Wiki research — academic

Fetches academic papers and metadata from arXiv, DOI resolvers, or Semantic Scholar into raw/academic/. Use this sub-skill when the input contains an arXiv ID, DOI, or phrases like "paper", "preprint", "study", "literature review".

Step 1 — Identify the source type

Step 2 — Fetch paper metadata and content

arXiv (primary for CS / ML / physics)

Fetch metadata:

curl "https://export.arxiv.org/api/query?id_list=<ARXIV_ID>&max_results=1" > /tmp/arxiv.xml

Parse from XML: <title>, <author>, <published>, <summary> (abstract), <id> (canonical URL), <link> (PDF).

Fetch full paper (HTML version, preferred over PDF):

llm-wiki ingest firecrawl "https://arxiv.org/html/<ARXIV_ID>" \
  --out academic/<year>/<arxiv-id>.md
# Fallback to abstract page if HTML not available:
llm-wiki ingest firecrawl "https://arxiv.org/abs/<ARXIV_ID>" \
  --out academic/<year>/<arxiv-id>.md

Search by keyword:

curl "https://export.arxiv.org/api/query?search_query=all:<QUERY>+AND+cat:cs.AI&max_results=10&sortBy=relevance" > /tmp/arxiv_search.xml

See skills/wiki-research/references/query-design.md for query construction.

DOI resolver

# Resolve DOI to landing URL
curl -sI "https://doi.org/<DOI>" | grep -i location
# Then fetch the landing URL with wiki-research-web

Semantic Scholar API

# Search
curl "https://api.semanticscholar.org/graph/v1/paper/search?query=<QUERY>&fields=title,authors,year,abstract,externalIds,openAccessPdf" > /tmp/s2.json

# Fetch by arXiv ID
curl "https://api.semanticscholar.org/graph/v1/paper/arXiv:<ARXIV_ID>?fields=title,authors,year,abstract,citationCount,references" > /tmp/s2_paper.json

Step 3 — Write to raw/

Output path: raw/academic/<year>/<arxiv-id-or-slug>.md

For non-arXiv papers, use: raw/academic/<year>/<first-author-last-name>-<short-title>.md

Required frontmatter:

---
title: "<Paper Title>"
source_type: academic
source_url: https://arxiv.org/abs/<id>
fetched_date: YYYY-MM-DD
arxiv_id: "2301.12345"       # omit if not arXiv
doi: "10.xxxx/..."            # omit if not known
authors:
  - "First Author"
  - "Second Author"
year: YYYY
venue: "NeurIPS 2024"         # conference/journal if known
abstract: |
  <abstract text>
---

<full paper content or abstract + key sections>

Step 4 — Check for related work

For important papers, optionally fetch:

• Cited-by papers (if high-impact): Semantic Scholar citations endpoint
• HN discussion (if notable): search site:news.ycombinator.com <paper title> via wiki-research-social
• Blog post explaining it: search <paper title> explained via wiki-research-web

Limit to 2–3 supplementary sources to avoid scope creep.

Step 5 — Evaluate and return to orchestrator

Run each file through skills/wiki-research/references/source-eval.md. Then return to wiki-research Step 3 (post-process).

Wiki citation format:

Chain-of-thought prompting improves multi-step reasoning (Wei et al., 2022 — `raw/academic/2022/wei-chain-of-thought.md`).

Done looks like

• Papers or abstracts land under raw/academic/ with structured frontmatter; source-eval applied.
• Control returned to wiki-research post-process (merge not done inside this skill alone).

Troubleshooting

Related skills

• wiki-research — orchestrator that invoked this skill
• wiki-research-web — for non-academic URLs
• wiki-research-social — for HN discussions of papers
• wiki-raw-prepare — for cleaning PDF/OCR output

Smoke check

• CLI: Run llm-wiki integrations status and any llm-wiki line from Step 1 of this skill (from the vault root).
• Prompt: Invoke this skill by name in Claude Code; complete Step 1 only and confirm expected CLI or file output.