Community Скиллы Российские сервисы

Миграция с OpenAI на GigaChat

Cursor
Открыть источник ↗ GitHub ↗
Установка 
git clone (см. README)

Описание

migrate-to-gigachat-skill — скилл миграции для Cursor: определение стека — находит OpenAI-вызовы в проекте на Python, JS или TS, выбор адаптера — подбирает способ перехода на GigaChat под кодовую базу, переписывание вызовов — аккуратная замена эндпоинтов, моделей и параметров, проверка после замены — что прогнать, чтобы убедиться, что ничего не сломалось. Для команд, выполняющих требования о переходе на российские ИИ.

Что умеет:
— находить OpenAI-вызовы в коде;
— подбирать адаптер под стек;
— переписывать вызовы на GigaChat;
— давать чек-лист проверки после миграции.

Полная документация ниже — автоматически из репозитория asya02/migrate-to-gigachat-skill (ветка main). Оригинальные материалы проекта распространяются на условиях его лицензии.

migrate-to-gigachat-skill

A Cursor skill that automatically migrates Python / JS / TS projects from OpenAI to GigaChat.

Why

If a project uses OpenAI (directly via SDK, through LangChain/LangGraph, or via an OpenAI-compatible endpoint), this skill swaps the provider to GigaChat, picks the right adapter, and verifies the project still starts.

Migration paths

Stack Adapter Package
Python + LangChain / LangGraph Native integration langchain-gigachat
Python without LangChain (OpenAI SDK, etc.) gpt2giga proxy gpt2giga
JS/TS + LangChainJS / LangGraphJS Native integration langchain-gigachat (npm)
JS/TS without LangChain GigaChat SDK gigachat (npm)

See SKILL.md (sections Path A–D) for detailed instructions on each path.

Project structure

SKILL.md                                  # Skill definition (consumed by the Cursor agent)
scripts/
  detect_stack.py                         # Detects the stack and suggests a migration path
  collect_cred_requirements.py            # Extracts env vars from README/.env/docs
  run_verify.py                           # Baseline/after verification (tests, lint, smoke)
  live_canary_openai_via_proxy.py         # Minimal script to smoke-test the gpt2giga proxy
examples/
  gpt2giga.env.example                   # Template .env for gpt2giga proxy

Quick start

1. Install the skill into your project

Copy the repository contents into .cursor/skills/migrate-to-gigachat-skill/ of the target project:

mkdir -p .cursor/skills/migrate-to-gigachat-skill
cp -r /path/to/this/repo/* .cursor/skills/migrate-to-gigachat-skill/

2. Detect the stack

python3 .cursor/skills/migrate-to-gigachat-skill/scripts/detect_stack.py

Outputs a JSON report: detected languages/frameworks and the recommended migration path (A/B/C/D).

3. Collect env-variable requirements

python3 .cursor/skills/migrate-to-gigachat-skill/scripts/collect_cred_requirements.py

Shows which env variables are mentioned in the README, .env.example, docker-compose, and docs.

4. Configure GigaChat credentials

Set environment variables locally (never paste secrets into the agent chat):

export GIGACHAT_CREDENTIALS="your_key"
export GIGACHAT_SCOPE="GIGACHAT_API_PERS"   # or GIGACHAT_API_B2B / GIGACHAT_API_CORP

For the gpt2giga proxy, use a dedicated env file (see examples/gpt2giga.env.example):

cp examples/gpt2giga.env.example .env.gpt2giga
# fill in the values
gpt2giga --env-path .env.gpt2giga

5. Baseline verification

VERIFY_MODE=baseline python3 .cursor/skills/migrate-to-gigachat-skill/scripts/run_verify.py

6. Migration

Ask the Cursor agent: "Migrate this project from OpenAI to GigaChat." The agent picks up SKILL.md and follows the migration instructions.

7. Post-migration verification

VERIFY_MODE=after python3 .cursor/skills/migrate-to-gigachat-skill/scripts/run_verify.py

Strict mode (tests and lint are required to pass):

VERIFY_STRICT=1 VERIFY_MODE=after python3 .cursor/skills/migrate-to-gigachat-skill/scripts/run_verify.py

Live canary (one real model call):

VERIFY_LIVE_CMD='your_command_here' VERIFY_MODE=after python3 .cursor/skills/migrate-to-gigachat-skill/scripts/run_verify.py

run_verify.py environment variables

Variable Description Default
PROJECT_ROOT Root of the target project .
VERIFY_MODE baseline or after baseline
VERIFY_STRICT 1 — lint and tests are required to pass 0
VERIFY_TIMEOUT_S Per-step timeout (seconds) 45
VERIFY_LIVE_CMD Command for the live canary check —
VERIFY_LIVE_REQUIRED 1 — canary must pass 0

Security

  • Secrets (API keys, tokens, certificates) are never committed.
  • .gitignore excludes .env / .env.* files.
  • The agent never asks you to paste secrets into chat — only to confirm that the variables are set.
  • For gpt2giga, a separate .env.gpt2giga is recommended to isolate proxy credentials from the application’s .env.

References