How to use the DeepXiv paper search and reading agent skill

What the deepxiv skill makes the agent do

An agent skill is a folder with a SKILL.md file that tells an agent when and how to perform a task. This one covers DeepXiv, a source of open-access papers. Its main idea is progressive reading: the agent avoids loading whole papers into the conversation too early.

Beyond paper search and layered reading, the skill can list trending papers, run a DeepXiv web search, and fetch Semantic Scholar metadata by ID.

• Search papers by topic
• Get a brief summary of a paper by its arXiv ID
• Get a section map (metadata and section overview)
• Read one named section, such as Introduction or Methods
• List trending papers over a time window
• Run a DeepXiv web search
• Fetch Semantic Scholar metadata by ID

Install the deepxiv skill

Download the skill ZIP from this page and unzip it into one of two places. For a personal setup, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Install the deepxiv-sdk command-line tool

The skill depends on the separate deepxiv-sdk command-line tool, which needs Python 3. Install it with pip. The skill itself does not replace the CLI.

The skill also expects Bash and git to be available so it can locate its helper script.

Set up your DeepXiv token

The CLI authenticates with a DeepXiv token stored as DEEPXIV_TOKEN. If no token is configured, a command that needs one auto-registers a token and saves it to ~/.env. You do not need to create or paste a token yourself to get started.

If you install the optional python-dotenv package, the CLI also loads ~/.env and the current project's .env file on startup. Setting DEEPXIV_TOKEN in the CLI's environment skips automatic registration and saving, but file loading still happens. The skill does not read these files or display tokens.

• Required: DEEPXIV_TOKEN, auto-registered and saved to ~/.env if not set
• Optional: python-dotenv, so the CLI loads ~/.env and project .env files

Trigger the skill with natural requests

You can ask for the skill in plain language. The skill description covers layered paper access, section-level reading, trending papers, and DeepXiv-backed literature retrieval, so requests on those topics are the ones it is built for.

You can also use the explicit command form with options appended to the arguments. The skill supports these options: max, brief, head, section, days, web, and sc.

• Find papers on a topic: "Search DeepXiv for recent papers on agent memory and give me the top 5."
• Quick summary: "Give me a brief of arXiv paper 2409.05591."
• Section map: "Show the section overview for 2409.05591."
• One section: "Read only the Introduction of 2409.05591."
• Trending: "What papers are trending on DeepXiv over the last 14 days?"
• Web search: "Run a DeepXiv web search for karpathy."
• Semantic Scholar: "Fetch Semantic Scholar metadata for ID 258001."

Read a paper in layers

The skill follows a fixed order so you only pay for the depth you need. Start with a search, move to a brief, then a section map, then one section. Read the full paper only when a brief or a single section cannot answer your question.

When reading a paper, the agent shows the title, arXiv ID, authors, venue or date if available, a TLDR or abstract summary, and a suggested next step.

• 1. Search to find candidate papers
• 2. Brief for a short summary
• 3. Head for metadata and the section map
• 4. Section for one named part
• 5. Full paper only if necessary

Safety notes: scripts and network access

This skill runs scripts and uses the network. It calls the DeepXiv service through the deepxiv-sdk CLI, and it can call Semantic Scholar metadata through the same tool. Its allowed tools include Bash, Read, and Write.

Credentials are handled by the CLI, not the skill. If you do not want a DeepXiv token saved in ~/.env, set DEEPXIV_TOKEN in the CLI's environment so automatic registration and saving are skipped. Be aware that loading .env files is still active when python-dotenv is installed.

• Runs scripts: yes (Bash commands)
• Uses network: yes (DeepXiv service and Semantic Scholar via the CLI)
• Token may be saved automatically to ~/.env unless DEEPXIV_TOKEN is set

Tips for better results

Give an arXiv ID when you know the paper, since the brief, head, and section commands work from IDs. Use a topic query when you are exploring. The skill is limited to open-access sources, so coverage of paywalled papers is not expected.

If you use the ARIS research workflow and have a research-wiki/ folder, the skill can record papers you read in depth into the wiki. Papers you only saw in search results are not ingested.

• Use arXiv IDs for brief, head, and section requests
• Ask for one section at a time to keep context small
• Set a shorter max for quick scans
• Check that deepxiv-sdk is installed if the CLI is not found