Cleverence Mobile SMARTS (.mslx) configuration

SkillDev tools

Cleverence "Склад 15" / "Магазин 15" runs handheld-terminal (ТСД) logic as algorithms: a directed graph of actions. The Mobile SMARTS Panel stores each document type and each operation as a single-line XML file with the .mslx extension.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Cleverence Mobile SMARTS (.mslx) configuration skill

About this capability

>-

What this skill tells your AI

The instructions your AI receives, as published by desko77/cursor-1c-skills in skills/cleverence-mslx/SKILL.md and read by ahel’s review.

Cleverence "Склад 15" / "Магазин 15" runs handheld-terminal (ТСД) logic as algorithms: a directed graph of actions. The Mobile SMARTS Panel stores each document type and each operation as a single-line XML file with the .mslx extension. Editing these by eye is error-prone; the failures are almost always broken transitions or wrong variable / field names, not XML syntax. This skill encodes how the graph works and how to change it without breaking it.

Mental model

  • DocumentType .mslx (под Documents/DocumentTypes/...) - a whole terminal workflow (e.g. "Сборка контейнера"). Its root is <DocumentType>.
  • Operation .mslx (под Documents/Operations/...) - a reusable sub-routine called by documents or other operations. Root is <Operation>.
  • Both contain <Actions> - the ordered list of graph nodes.
  • The Panel and the MS Server work off a server database, not necessarily the git working copy. Editing a file on disk changes nothing until the config is transferred to the server (Сервис -> Сравнение конфигураций, or copying into the live DB + restart). If the Panel shows old behaviour after your edit, suspect a stale/never-transferred file before suspecting your change.

The single most important rule: CL_ copies of vendor operations

Vendor operations (typical Cleverence обработчики, usually under Operations/EN/..., Operations/Основные/... etc.) get overwritten on Cleverence updates. Editing them directly is like editing a vendor 1C extension - your change disappears on the next update.

So:

WhatRule
Our own document type (e.g. "Сборка контейнера")Edit in place. Do not clone.
Our own operation (in the project root Operations/)Edit in place.
A vendor/typical operation we need to changeMake a copy CL_<Name> (placement below), edit the copy, and in the caller switch operationName to CL_<Name>.
A new operationPrefix CL_ (latin).

Copy only the operations you actually change. If the change is inside a nested vendor operation, copy that one and repoint its CL_ parent at it - do not clone the whole chain.

Where to put a CL_ copy. Put it in your own branch of the operations tree - Documents/Operations/ (the root) - not inside the vendor subfolder it was copied from (Documents/Operations/EN/...). The Panel resolves a call by the operation's name, not by its file path, so a CL_<Name> in Operations/ fully serves callers that use operationName="CL_<Name>". Keeping the copy in the vendor subfolder (a) puts it among files that get overwritten on update, and (b) risks two files with the same name if the Panel later re-exports the operation into the root - a duplicate that conflicts on import. So: after creating/moving the copy to Operations/, delete any stale same-name file left in the vendor subfolder.

How the action graph flows

Every action decides where control goes next. Read references/graph-mechanics.md for the full rules; the essentials:

  • Sequential fall-through: an empty nextDirection="" means "go to the next action in document order". This is why the physical order of <Actions> matters - inserting an action in the wrong place silently changes the flow.
  • Named transitions point to an action's name. Resolution is case-insensitive (a button direction "Просмотр факт" resolves to action name="Просмотр Факт"). So if a transition "is not found", the cause is almost never letter-case - it is a genuinely missing action (often a stale file on the server).
  • ConditionAction branches with yesDirection (TRUE) and noDirection (FALSE); empty means fall through to the next action.
  • QuestionYesNoAction branches with yesDirection / noDirection.
  • QuestionAction (a menu) has three parallel arrays - Buttons, ButtonTexts, ButtonDirections - matched by index. They must stay the same length; a button whose ButtonDirection names a non-existent action throws "действие для перехода не найдено".
  • Built-in targets that are not action names: back, return, abort, exit, home, process zero, and "".
  • indent = visual nesting in the Panel tree, not control flow. Each action carries an integer indent. Logically subordinate actions (a process zero branch under the scan window, the body executed inside a condition) must have indent greater than their parent - they appear shifted right in the Panel algorithm tree. Flow is driven by nextDirection/named transitions, not by indent; but leaving a flat indent=0 on nested actions renders the structure wrong/flat in the Panel and is a real defect to fix. When you copy or hand-edit a .mslx, preserve/set indent on the nested actions (it is easy to lose it on a copy). mslx_inspect.py indents its dump by this attribute, so a flat structure shows up at a glance.

Calling another operation: parameter mapping

OperationAction invokes a sub-operation by operationName and maps variables explicitly:

  • InKeys / InValues - the sub-operation's variable name / the caller's expression. Expressions are wrapped in braces: InValues = {GO.GetBarcodeData(ScannedBarcode)}.
  • OutKeys / OutValues - the sub-operation's result variable / the caller's variable.

Many shared variables (BarcodeData, IdentificationCode, CurrentItem, ...) are global to the session, so some calls pass nothing and rely on globals. When in doubt, look at how the operation is already called elsewhere (grep for its operationName) and copy that contract verbatim - guessing variable names is the #1 cause of "scan does nothing".

Calling 1C online from the terminal

To run a 1C function live during a scan, use an InvokeMethodAction against the 1C connector. The function name you pass must be a method of the integration data processor (вендорская ИнтеграционнаяОбработка_*), and it must return a ТаблицаЗначений (not JSON). The full mechanism, the thin-wrapper pattern (with vendor markers), and what is not an online call (the "Список произвольных кодов" is field mapping, not a call) are in references/online-1c-call.md.

Marking codes (КМ / КИЗ) and document lines

The marking code on a document line lives in Item.МаркаИСМП (and Item.ГрупповаяУпаковка for aggregates) - not Item.Марка. You match against the recognised IdentificationCode, which is produced by GetIdentificationCode from BarcodeData, not against the raw scanned string. Reuse the ready operations (FindMCInDocument, DeleteCurrentItemFromDocument, IsMarkingCode) instead of hand-writing queries. See references/km-and-document-operations.md.

Editing methodology (do this, in this order)

  1. Inspect first. Run the bundled tool to see the real graph and current links:
PYTHONIOENCODING=utf-8 python scripts/mslx_inspect.py "<file>.mslx"

It dumps every action (type, name, operationName, directions, In/Out mapping) and then validates: XML well-formedness, that every transition resolves, and that menu button arrays are balanced. 2. Find the proven pattern. Before writing new graph logic, grep the existing operations for one that already does it and copy its action structure and variable contract. The codebase almost always has a working precedent (scanning, finding a line, deleting a line, confirming, calling 1C). Reusing it beats inventing. 3. Make point edits. .mslx is one long line; use exact-string Edit, keep ids stable where you can, and keep the action order consistent with the fall-through you intend. 4. Re-inspect. Run mslx_inspect.py ... --validate again. Zero dangling transitions and balanced button arrays is the bar before you hand it back. 5. Remember the transfer step. A validated file on disk is not yet live - it must reach the MS Server. Tell the user to transfer it and to confirm by checking that a changed operationName (e.g. CL_...) shows up in the Panel.

When a scan "does nothing" on the TSD

Before suspecting the graph, rule out the input itself - the graph is usually innocent:

  • Ctrl+V into the debugger window is NOT a scan event. A scan window catches input via a KeyToAction barcode="{any}" KeyJump, which fires on a scan event. Pasting text into the field with Ctrl+V is keyboard input - it does not trigger the barcode KeyJump, so "nothing happens". Use the debugger's emulate-barcode function (a dedicated barcode input), or a real scanner.
  • DataMatrix loses its GS separators on copy-paste. A ЧЗ marking code (DataMatrix) contains the GS control char (ASCII 29). Clipboard/Notepad strips it, so even if the input lands, GO.GetBarcodeData won't recognise the КМ structure. Only a real scanner (or the emulator) preserves GS.
  • The config may not be on the server. The debugger runs the server copy; an edited file in the repo is not live until transferred (Сервис -> Сравнение конфигураций).
  • Only after these: check the graph. Confirm the scan window has KeyToAction barcode="{any}" pointing at the scan-handling action, and (with a breakpoint in the debugger) whether that action is even reached. If it is reached but bails out, inspect BarcodeData - the code was read but not recognised.

Linking a TSD document back to its 1C document (dedup on completion)

When a TSD document must, on completion, update an existing 1C document instead of creating a new one, the back-link is a Panel setting, not graph logic: in the БП open "Настройка загрузки полей шапки документа" and the rule whose target (приемник) is the Ссылка field, match mode Поиск по GUID. The source attribute of that rule is a classic trap:

  • Идентификатор - the MS document's own Id. This is the correct source for dedup. It is stable for one TSD document and is exactly what the vendor completion code resolves the 1C reference from (ШапкаДокумента.Ид). Use this.
  • Идентификатор исходных документов (ИдИсходныхДокументов) - wrong for the back-link of a TSD-created document. If an online step creates/re-creates the 1C document and writes its GUID here, a fresh GUID can arrive on every completion -> the GUID search never matches -> a new 1C document is created each time = duplicates (often without a warehouse). This attribute is for the "document assembled from several 1C documents" case, not the back-link.

Symptom this fixes: every завершение on the TSD spawns a second 1C document. Root cause is almost always the Ссылка rule pointing at Идентификатор исходных документов instead of Идентификатор. The vendor completion path only ever resolves the 1C ref from ШапкаДокумента.Ид - writing a GUID into ИдИсходныхДокументов does nothing for dedup.

Caveat for TSD-created docs: their MS Id is new_<guid>. A hand-rolled Документы[Тип].ПолучитьСсылку(Новый УникальныйИдентификатор(Ид)) throws on the new_ prefix (invalid GUID) and, inside a bare Попытка, silently falls through to creating a duplicate; the Поиск по GUID rule handles the prefix correctly.

Reference files

  • references/graph-mechanics.md - action types, every transition attribute, fall-through, button arrays, KeyJumps, common caveats. Read when editing graph flow.
  • references/online-1c-call.md - the InvokeMethodAction -> integration-processor -> ТаблицаЗначений mechanism, the wrapper pattern, what is not an online call.
  • references/km-and-document-operations.md - КМ line fields, IdentificationCode, the ready find/delete/recognise operations and how to chain them.

Constraints in 1C-integration projects

Vendor extensions and 1C extensions are not edited directly - mark project insertions and prefer copies/wrappers. In this family of projects code and docs avoid the letter "е" and long dashes. Confirm destructive or shared-state actions (config transfer to Test/Prod) before doing them.

Signals

GitHub stars
57
Forks
13
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
cleverence-mslx-desko77
Source
github.com/desko77/cursor-1c-skills