Read tool handles long PDFs better: total page count reported, and oversized PDFs can return first pages

What

The Read tool is what Claude uses to open files. It changes how it handles PDFs with many pages:

• When a PDF is read in page ranges, the result now tells Claude the total page count and suggests what to read next. For example: "This PDF has N pages. These are pages 1-K. Use the pages parameter to read other pages". Before, it said only "PDF pages extracted: N page(s) from path (size)".
• When a PDF has more pages than can be read at once, the Read tool can now read the first pages and return them with the total page count, instead of failing with the "too many to read at once" error. Before, this error was always thrown. This behaviour sits behind the flag tengu_misty_sunbeam, which has not been read, so whether it is active is not known.
• PDF page-range reads now check page counts across the session, and the number of pages in the document is now recorded.
• The pages argument is now checked differently for image files.

The Read tool's description shown to Claude is unchanged in this release.

Why

Claude can work through a long PDF section by section, knowing how long it is, instead of guessing. Where the new behaviour applies, a very large PDF gives a usable first part rather than an outright error.