Automate a full 1C change request from reading the existing config through writing a design doc, implementing code, and getting an AI audit verdict.
Watch multiple 1C agent sessions run simultaneously in separate Windows Terminal tabs and see each step in real time.
Use the auditor agent to review code changes made by the implementer before committing them to a git branch.
Requires 1C installed locally, Claude Code configured, and Windows Terminal with PowerShell on Windows 11.
This project is an orchestrator, meaning a coordinator that runs a series of AI agents in a specific order to complete a software development task. The README is written in Russian, so this explanation is based on a translation of its contents. The target domain is 1C, a popular enterprise accounting and business management platform widely used in Russia and surrounding countries. Modifying 1C projects involves reading configuration metadata, writing design documents, making code changes, and verifying the results. This orchestrator automates that pipeline by spinning up a chain of four AI agents: an analyst that reads the 1C configuration and produces a report, an SDD writer that converts the analysis into a formal design document, an implementer that makes the actual code changes and pushes them to a git branch, and an auditor that reviews what was implemented and issues a verdict. The architecture uses three levels. The user is the top level. A central Claude Code session acts as the router and task-package generator. Each specialized agent runs in its own tab of Windows Terminal, so you can see what each one is doing at the same time. The whole setup is designed for a Windows 11 laptop running PowerShell. All five phases of the pipeline are marked as shipped. PowerShell scripts handle spawning each agent tab, validating the outputs from each phase, and checking the status or stopping any agent mid-run. A project registry file lists the 1C projects you want to work on, each with a local path and a list of external tools the agents can call. Optional hooks are available for more autonomous behavior, such as automatically re-prompting agents and saving session state, but these are all turned off by default and require explicit configuration to enable. The project is released under the Apache 2.0 open-source license.
← arman-kudaibergenov on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.