Deepxiv

DeepXiv

Use this skill when the user needs a paper-native search and reading surface, not a broad web crawl.

DeepXiv is best treated as:

• a literature access substrate
• a progressive reading interface
• a shortlist and identifier recovery lane

Do not treat it as:

• a replacement for the user's long-term literature library
• the only acquisition path
• a project-specific knowledge base

Best-Fit Use Cases

Use this skill for:

• arXiv / PMC friendly paper search
• quickly shortlisting papers before deep reading
• brief -> head -> section -> full progressive access
• recovering arXiv IDs, DOI-based bioRxiv / medRxiv access, or PMC handles
• preparing inputs for literature-finder-downloader or a project-local library skill

Prefer another lane when:

• the task is broad web discovery across blogs, repos, news, and forums
• the user already has BibTeX / RIS / DOI batches and wants end-to-end downloading
• the main job is bibliography management or long-term attachment organization

Setup

Prefer the bundled isolated runtime first:

deepxiv-local --help

Fallback wrapper path:

$CODEX_HOME/skills/deepxiv/scripts/deepxiv_local.sh --help

If that wrapper is unavailable, check whether the CLI is already available:

command -v deepxiv

If not installed, install it with Python user packages:

python3 -m pip install --user -U "deepxiv-sdk[all]"

Minimum package if only the reader/CLI lane is needed:

python3 -m pip install --user -U deepxiv-sdk

Recommended local-isolated install for this skill:

python3 -m venv $CODEX_HOME/skills/deepxiv/.venv
$CODEX_HOME/skills/deepxiv/.venv/bin/python -m pip install -U pip setuptools wheel
$CODEX_HOME/skills/deepxiv/.venv/bin/python -m pip install -U "deepxiv-sdk[all]"

Important token behavior:

• on first CLI use, DeepXiv may auto-register an anonymous token and write it to ~/.env
• if the user wants explicit control, prefer configuring the token manually instead of relying on auto-registration

Explicit token options:

deepxiv config --token YOUR_TOKEN
export DEEPXIV_TOKEN="YOUR_TOKEN"
deepxiv paper 2409.05591 --token YOUR_TOKEN --brief

Default Workflow

1. Search

Start with a small search pass:

deepxiv-local search "trajectory recovery" --limit 5
deepxiv-local search "trajectory representation learning" --limit 10
deepxiv-local search "image generation" --date-from 2026-03-01 --limit 20

If machine-readable output is helpful:

deepxiv-local search "trajectory recovery" --limit 20 --format json

For web-profile search:

deepxiv-local wsearch "karpathy"
deepxiv-local wsearch "karpathy" --json

2. Progressive Reading

Once you have a candidate paper id, read in layers:

deepxiv-local paper 2505.13857 --brief
deepxiv-local paper 2505.13857 --head
deepxiv-local paper 2505.13857 --section Introduction
deepxiv-local paper 2505.13857 --section Experiments
deepxiv-local paper 2505.13857

Interpretation:

• --brief
  • low-cost paper triage
• --head
  • section map and token distribution
• --section
  • load only the useful part
• bare paper
  • full content access when the cheaper layers were not enough

Use --brief before --head, and --head before section or full reads unless the user explicitly asks for the full paper immediately.

3. Non-arXiv Open Literature

PMC:

deepxiv-local pmc PMC544940 --head

bioRxiv / medRxiv by DOI:

deepxiv-local biorxiv 10.1101/2021.02.26.433129 --section Introduction,Methods
deepxiv-local paper 10.1101/2021.02.26.433129 --biorxiv
deepxiv-local paper 10.1101/2021.02.26.433129 --biorxiv --section Introduction

Source-specific search:

deepxiv-local search "protein design" --biorxiv --limit 5
deepxiv-local search "Alzheimer" --medrxiv --date-from 2024-01

Acquisition Boundaries

DeepXiv is excellent for shortlist construction and progressive reading, but it is not the user's durable library home.

When the user wants:

• many papers downloaded locally
• a seed folder with PDF attachments
• metadata tables and unresolved queues

route from DeepXiv into literature-finder-downloader.

When the user wants:

• a project-specific paper library
• both local PDF and Markdown under a stable repo path

route from DeepXiv into the relevant project-local literature skill.

Practical Patterns

Pattern A: shortlist first, acquire later

1. deepxiv search ...
2. deepxiv paper <id> --brief
3. deepxiv paper <id> --head
4. keep only the papers worth acquisition
5. pass the identifiers into literature-finder-downloader

Pattern B: compare methods cheaply

1. search one focused topic
2. run --brief on 3 to 10 candidates
3. run --head only on the strongest subset
4. read Introduction, Method, and Experiments sections only

Pattern C: prepare a project library update

1. use DeepXiv to recover strong candidates and identifiers
2. acquire durable local attachments through the project's literature lane
3. keep DeepXiv as the external access layer, not the final storage layer

Output Contract

When using this skill, report:

• what was searched
• which identifiers or papers were shortlisted
• which progressive-reading layer was used
• what remains uncertain
• whether the next step should be:
  • more DeepXiv reading
  • durable acquisition via literature-finder-downloader
  • handoff into a project-local paper library

Common Triggers

• “用 DeepXiv 先搜几篇相关论文”
• “先别整篇读，先给我 brief/head 看看值不值得深读”
• “帮我用 DeepXiv 找到这篇论文的 arXiv / PMC 入口”
• “把 DeepXiv 当成论文搜索和预读入口，再接到本地下载链”