More screenshots
Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsPaper Ask: Answers From Your Papers, With Page Numbers
Ask Claude about the PDFs in your reference library. Every answer cites the page and quotes the original sentence, and Claude can highlight those passages in grey right in your PDF so you can check them.
$5
Paper Ask: Answers From Your Papers, With Page Numbers
Example session with this skill installed
In Fu et al.'s optical neural network review, what holds back integrated ONNs from scaling up?
- Read your context and instructions
- Compiled the paper ask
Found it in your library: Fu et al., Optical neural networks: progress and challenges (Light: Science & Applications 13:263, 2024). I searched the PDF for "scalab|challenge" and read pages 18–20.
Short answer: the review names three bottlenecks for integrated optical neural networks: too few inputs, computing units that are hard to power and drive at speed, and no good way yet to cascade chips.
What the paper says
| Bottleneck | Evidence | Where |
|---|---|---|
| Input dimension | most integrated ONNs have "the input waveguide of less than 10" | p.19 |
| Computing units (MZI mesh, MRR weight banks) | they "require additional energy supply during the operation of the computing units, and many computing units face difficulties in synchronous modulation under high-speed situations" | p.20 |
| Cascading chips | "At present, all on-chip ONNs have not achieved good scalability in the cascading method." | p.20 |
The exception is on-chip diffractive ONNs
- p.20 "its scalability is superior to other on-chip ONNs"
- p.20 "However, the computing units of on-chip DONN are difficult to achieve reconfigurable and nonlinear functions."
On p.18 the authors also tie the limit on MZI/MRR designs to the "adjacent modulator's thermal crosstalk".
Inference (not stated in the paper)
Taken together, the near-term route to larger systems looks like optoelectronic hybrids rather than all-optical chips. The paper proposes hybrids (p.20) but doesn't show one overcoming all three limits.
What the paper doesn't say
It gives no hard limit on how many MZI or MRR units one chip can hold; the thermal-crosstalk point (p.18) is qualitative.
Read p.18–20 | 5 grey highlights (p.19, p.20)
The passages above are now highlighted in grey in your PDF, each with a [Q: what limits scaling of integrated ONNs] comment. Say "open page 20" to jump there in the reader.
Quotes from T. Fu, J. Zhang, R. Sun, Y. Huang, W. Xu, S. Yang, Z. Zhu, H. Chen, "Optical neural networks: progress and challenges", Light: Science & Applications 13, 263 (2024), doi:10.1038/s41377-024-01590-3, licensed under CC BY 4.0. Excerpted.
Connects securely to your tools. The creator never sees your data.
About this skill
Free and open source on GitHub. This listing is the same skill as a ready-to-install package: one zip, no repo to clone, nothing to build. If you'd rather set it up from source, it's free at https://github.com/nebula0/zotero-llm-wiki. Buying here is the convenient route and supports development.
When you ask an AI about a paper, you can't tell whether it read the text or is guessing. Paper Ask makes Claude show its work.
What it does
- Finds the paper in your library by author or title keywords and reads the PDF directly. No uploading, no copy-pasting.
- Every answer cites the page and quotes the sentence:
p.5 "…the original sentence…". - Keeps three things apart: what the paper says, what Claude infers (labelled as inference), and what the paper doesn't say.
- Grey highlights in your PDF (optional, with a free API key): the passages Claude relied on are highlighted in grey with a short comment, so you can check them where you read. Your own coloured highlights are never touched.
- Compares papers: ask how two or three papers differ in method or results and get a side-by-side answer, each point cited.
- "What did I highlight?": lists your own highlights and comments in a paper.
- Saves notes only when you ask: one Markdown file per paper, new Q&As appended.
- Replies in your language.
- Never writes to your library database or PDF files. Reading is read-only; highlights go through the official web API.
Requirements
- Claude Code
- Python 3.9 or newer, plus
pymupdf(setup installs it) - The Zotero 7 desktop app, with your PDFs in it
- Optional: a free Zotero API key, for grey highlights
Setup: Unzip into ~/.claude/skills/paper-ask/, then tell Claude "set up paper-ask". It checks everything step by step, and opens an illustrated guide for the API key.
What it doesn't do
- No web search: it only reads papers that are already in your library.
- Scanned PDFs without a text layer can't be read.
Pairs well with Paper Search (find papers with verified citations and add them to your library).
Zotero is a registered trademark of the Corporation for Digital Scholarship. Paper Ask is not affiliated with or endorsed by Zotero or the Corporation for Digital Scholarship.
How to use
Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scanned
Verified clean 10 days ago
- Passed all security checks, Safe to install
Requires
Python
macOS