GigaChat Skills

Install

Copy the skill folders from github.com/trashchenkov/gigachat-skills into the project skills directory, keeping references inside each folder. Start with gigachat-navigation and gigachat-setup.

This is third-party code. Review the repository files before installing.

What it does

A set of portable folders with SKILL.md and references that teaches an agent to build GigaChat integrations. gigachat-navigation helps choose between the official SDK, langchain-gigachat, the gpt2giga proxy and deepagents-gigachat. The other skills cover credentials, OAuth scopes and TLS certificates, chat and streaming, function calling, files and embeddings, LangChain RAG and OpenAI client compatibility.

Who it is for. For Python developers connecting GigaChat to their applications or agents.

Example request

Write a Python service that answers questions over a document base using langchain-gigachat

Limitations

There is no universal install path; copy the folders into your agent's skills directory or adapt them to its instruction format. The verification scripts are smoke tests, not a full benchmark. File requests must not mix modalities in one call.

How to disable. Remove the copied skill folders from your agent's directory.

Security check

• The skills are instructions and references
• Smoke scripts call the API only with --live

README in short

The README describes tool-agnostic skills for the GigaChat API and GigaChain. Each folder holds a SKILL.md and on-demand references. The verification folder has smoke scripts that skip live calls by default and run with --live. It notes that gpt2giga is a compatibility layer, not a byte-for-byte OpenAI API replacement.

SKILL.md

---
name: gigachat-navigation
description: >-
  Routes work across the GigaChat ecosystem: official gigachat SDK,
  langchain-gigachat, gpt2giga, and deepagents-gigachat. Use when deciding which layer to choose,
  how capabilities differ, which limitations are smoke-covered or source-backed, and which skill or
  reference path to use next.
---

# GigaChat Navigation

Use this skill first when the user is unsure which GigaChat integration layer fits the task.

## What this skill covers

- choosing between `gigachat`, `langchain-gigachat`, `gpt2giga`, and `deepagents-gigachat`
- mapping user intent to the right skill
- understanding smoke-covered capabilities and limitations
- choosing the right implementation skill next

## Workflow

1. Classify the task by integration style, not by API buzzwords.
2. If the user is writing a native Python integration, prefer `gigachat`.
3. If the app already uses LangChain primitives, use `langchain-gigachat`.
4. If the client must keep an OpenAI-compatible or Anthropic-compatible SDK, use `gpt2giga`.
5. If the app is a Deep Agents or `deepagents-code` harness using GigaChat, use `deepagents-gigachat`.
6. If more than one layer could work, choose the simplest source-backed path.
7. For files, keep one request limited to one modality.
8. Prefer the simplest smoke-covered or source-backed path described in the relevant skill and references.

## Read these references as needed

- For routing logic: `references/decision-rules.md`
- For smoke-covered feature coverage: `references/feature-matrix.md`

## Default output

- name the recommended layer explicitly
- state why the other layers are not the default
- point to the next implementation skill or reference file

## Boundaries

- This skill helps choose and navigate.
- It does not replace the implementation skills for SDK chat, files, LangChain, or proxy work.

Related