The LLM view is the half of rendergit I use most, but the CXML text it builds (generate_cxml_text) is only reachable by opening the HTML page and copying the textarea. This adds an optional --cxml flag that emits that same text directly, so it can be scripted:

# write CXML to a file
rendergit https://github.com/karpathy/nanogpt --cxml nanogpt.txt

# or stream to stdout ('-' or bare --cxml) and pipe straight into whatever
rendergit https://github.com/karpathy/nanogpt --cxml - | llm "what does this repo do?"

Same UX ask as #10 (--copy --llm), minus the clipboard dependency.

Details:

• ~20 lines in rendergit.py, no new dependencies. Without the flag, behavior is completely unchanged.
• In --cxml mode no HTML is built or opened at all. Nice side effect: since READMEs render unsanitized into the auto-opened file:// page, this is also the safer mode for flattening untrusted repos.
• Progress messages already go to stderr, so stdout carries pure CXML — pipes stay clean.
• stdout mode reconfigures sys.stdout to UTF-8 with errors="replace" — Windows consoles default to cp1252 and would otherwise raise UnicodeEncodeError on the first emoji in someone's README.
• Temp clone cleanup still runs through the existing finally.

🤖 Generated with Claude Code