Turned ArXiv into an AI-Agent-Friendly Interface (No Browser Vision Needed)

Install This Skill

npx skills add mediause/agent-skills/arxiv

This skill is synced from github.com/mediause/agent-skills and automatically maps to the matching site plugin data by plugin name. You only need to follow the CLI steps in the skill guide.

skill.md

arxiv

Discover arXiv research with a guest-only workflow for paper search, category recents, author lookups, and detailed paper retrieval.

Discover arXiv research with a guest-only workflow for paper search, category recents, author lookups, and detailed paper retrieval.

Scope

Use this skill when the task targets arXiv operations such as:

Account: plugin health

Search: papers by keyword

Recent: latest submissions in a category

Paper: full details for a specific paper ID

Author: papers by a named author

arXiv is a public read-only repository. No login required. All operations are read-only.

1. Install MediaUse CLI (Windows Only)

Use the official install script for Windows:

https://release.mediause.dev/install.ps1

Run:

powershell -C "iwr https://release.mediause.dev/install.ps1 -UseBasicParsing | iex"

Then verify :

mediause --version

Current support status:

Windows: supported

Linux: not supported yet

macOS: not supported yet

Recommended skill install path:

.mediause/skills/arxiv/SKILL.md

1. Get and Configure MediaUse Key

2.1 Apply for key

Open https://mediause.dev/

Sign in to your account.

Open Project.

Create or copy your API key.

2.2 Configure key in CLI

mediause manage key --json

1. Core Flow (Mandatory Order)

Always follow this order:

Discover site and commands.

Bind account context with use account (guest is the only mode for arXiv).

Execute dynamic site actions.

Verify with trace/task.

arXiv is a fully public API — no login required. Always use arxiv:guest as the account context. Skip auth health for guest mode.

3.1 Discover and plugin setup

mediause plugin list --json mediause plugin add arxiv --json mediause arxiv -h mediause arxiv search -h mediause arxiv get -h mediause arxiv user -h mediause arxiv account -h

3.2 Bind guest context

arXiv does not require login. Use guest mode.

mediause use account arxiv:guest --show --json

Guest mode rules:

All arXiv operations are read-only (account health, search papers, get paper, get recent, user author).

No write operations exist for arXiv.

If page shows unusual traffic or captcha, repeat with --show to manually resolve.

3.3 Auth health

Not required for guest mode. Skip this step for arXiv.

1. arXiv Dynamic Command Map (v1)

public arXiv API plugin with guest default account and read-only commands.

4.1 account.health

Check plugin/runtime health for current arXiv context.

mediause arxiv account health --json

4.2 search.papers

Search papers by keyword across all fields (title, abstract, authors, etc.).

mediause arxiv search papers --query "" [--limit ] --json

--query (required): keyword or phrase, e.g. "attention is all you need"

--limit: max results, default 10, max 25

Columns returned: id, title, authors, published, primary_category, url

Example:

mediause arxiv search papers --query "transformer language model" --limit 10 --json mediause arxiv search papers --query "diffusion model image generation" --limit 5 --json

4.3 get.recent

List recent submissions in a specific category, sorted by submission date descending.

mediause arxiv get recent --category [--limit ] --json

--category (required): arXiv category code, e.g. cs.CL, cs.LG, math.PR, q-bio.NC

--limit: max results, default 10, max 50

Columns returned: id, title, authors, published, primary_category, url

Common categories:

CategoryDescription

cs.CLComputation and Language (NLP)

cs.LGMachine Learning

cs.CVComputer Vision

cs.AIArtificial Intelligence

cs.CRCryptography and Security

math.STStatistics Theory

q-bio.NCNeurons and Cognition

physics.comp-phComputational Physics

Example:

mediause arxiv get recent --category cs.CL --limit 20 --json mediause arxiv get recent --category cs.LG --limit 10 --json

4.4 get.paper

Get full details for a specific paper by arXiv ID.

mediause arxiv get paper --id --json

--id (required): arXiv paper ID, e.g. 1706.03762 or 2303.08774

Columns returned: id, title, authors, published, updated, primary_category, categories, abstract, comment, pdf, url

Example:

mediause arxiv get paper --id 1706.03762 --json mediause arxiv get paper --id 2303.08774 --json

