claude-code-skills-1c

What the skill set does

The README describes a set of skills, rules and commands for Claude Code that let an agent edit 1C:Enterprise by the platform format instead of text search across files. The skills give the agent ready made operations: assemble a metadata object, a form, a role, a data composition schema, an extension, or an external processing; and take them back apart. The rules dictate how the agent does this, covering BSL standards, anti patterns, EDT requirements, and work with the standard subsystem library. The scripts edit the XML dump directly, so basic operations run without a launched platform and without the Configuration Extension tool. Alongside the skills, the set ships API references so the agent does not guess a method name from memory but takes it from a reference and checks the call before writing code. The BSL reference covers 2624 methods in 284 modules of the standard subsystem library versions 3.1.11 and 3.2.1, and a payroll reference covers 1C:ZUP staffing, average earnings, accruals, benefits and reporting. The README targets a 1C developer who keeps sources under git and wants to hand routine work to an agent rather than typing every object by hand. The target user is a 1C developer who keeps sources under git and wants to hand routine object work to an agent instead of typing each one by hand.

Composition and rules

The README breaks the set into groups with skill counts: a router that picks the right skill, configuration tasks such as create or analyze, metadata objects, managed forms, extensions, processings and reports, database and storage work, roles, subsystems, layouts, data composition schemas, XDTO, web publishing and testing, the standard subsystem library and 1C 7.7, and reference or utility skills. In total the README states 117 skills and 40 rules. The rules attach once and apply to every task without repeating them in each prompt, covering code and queries, forms and metadata, the standard library, and process and the agent itself such as a workflow, model choice, working memory and git safety. The skills are layered: basic XML generation works without the platform, while advanced skills need 1C or an MCP server. Fourteen validators parse the XML of an object, form, role, schema, layout or extension, and a configuration index lets validators compare an object against the rest of the configuration, checking reference types, form data paths, subsystem makeup, role rights, query table and field names, and common module calls. That index is what turns a pile of files into a checked whole. That configuration index is what turns a folder of files into a checked whole, catching broken references that a plain text edit would miss.

How a task looks

The README shows a concrete example: a catalog object described in about fifteen lines of JSON becomes nearly six hundred lines of correct XML on disk from a single skill call. The output includes the catalog XML with required sections and generated UUIDs, an object module stub, and a line in Configuration.xml so the object is part of the configuration rather than an orphan file. A validator then reports success with a check count. The README stresses the file structure is not invented or copied from other dumps, because the format is fixed in the skill and locked by snapshot tests that compare output to a reference. Installation offers two paths: ask the agent to clone the repo, inventory the machine, show a plan and install only what you pick, or copy the skill, rule and command folders by hand. The agent led path shows diffs and asks before overwriting, which the README recommends on a machine where you already customized skills. A first task can be as simple as asking in plain words to create a counterparty catalog with an INN and a contact table, and the router skill selects the right workflow for that request. The router skill selects the right workflow for a plain language request, so a beginner does not need to memorize the name of every skill up front.

AI-EDT and dependencies

The set is meant to work fully with AI-EDT, an MCP server that runs as a plugin inside a launched 1C:EDT and gives the assistant a semantic model of the environment instead of file text. With it, the assistant resolves metadata references, builds call hierarchies, asks EDT for inferred types, edits metadata without hand writing XML, sets breakpoints and reads variables, then validates through EDT own checks so errors surface on the spot. The README notes more than a hundred operations are wrapped in facades such as code search, edit metadata, launch debugger, diagnostics, insights and security audit. Some of the set does not depend on the plugin at all: 1C 7.7, web publishing, browser testing, data conversion rules, and the basic source operations run without EDT or the platform. Dependencies listed include PowerShell, Python, v8unpack, 1C:Enterprise, Node with docx, ffmpeg, and optionally an NVIDIA GPU with CUDA for local transcription. The README also names referenced MCP servers for BSL help, code review, platform docs and template search. The project is published under the MIT license and its most recent commit was on 2026-08-26. The README also names the referenced MCP servers for BSL help, code review, platform docs and template search that the skills can call when configured.

Editorial conclusion

This skill set gives Claude Code agents structured operations for 1C:Enterprise source files, and it is published under the MIT license with its most recent commit on 2026-08-26.