Paper Query

Paper Query Skill

What I Do

• Search 7+ academic databases in parallel with automatic retry and fallback
• Filter: Only show published papers (journal/conference proceedings, exclude unpublished preprints)
• Deduplicate and merge results by DOI, arXiv ID, or title similarity
• Annotate venue tiers using CCF (A/B/C), CAS-JCR (Q1-Q4), and SCI (Q1-Q4)
• Translate paper titles from English to Chinese
• Generate 2-3 sentence bilingual summaries from abstracts
• Format results as Markdown with direct links to arXiv, DOI, Semantic Scholar
• Auto-save results to papers/query_.md

---

Workflow

Step 0: Check Configuration

Read paper-config.json from the project root. If the file is missing:

> I notice your paper query is not set up yet. Let me help you configure it. Please run /paper-setup to create your profile — it takes about 2 minutes.

If the file exists, extract:

• user.name, user.institution, user.field_of_study
• api_keys for rate-limited APIs
• preferences.databases, preferences.min_venue_tier, preferences.max_results
• preferences.sort_by, preferences.year_range, preferences.output

Dynamic Year Range: If preferences.year_range is null or not set, default to the last 3 years:

• year_start = current_year - 3 (e.g., 2023 if today is 2026)
• year_end = current_year
• User can specify explicit year range in query to override this default

Step 1: Parse Query

From the user's query string, extract:

• Keywords: core search terms — generate 2-3 query variants (exact phrase, broader terms, synonyms) to maximize recall
• Authors: if mentioned (e.g., "papers by Hinton")
• Year range: if user specifies (e.g., "--year 2023-2025"), use it. Otherwise, use config default (last 3 years). CRITICAL: Always apply a year filter to avoid overwhelming results.
• Venue filter: if user mentions a specific conference/journal
• Topic mapping: map keywords to CS subfields for venue prioritization

Important: The year range defaults to the last 3 years. If the user does not specify a year and the config has null, use current_year - 3 to current_year.

Step 2: Parallel Database Search

Query all configured databases simultaneously using webfetch. IMPORTANT: Request 2x preferences.max_results from each source to compensate for unpublished preprints that will be filtered later.

General Retry Strategy for all APIs:

• If 429 (rate limited): wait 5 seconds, retry once. If still fails, try the fallback endpoint.
• If 500/502/503 (server error): wait 2 seconds, retry once. If still fails, skip that source and note it.
• If timeout (>15s): skip and try fallback.
• If 403 (forbidden): skip immediately — this source requires authentication.

Data Coverage Note: Semantic Scholar and OpenAlex are meta-search engines that aggregate papers from hundreds of publishers including IEEE Xplore, ACM Digital Library, Springer, Elsevier, and others. Even when IEEE/ACM direct APIs fail, their papers ARE still represented in the results — you just get them through S2/OpenAlex instead of directly. Do NOT treat IEEE/ACM direct API failure as missing data.

2.1 arXiv API (with rate-limiting discipline)

arXiv enforces strict rate limits: no more than 1 request per 3 seconds. Violations result in 429 + temporary ban.

Query strategy — use a single targeted request:

URL: http://export.arxiv.org/api/query?search_query=all:{encoded_keywords}&start=0&max_results={2x_max_results}&sortBy=relevance

Parse XML response. Extract: id, title, summary, authors (list), published (date), doi, journalref, primarycategory, link.

Rate limit handling:

• If 429: wait 20 seconds (arXiv ban cooldown), retry once. If still fails, skip — S2 and OpenAlex already cover arXiv papers.
• Do NOT send multiple rapid requests to arXiv. One query per search is sufficient.

Published-only check: After parsing, flag papers that have a journal_ref or doi as PUBLISHED. Papers with neither are preprints — keep them for dedup but mark as unpublished candidate.

2.2 Semantic Scholar API (with rate-limit handling)

Primary endpoint — try this first:

URL: https://api.semanticscholar.org/graph/v1/paper/search?query={encoded_keywords}&limit={2x_max_results}&year={year_start}-{year_end}&fieldsOfStudy={topic}&fields=title,year,authors,externalIds,url,abstract,venue,citationCount,publicationTypes,openAccessPdf,publicationDate,journal

If API key configured, add: x-api-key: {key}

Fallback endpoint — use if primary returns 429:

URL: https://api.semanticscholar.org/graph/v1/paper/search/bulk?query={encoded_keywords}&year={year_start}-{year_end}&fields=title,year,authors,externalIds,url,abstract,venue,citationCount,publicationTypes,journal

No-key workaround — if both endpoints rate-limit and no API key is configured:

• Inform user: "Semantic Scholar API 请求被限。建议在 https://www.semanticscholar.org/product/api 免费注册 API key 以提高限额。"
• Skip Semantic Scholar, proceed with other databases.

Parse JSON response (data array). Extract: title, year, authors, externalIds (DOI, ArXiv), url, abstract, venue, citationCount, publicationTypes, journal.