4.5 user.author

List papers by a named author, newest first. Author name matching is fuzzy — try alternate spellings if no results.

mediause arxiv user author --author "" [--limit ] --json

--author (required): author full name or initials, e.g. "Yoshua Bengio" or "Y Bengio"

--limit: max results, default 20, max 50

Columns returned: id, title, authors, published, primary_category, url

Example:

mediause arxiv user author --author "Yoshua Bengio" --limit 20 --json mediause arxiv user author --author "Andrej Karpathy" --limit 10 --json mediause arxiv user author --author "Y LeCun" --json

1. Workflow Examples

Workflow A: Discover recent NLP papers and retrieve one in full

A1. Setup

mediause plugin add arxiv --json mediause use account arxiv:guest --show --json

A2. Browse recent submissions in cs.CL

mediause arxiv get recent --category cs.CL --limit 15 --json

A3. Get full details and abstract for a specific paper

mediause arxiv get paper --id 2303.08774 --json

A4. Verify

mediause trace last --json

Workflow B: Keyword search and author follow-up

B1. Setup

mediause use account arxiv:guest --show --json

B2. Search for a topic

mediause arxiv search papers --query "large language model reasoning" --limit 10 --json

B3. Find more papers by the first author

mediause arxiv user author --author "Jason Wei" --limit 20 --json

B4. Get full paper details

mediause arxiv get paper --id 2201.11903 --json

B5. Verify

mediause trace last --json

Workflow C: Monitor a research area daily

C1. Setup

mediause use account arxiv:guest --show --json

C2. Pull recent AI papers

mediause arxiv get recent --category cs.AI --limit 50 --json

C3. Pull recent ML papers

mediause arxiv get recent --category cs.LG --limit 50 --json

C4. Search for a specific concept

mediause arxiv search papers --query "in-context learning" --limit 25 --json

C5. Verify

mediause trace last --json

1. Operational Constraints (Mandatory)

6.1 Read-only

arXiv has no write operations. All commands are fetch-only. Do not attempt post, reply, or engage actions.

6.2 Frequency limits

arXiv public API has a soft rate limit. Apply pacing between repeated calls.

OperationDefault limitMinimum spacing

search papersmax 25/call>= 3 seconds between calls

get recentmax 50/call>= 3 seconds between calls

get paper1/call>= 1 second between calls

user authormax 50/call>= 3 seconds between calls

Do not run bulk loops without delay (e.g. fetching 100+ papers in rapid succession).

If you receive HTTP 429 or arXiv API errors, wait at least 30 seconds before retrying.

6.3 Content use constraints

Do not republish paper abstracts or full text without proper attribution.

Do not scrape arXiv to build competing paper indexes.

Respect arXiv's usage policies.

6.4 Failure handling

Always use --json for structured error output.

Common errors:

ErrorCauseAction

No papers foundQuery too specific or wrong spellingBroaden query or check spelling

Paper was not foundInvalid arXiv ID formatUse format YYMM.NNNNN or subj-class/YYMMNNN

Invalid arXiv categoryCategory code typoCheck arXiv category list

arXiv API HTTP 4xx/5xxAPI error or rate limitWait 30s and retry

unusual traffic / captchaGuest session flaggedRerun use account arxiv:guest --show --json to resolve manually

Recovery pattern:

On arXiv API error or unusual traffic

mediause use account arxiv:guest --show --json mediause trace last --json

1. Quick Reference

always run once before each workflow (auto-upgrade latest)

powershell -C "iwr https://release.mediause.dev/install.ps1 -UseBasicParsing | iex" mediause --version

Install

powershell -C "iwr https://release.mediause.dev/install.ps1 -UseBasicParsing | iex" mediause plugin add arxiv --json

Context

mediause use account arxiv:guest --show --json

Commands

mediause arxiv account health --json mediause arxiv search papers --query "" [--limit ] --json mediause arxiv get recent --category [--limit ] --json mediause arxiv get paper --id --json mediause arxiv user author --author "" [--limit ] --json

Verify

mediause trace last --json mediause task status --task-id --json

Skill Metadata

Maintainer: @mediause-demo

Last-Updated: 2026-05-12

Version: v1