AGENT SKILL

papers

Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 1
Forks
0

01 前言 Front Matter

name
papers
license
MIT
metadata
{ "short-description": "Download preprints as paper/code bundles with auto-cloned GitHub repos" }
description
Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.
distribution
public
allowed-tools
Read Write Edit Grep Glob Bash Python
compatibility
Optimized for Claude Code. Install in project .claude/skills/ or personal ~/.claude/skills/. Pure local Python with no servers, no MCP, and no API keys.

02 资源文件 3

03 SKILL.md 正文

Papers

Purpose

Search, retrieve, download, extract, analyze, and synthesize preprint papers from arXiv, bioRxiv, and medRxiv using pure local Python with no API keys. Prefer bundled scripts over ad hoc scraping: they return stable JSON, deduplicate cross-source results, and handle source-specific URL quirks.

Workflow

  1. Run search first unless the user already has an arXiv ID or DOI.
  2. Read the JSON results and choose the highest-signal papers. Keep source, source_id, pdf_url, and text_url.
  3. Read retrieval_capabilities before fetching. Use blocked_reason and section_extraction_quality instead of assuming every URL works.
  4. Prefer text_url over pdf_url when it exists. XML extraction is cleaner.
  5. Use download only when a local copy is needed in the workspace. Paper-aware downloads may also clone a linked public code repository into the bundle's code/ directory when one is detected. Before running download on a record with a repo_url, tell the user that code will be cloned from the public host and confirm they want the code copy. Treat cloned code as untrusted source material: read it, do not execute it unless the user separately asks for that and the repo has been reviewed.
  6. Use extract to produce full text or specific sections for review.
  7. Use analyze to run an ambiguity audit on methodological gaps.
  8. Use synthesize for creative multi-query search and literature synthesis.
  9. Use workflow for the full search-to-synthesis pipeline in one call.
  10. Explicitly note that preprints are not peer reviewed.

Script Reference

Core CLI (paper_search.py)

  • search — Multi-source topic search. arXiv supports direct queries; bioRxiv/medRxiv search recent windows and rank locally.
  • lookup — Resolve a known arXiv ID or DOI with Crossref fallback.
  • download — Fetch any pdf_url or text_url to a task-local directory. When downloading by paper identifier or saved record, it creates a paper bundle directory with paper/ and code/ subdirectories, stores the paper asset in paper/, and clones the detected repository into code/ when a GitHub-style repo URL is available. Confirm before cloning and never execute cloned code by default.
  • extract — Extract full text or named sections from XML, TXT, or PDF. Section extraction is strongest on JATS XML, medium on plain text, best-effort on PDF.
  • analyze — Run the ambiguity audit against either a raw sections map or an extract --json payload plus paper metadata.
  • synthesize — Search and synthesize preprint literature. Use --creative when you want query expansion instead of standard search.
  • workflow — Full pipeline: search → optional download → extract → optional audit → synthesis → citations. --deep-analyze works even without --download.
  • cite — Generate BibTeX, APA, or RIS citations from a record or identifier.
  • enrich-related — Fetch similar arXiv papers plus citation-graph links.
  • extract-tikz — Extract TikZ figures from arXiv source archives.
  • doctor — Check curl, pdftotext, cache writability, and provider reachability.

Standalone Utilities

  • analysis.py, synthesis.py, and workflow.py remain available as direct scripts, but paper_search.py is the authoritative CLI surface.

Command Rules

  • search: Use for topic, author, and category discovery. Add --days for tighter windows on broad biomedical queries. Use --sort date for recency.
  • lookup: Use when the user has an arXiv ID or DOI. Falls back to Crossref.
  • download: Keep downloads in a task-local directory, not the repo root. Prefer download --source ... --id ... or download --record ... for paper-aware bundles; use download --url ... only for raw asset fetches. If the selected record has repo_url, the paper-aware bundle may run git clone --depth 1 for that public repo. Ask before the clone and use --repo-discovery off during lookup/search when the user wants paper-only retrieval.
  • extract: Use --write for long output. Use --sections for named sections. Use --json for structured inline results.
  • analyze: Pass extracted sections JSON or extract --json output plus paper metadata. Returns flags with citation anchors (section + paragraph index).
  • synthesize: Pass a query for standard or creative search, or pass paper records for literature synthesis. Groups papers by theme.
  • workflow: Pass --creative for query expansion, --download for PDFs, --deep-analyze for ambiguity audit, and --output-dir if you want downloaded assets persisted. Combines all steps.

