read-tool-enhanced

File reads show the first 7 lines while loading, then collapse to a compact summary with filename, line count, and size. The full content is always sent to the LLM’s context regardless of what’s displayed.

SKILL.md files get special rendering with a distinct visual treatment. Press ctrl+o to expand any read to its full content.

PDF support

The read tool now parses PDF files and extracts text instead of returning binary garbage.

Read full PDF

read path/to/spec.pdf

Returns extracted text plus a summary (<path> (<N> pages, <size>KB extracted)).

Read specific pages

read path/to/spec.pdf pages="1-5"
read path/to/spec.pdf pages="1,3,7-10"

pages is ignored for non-PDF files.

Error handling

• Encrypted/password-protected PDFs return a clear error
• Corrupt PDFs return parse errors
• Image-only PDFs report that no extractable text was found

Jupyter notebook support (.ipynb)

Notebook files are parsed as structured cells instead of raw JSON:

• Markdown cells render as plain text
• Code cells render in language fences with execution counts (for example # [5])
• Outputs are summarized safely:
  • text output is included inline
  • image outputs are represented as metadata placeholders (no base64 payload)
  • HTML outputs are labeled as [HTML output]
  • error tracebacks are shown in fenced blocks

read analysis.ipynb

The collapsed summary includes cell/type counts and extracted size; expand to view full formatted cells.

Interaction with @file references

file-reference now understands PDFs too:

• Small PDFs are inlined directly into prompt context
• Large PDFs are summarized with metadata + first-page preview
• The summary suggests read <file> pages="..." for targeted extraction