AI toolkit for 1C-Bitrix development
bitrix-ai-toolkit
A set of rules, skills, linters and an MCP profile for AI development on 1C-Bitrix, with one core shared across Claude Code, Codex and Gemini CLI
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Hooks automatically run linters and fixes after every file edit by the agent
- The init_project.py and install_git_hooks.py scripts change git settings and project configuration
Install
Manual install
git clone https://github.com/vgtitov/bitrix-ai-toolkit && cd bitrix-ai-toolkit
sh onboard/install.shInstalls linters, git hooks, copies skills to ~/.claude/skills and builds the configs.
This is third-party code. Review the repository files before installing.
What it does
The toolkit addresses a real gap: Bitrix has no official IDE stubs or public core repository, so changes are checked against the real core code in vendor (bitrix-toolkit/bitrix-ci) and bxApiDocs stubs, not from the model's memory. A single source of truth in core holds the AGENTS.md rules, five skills (development, analyst, performance, admin/devops, tester), linter configs for PHPStan, PHP-CS-Fixer, ast-grep and Rector, and an MCP profile with bitrix24/mcp-rest-doc and Serena. The build.sh script generates per-agent configs from core: CLAUDE.md with PostToolUse hooks for Claude Code, AGENTS.md for Codex and Gemini CLI, and .rulesync for Cursor, Copilot and Cline. A core rule baked into the toolkit: distinguish Bitrix's procedural and D7 cores, write only into /local, and manage infoblock structure through sprint.migration rather than by hand in the database.
Who it is for. For 1C-Bitrix Site Management developers who want an agent to write production-ready code against the real core, rather than guessing the API from memory.
Good fit when
- You want an agent to write Bitrix code against real core signatures, not the model's outdated memory
- You need shared rules and skills across several agents at once: Claude Code, Codex, Gemini CLI
- You want PHPStan and PHP-CS-Fixer to run automatically after every code edit the agent makes
Not a fit when
- The project isn't on 1C-Bitrix Site Management: the rules and stubs are specific to this CMS
- You can't connect the vendor bitrix-ci repository with the real core code: without it the tool's main point is lost
- You need integration beyond Claude Code, Codex, Gemini CLI, Cursor, Copilot, Cline: the README calls Windsurf and Aider manual, not out of the box
Example request
Add a new catalog event handler in the /local module following D7 standards, and check it with PHPStan and the style linter before committingLimitations
Claude Code is stated as the reference implementation; other agents get minimal adaptation from the shared core, so hook and MCP coverage may vary. After updating the toolkit, a plain git pull isn't enough: sh onboard/install.sh --no-tools must be run again, or changes to hooks, skills and generated configs won't apply. The /local-only rule relies on discipline: edits to the /bitrix/modules core are technically possible but get overwritten on a Bitrix update. The tool is independent and not affiliated with 1C-Bitrix; the trademark belongs to its rights holder.
How to disable. Remove the generated CLAUDE.md, AGENTS.md, .mcp.json and .claude from the project, disable the git hooks via scripts/install_git_hooks.py, and delete the skills from ~/.claude/skills.
Security check
- Hooks automatically run linters and fixes after every file edit by the agent
- The init_project.py and install_git_hooks.py scripts change git settings and project configuration
README in short
The README explains the project's niche: Bitrix has no official stubs or public core, so the lever is real code in vendor plus local search and deterministic linter hooks. It describes the repository layout (core, adapters, build.sh, scripts, onboard, tests, demo, docs), key methodology principles (ask the tool, two cores, the /local gate, reuse-first), three integration channels with an agent (hooks, LSP, MCP), a table of supported agents, and a stack of free open-source static-analysis and refactoring tools. It gives the install via onboard/install.sh and a warning that an update needs the install rerun. MIT license, modeled after bsl-ai-toolkit.
FAQ
Does the toolkit only work with Claude Code?
No, build.sh generates configs for Claude Code, Codex and Gemini CLI out of the box, and Cursor, Copilot and Cline connect via rulesync.
Can the agent edit Bitrix's core code directly?
The gate rule requires writing only into /local; the /bitrix/modules core is treated as read-only and gets overwritten on a platform update.
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
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything