comol/ai_rules_1c/1c-live-ib

1c-live-ib — the running infobase

Escalate here only when the configuration dump, the static validators and the docs cannot answer: "does it run now in this IB". Never a replacement for 1c-validate. Available only when the tools are exposed in the session; a configured URL proves nothing, and fake "execution output" is a defect.

Tools and exact arguments

Calls

{"tool": "vcvalidatequery", "args": {"querytext": "ВЫБРАТЬ Т.Ссылка ИЗ Справочник.Контрагенты КАК Т ГДЕ Т.ИНН = \"7701234567\""}}
{"tool": "vcexecutequery", "args": {"querytext": "ВЫБРАТЬ ПЕРВЫЕ 5 Т.Ссылка, Т.ИНН ИЗ Справочник.Контрагенты КАК Т"}}
{"tool": "vcexecutecode", "args": {"bslcode": "Результат = Строка(Метаданные.Справочники.Контрагенты.Реквизиты.ИНН.Тип.КвалификаторыСтроки.Длина);"}}
{"tool": "vcloggetlasterror", "args": {}}

Safety

• Read-only first: vcvalidatequery → vcexecutequery; non-mutating fragments for vcexecutecode.
• No Записать(), Удалить(), НачатьТранзакцию, register movements without asking the user, naming the object and having a rollback plan; on a production IB (.dev.env INFOBASE_ROLE=prod) refuse and request a copy.
• No secrets or personal data in bslcode / querytext — they travel over HTTP and may be logged.
• Validate AI-generated text first (vcvalidatequery, syntaxcheck), then run it here.
• Gate 3a (content/rules/verification-gates.md) uses only the read-only tools.

Where it sits in the playbooks: error fixing step 2 and 6, performance step 5, new-code step 5 — content/rules/tooling-playbooks.md. Setup and 401/403 troubleshooting — content/commands/checkmcp.md. Details: content/skills/mcp-1c-tools/docs/1c-data-mcp.md.