doc-splitter-launcher

name	doc-splitter-launcher
description	Launch the document splitting stage for the COMANDOS bundle. Use this when the user has a folder of raw source files and wants one clean `Base` with `prepared_docs`, optional `product_memory`, and canonical `products_live` when factual live rows exist.

Doc Splitter Launcher

What this launcher does

This launcher is a user-facing entrypoint for the document preparation stage.

Use it when the user says things like:

• раздели документы
• подготовь Base
• запусти doc splitter
• собери docs и products_live

Required behavior

1. Ask only for the source folder path if it is not already known.
2. Create or update one Base.
3. Ensure the workspace layout exists:

• new_files/
• prepared_docs/
• vectorized_docs/

4. Run the bundled document splitting logic from:

• ../doc-splitter/SKILL.md
• and use its bundled references/ files, not an external or globally installed variant.

5. Produce:

• prepared_docs/
• optional product_memory
• canonical products_live only if factual live rows really exist

6. Keep the output limited to the current bundle artifacts.
7. Do not offer a separate ad hoc table after the main run.
8. Do not place cleaned files back into new_files/ or directly into vectorized_docs/.
9. Treat new_files/ as a transient queue: after successful normalization, processed source files must be removed from new_files/.

Important implementation rule

Do not treat document preparation as a light copy or a mild markdown cleanup.

The bundled doc-splitter must perform real structural normalization, including:

• remove decorative H1 from final document body;
• remove **Теги:** from final document body;
• start from the first useful ## block;
• avoid one giant ## Описание blob when it contains multiple semantic sections;
• avoid --- as a structural separator;
• normalize product docs into retrieval-friendly sections when source semantics allow it.
• do not treat a near-verbatim copy of the source file as a valid prepared_docs result.
• remove empty headings and headings that have no useful text before the next heading.

After finishing

Always report:

• which docs were created in prepared_docs/;
• how many source files were removed from new_files/;
• whether products_live was created;
• if products_live was not created, why.

Then suggest the next step:

• Запустить ingestion flow