vector-store contracts, and standard-test framework changes belong to sibling skills.
4. Add deterministic unit tests for new behavior. Network-backed tests must be opt-in, credential-gated, and isolated from default no-network unit test runs.
5. Validate narrowly first with package-local `uv` commands when `uv` is available; do not use `pip`, `poetry`, `conda`, or ad-hoc virtual environments for this monorepo.
## Safe Validation
From a partner package directory, prefer targeted checks such as:
```bash
uv run --group test pytest tests/unit_tests/test_imports.py
uv run --group test pytest tests/unit_tests/test_chat_models.py
uv run --group test pytest tests/unit_tests/test_embeddings.py
uv run --group test pytest tests/unit_tests/test_vectorstores.py
uv run --group lint python scripts/check_imports.py $(find langchain_* -name '*.py')
uv run python scripts/check_version.py
```
Adjust test filenames to the package. Some packages use `make test`, `make check_imports`, `make check_version`, `make lint_package`, or `make type`; run these from the package directory only after confirming the package Makefile supports them. Skip network-heavy integration tests unless the user explicitly provides credentials or a local service and asks for those checks.
For a safe metadata/layout check that does not import provider SDKs or contact services:
```bash
python skills/langchain/sub-skills/integrations/scripts/partner_package_check.py libs/partners/openai
```
## Guardrails
- Preserve public provider class signatures and exported names unless the user explicitly accepts a breaking change.
- Do not add a provider SDK dependency to an unrelated package; partner SDKs belong in that partner package's metadata.
- Do not run credentialed examples, cassette recording, vector database services, or live provider calls without explicit user permission.
- Do not manually edit generated `_profiles.py` as the primary fix for profile drift; refresh through the model-profile workflow and review the generated diff.
- Do not link runtime instructions to source checkout docs, examples, or scripts; use the bundled references and scripts in this sub-skill.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!