1C BSL Agent Development Framework
A modular agent development framework for 1C: skills, rules, subagents and SDD and TDD workflows with cross-review
Install
python tools/install.pyRun from the framework folder. The installer asks for the IDE, project folder and files. For Claude and Codex, create a root CLAUDE.md or AGENT.md.
This is third-party code. Review the repository files before installing.
What it does
The framework configures agents to work with 1C configurations through MCP servers. In subagent mode an orchestrator runs the full cycle: spec, technical requirements, unit tests, code, test runs in a Docker sandbox and review by a second agent. Skills are bound to capabilities rather than specific MCP servers and are linked into projects via symlinks by an installer, so one edit applies everywhere. Skills are written in Russian, and projects get an automatically synced English mirror.
Who it is for. For 1C developers and teams who want a repeatable multi-agent development process.
Good fit when
- You want agents to take 1C tasks from spec to tested code
- You need the same skills and rules across several 1C projects
- You want decisions reviewed by a second model, for example via Codex
Not a fit when
- You need one-off help with a single data processor without infrastructure setup
- You cannot deploy the sandbox and MCP servers
Example request
Add a Purchase Request document that posts to the stock balance register, starting with a spec and testsLimitations
The full cycle requires a Docker sandbox, a 1C server side, PostgreSQL and Apache, plus the MCP servers listed in docs/info/mcp.md. The author notes that not all scripts are cross-platform, multi-agent mode is untested and Vanessa GUI testing is unfinished. Some docs in docs/specs-and-analisys may be outdated.
How to disable. Run python tools/install.py from the framework folder and choose uninstall for the project. This removes the symlinks without touching the sources.
Security check
- Assumes autonomous agent work on code and databases in a Docker sandbox
- Requires installing several third-party MCP servers
- The installer creates symlinks in project and IDE folders
README in short
The README describes an agent development framework for 1C BSL with modular installation of skills, rules, subagents and workflows via symlinks. Infrastructure is a client sandbox and a VM for server services from the separate 1c-ai-sandbox-client-server repository. The workflow starts from the framework-bootstrap rule and the full-cycle and orchestrator workflows. The project accepts only open source, free solutions. MIT licensed.
SKILL.md
--- name: framework-bootstrap description: "На старте или компакте загрузить профиль оркестратора" alwaysApply: true --- # 1C BSL Agent Development Framework Это фреймворк агентной разработки для 1С BSL. Минимальный always-on ориентир + портативный кросс-харнес мост, поднимающий роль Lead/оркестратора в главном потоке. Детальный роутинг и манинг оркестрации живут в **профиле оркестратора** (`framework/subagents/orchestrator.md`), а не здесь. ## Само-промотирующий стаб (главный поток) ``` - Если ты САБАГЕНТ (системный промпт явно называет роль: analyst / architect / developer-code / developer-tests / scenario-author / scenario-coder / tester / reviewer / explorer / debugger) → это правило НЕ для тебя. ВЫХОД. - Если тело манинга оркестратора СЕЙЧАС в контексте → ты — Lead. Действуй по нему: классифицируй задачу → выбери цикл (short/full) → дальше по профилю. - Иначе → ПРОЧИТАЙ СЕЙЧАС framework/subagents/orchestrator.md, затем действуй как Lead. ``` ## Куда идти (короткая карта) - Простая задача (баг в одном файле, < 20 строк, без новых фич, без новых объектов метаданных) → short-цикл: навык **`quick-fix`** (через Skill-тул). - Средняя / сложная (новые фичи, несколько файлов, архитектура, новые объекты метаданных) → full-цикл: работа оркестратора по профилю + фазовая механика **`full-cycle`**. - **Если сомневаешься** — трактуй как сложную (full). ## Инструменты - Агент обнаруживает доступные инструменты динамически через MCP (`tools/list`) — не hardcode имена tool-ов
FAQ
Why do projects get English files?
Cyrillic costs more tokens, so symlinks point to the English mirror while the Russian version is edited.
Which agents does the author use?
Claude CLI for specs and code, Codex CLI as reviewer, and Cursor as the shell.
Related
A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything
Up-to-date, version-specific library docs and code examples in your agent's context, via MCP or a CLI plus skill