If you're using Claude Code for research: stop making it read directly from PDFs We've introduced a SKILL.md that fetches structured, AI-friendly paper overviews from alphaXiv 👀

Paste the following into Claude Code!

Solid as a triage layer for quickly checking if a paper's relevant. But for reproducing results, the raw paper is irreplaceable - equations and hyperparameters don't survive summarization well. Curious if the structured overviews include enough detail for implementation or if it's more of a 'should I read this?' filter.

Just added the full text as a fall back:

Can you expose an end point to actually get the pdf too? Even if it's a proxy to arxiv, it'll be helpful

Coming soon!

Worked on my paper :)

This is cool

Interesting shift—does the SKILL.md approach preserve citation granularity and equations as well as direct PDF parsing when comparing nuanced claims across sections?

@rehaanahmad171 🔥🔥🔥

@rajpalleti314 Definitely needed this 🤌

Hey, we should talk. I'm working on Papyrus,

Claude Web doesn't like the default name. Easy to fix on my side but thought you'd like to know.

I recently found a way for paper reading, swipe like Tinder😈🔥

nice!

This is really nifty. Direct PDF dumps flood Claude Code's context with noise. Structured extraction is the the best way to make its research reliable.

oh this is great, i'll use this in my class

Interesting

Or you could hit it locally with or for instance

wow

Cool

How does SKILL.md handle version control when papers get updated? I'm dealing with research that references constantly evolving datasets and the PDF approach breaks every few weeks when authors push revisions.

Thanks! As a regular reader who really wants to dig into paper content, this is incredibly useful.

damn lovely. converting to md is the best thing for AI parsing.

Skills are underrated for this. the SKILL.md pattern works bc the agent reads it once and knows what tools to reach for. we build all our automation around skills now, each one gets its own tools and hooks. structured input beats dumping raw PDFs into context every time

pdfs to structured data? that's a solid upgrade. doing the heavy lifting for us.

Interesting approach. Structured research summaries will likely make AI agents far more effective than raw PDF parsing.

Structured paper overviews make a huge difference for AI agents compared to raw PDFs.

Structured paper overviews game-changer for research velocity. Converting PDFs to AI-friendly formats reduces friction. Perfect integration with Claude workflows

AI tools work better when papers become data not messy documents

This is exactly what research workflows need. PDF parsing has always been the bottleneck—structured metadata unlocks a completely different level of utility for LLMs. The difference between 'reading' and 'understanding' a paper comes down to context hierarchy.

😎

Why don't we use the HTML file in arXiv?

Awesome !! Actually, I was going to make the same skill yesterday and the origin launched origin ~~ One thing I want to share, both Skill-Creator and Writing-Skills (in Superpowers) suggest that the description needs to start "Use when..." while the SKILL.md doesn't.

Yes! Would also appreciate an AlphaXiv CLI. I tried making an agent to get the relevant papers in my interest added to notion using Claude Code. Would be super useful for custom workflows researchers have

@JFPuget That is great an MCP or api end point would add even better reach to for LLM

I made that skill obsolete or just lwk steal it

Academic paper access is one of the most underrated capabilities for agents. Once they can read papers, they can actually reason about new ideas instead of just existing docs.