[DNM] Claude-generated changes to initial materialize-docs skill

[bosconi]

Not intended to be merged.

A lot of files in here... The gist is:

• skill_upgrade.md was initially generated by ChatGPT when prompted abstractly to suggest ways a skill could be upgraded to be most accessible and efficient for LLM use.
• It was then worked on by Claude Opus (in plan mode using ultrathink, and being told to research best practices from the web). The improvements it proposed are in meta_upgrade_plan.md. It cited at least some of its web sources:

Source	Key Finding
AWS RAG Best Practices	Tables break RAG; use lists instead. Add section summaries.
Bluestream Metadata Strategy	Controlled vocabularies, typed relationships, complexity levels
Nielsen Norman Group	Max 2 levels of progressive disclosure
Write the Docs	Consistency, style guides, version control

• The final skill_upgrade.md was applied by Claude Opus, again with ultrathink.

It wrote a Python script to apply many of the changes. From a quick inspection this is likely to be a "static" script, specialized to the current content of the docs, and hence not something we can just run at the end of gen-claude-skill. But maybe we can extract a lot of generic transformations out of it.

Only 3 new files:

• _GLOSSARY.md
• _SQL_COMMANDS.md
• _VOCABULARY.yaml