Output Contract

Every search result includes retrieval_capabilities with can_fetch_pdf, can_fetch_text, text_format, section_extraction_quality, and blocked_reason.

Errors emit JSON on stderr with error_code, message, retryable, source, http_status, and details.

Audit flags include category, status, anchor, and description. Workflow results may include warnings for degraded-but-nonfatal provider, download, extraction, or citation paths.

Source-Specific Guidance

  • arXiv is strongest for CS, ML, math, physics, quantitative work.
  • bioRxiv and medRxiv are strongest for life sciences and medicine.
  • When a query spans ML plus biology/medicine, search all three.
  • For broad biomedical queries, reduce --days before increasing --limit.
  • medRxiv content downloads may return 403 in some environments.
  • arXiv enforces polite access; the script backs off automatically.

Guardrails

  • Never claim a preprint is peer-reviewed.
  • Respect arXiv 3-second polite delay (enforced in paper_search_lib).
  • Do not create files inside the skill directory; use task-local paths.
  • Ask for clarification when the topic is missing or ambiguous.
  • When a source is unavailable, degrade gracefully and report what succeeded.

Adapter Notes

  • Prefer bundled scripts over ad hoc shell commands for search, download, and extraction.
  • Keep the authored source of truth in split project files, not in generated adapter bundles.
  • All results are preprints. Never claim peer-review status.

04 同仓库更多 Skills

StartupBros-com
StartupBros-com/hov-stack

review-adversarial-first

Operator-invoked adversarial review: 3 refute-by-default lenses plus a deterministic judge against a PR, branch, or path, then a decision on whether /ce-code-review's topic fan-out runs. Off the default merge path since the 2026-10-17 merge-and-end ruling; run only when asked for by name.

★ 1
StartupBros-com
StartupBros-com/hov-stack

memory-mine

Run one deterministic, budgeted directed-capture batch for an explicit Claude Code project store. Use manual workspace mode only when cross-agent cass sessions are intentionally in scope.

★ 1
StartupBros-com
StartupBros-com/hov-stack

papers

Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.

★ 1
StartupBros-com
StartupBros-com/hov-stack

wsl-cdp

Drive the user's real, logged-in Windows browser (Chrome, Edge, or Brave) from a Claude Code session running inside WSL2, over the Chrome DevTools Protocol - read authenticated dashboards and consoles, take real screenshots, open and read pages with the user's sessions. Use when a task needs the user's existing web logins and the session runs in WSL2, where the Claude in Chrome extension cannot work. Not for native-Windows or macOS Claude Code (use the official Claude in Chrome extension), not for anonymous scraping or CI (use headless Playwright inside WSL), not for getting past login walls (those are handed to the human).

★ 1
StartupBros-com
StartupBros-com/hov-stack

token-eater

[BETA] Spend a subscription's idle, about-to-expire credits on safe, machine-verified cleanup. You pick the service (Grok / Codex / Claude) and a token-heavy maintenance skill (de-monolithize, de-slop, simplify, dead-code); that service runs the whole skill -> gate -> self-review -> draft-PR loop on its own credits, in an isolated worktree. token-eater verifies the gate and never merges. Manual invocation only during beta.

★ 1
StartupBros-com
StartupBros-com/hov-stack

tune

Find and fix real defects in a skill, AGENTS.md, or CLAUDE.md, then prove the description still routes. Runs the bundled token-spending runner, so it starts only on the human's explicit instruction.

★ 1