Tighten Google Slides summary read path

[Aismit]

Summary

This tightens the main Google Slides skill so deck-reading and summarization workflows stay on the connector read path.

What changed

• clarify in the skill description that deck reading and summarization should stay on Google Slides read tools
• explicitly tell the model not to detour into generic code execution like js_repl for slide finding and summarization
• add a concrete read sequence: get_presentation_text for candidate slides, then get_slide and get_slide_thumbnail when visual evidence matters
• clarify that export is not the default path for finding content or understanding visuals

Why this was needed

In a recent codex exec run, the skill was loaded correctly but the model took a bad branch after get_presentation_text, attempted a js_repl detour, and then fell back to export instead of staying on the intended Slides read path.

Testing

• inspected the skill diff locally
• no automated tests run; prompt-only skill change