2.3 DBLP API (with fallback)

Primary endpoint:

URL: https://dblp.org/search/publ/api?q={encoded_keywords}&h={2x_max_results}&format=json

Fallback endpoint — use author-title search if primary fails:

URL: https://dblp.org/search/publ/api?q={encoded_keywords}&h={2x_max_results}&format=json&c=0

Second fallback — use simpler search with f=0:

URL: https://dblp.org/search/publ/api?q={encoded_keywords}&h={2x_max_results}&f=0&format=json

If DBLP returns 500 after all retries, skip it — DBLP has known stability issues. Note in results.

Parse JSON. Extract from result.hits.hit[]: title, year, venue (from info.venue), authors (from info.authors.author[]), doi (from info.doi or info.ee).

2.4 PubMed E-utilities

Step 1 (search): https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term={encoded_keywords}&retmax={2x_max_results}&retmode=json&mindate={year_start}&maxdate={year_end}&datetype=pdat
Step 2 (fetch): https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esummary.fcgi?db=pubmed&id={comma_separated_ids}&retmode=json

Parse JSON. Extract: title, authors, pubdate, source (journal), doi, articleids.

2.5 IEEE Xplore (API key required)

IEEE Xplore has no public REST API. The search page aggressively blocks automated access (403). There are two paths:

Path A — With API key (configured in paper-config.json.api_keys.ieee):

URL: https://ieeexplore.ieee.org/rest/search
Method: POST
Headers:
  Content-Type: application/json
  Accept: application/json
  apikey: {ieee_api_key}
Body: {"queryText": "{keywords}", "returnType": "SEARCH", "highlight": false, "rowsPerPage": {2x_max_results}, "ranges": ["{year_start}_{year_end}_Year"]}

Path B — Without API key (most users): Skip IEEE Xplore direct search entirely. The IEEE papers ARE already covered by Semantic Scholar and OpenAlex — they index IEEE Xplore content including all conference proceedings and journals. Your results will include IEEE papers via these sources.

How to get an IEEE API key:

1. Visit https://developer.ieee.org/
2. Register for a free account (requires institutional email or IEEE membership)
3. Request API access — approval may take days
4. Add the key to paper-config.json: "api_keys": { "ieee": "your-key-here" }

If 403 or auth error: Skip and note "IEEE Xplore 需 API key 授权。IEEE 论文已通过 Semantic Scholar/OpenAlex 覆盖。"

2.6 ACM Digital Library (limited access)

ACM Digital Library blocks automated web scraping (403). Direct search is unreliable.

With institutional access (if user can authenticate): Try a single webfetch attempt:

URL: https://dl.acm.org/action/doSearch?AllField={encoded_keywords}&pageSize={2x_max_results}&startPage=0&AfterYear={year_start}&BeforeYear={year_end}

Parse HTML for paper entries if response is successful.

Without institutional access (most users): Skip ACM DL direct search entirely. ACM papers (SIGMOD, SIGCOMM, CHI, CCS, OSDI, etc.) ARE indexed by DBLP, Semantic Scholar, and OpenAlex — all three cover ACM venues comprehensively. No data is lost by skipping ACM direct.

If 403 or blocked: Skip silently. Note "ACM Digital Library 拒绝自动访问 — ACM 论文已通过 DBLP/Semantic Scholar/OpenAlex 覆盖。"

2.7 Google Scholar (webfetch — last resort fallback)

Only use if fewer than 5 results returned from all other databases combined.

URL: https://scholar.google.com/scholar?q={encoded_keywords}&as_ylo={year_start}&as_yhi={year_end}&num={max_results}

Parse HTML. Extract title, authors, year, venue, cited-by count.

Warning: Google Scholar frequently blocks automated access. If blocked or returns empty, skip silently — do not treat as error.

2.8 Additional Open Access Sources (if results are still insufficient)

OpenAlex API (free, no key needed):

URL: https://api.openalex.org/works?search={encoded_keywords}&filter=publication_year:{year_start}-{year_end},type:article|proceedings&per_page={2x_max_results}&sort=publication_date:desc

This is an excellent fallback — OpenAlex has broad coverage and is fully open access.

CORE API (free, no key needed):

URL: https://api.core.ac.uk/v3/search/works?q={encoded_keywords}&yearFrom={year_start}&yearTo={year_end}&limit={2x_max_results}

Step 2.5: Published-Paper Filter (CRITICAL)

After collecting raw results from all databases, apply this filter to keep ONLY published papers:

A paper is considered PUBLISHED if it has ANY of the following:

• DOI assigned (from any source)
• Journal reference present (arXiv journal_ref field)
• Venue + year clearly specified (conference name + year, or journal name + volume/issue)
• In DBLP (DBLP only indexes published CS papers)
• In PubMed with PMID (PubMed only indexes published papers)
• publicationTypes from Semantic Scholar indicates "JournalArticle" or "Review"

A paper is UNPUBLISHED (exclude) if:

• Only has arXiv ID but NO DOI, NO journal_ref, NO venue
• Only appears on Google Scholar without verifiable venue

Filtering logic: Remove all unpublished papers from the candidate list. Keep a count of how many were removed for reporting.

Exception: If after filtering, fewer than 3 papers remain, fall back to the broader candidate list and mark the report header: "⚠️ 大部分结果为预印本/未发表论文，仅有 {count} 篇已发表。"

Step 3: Deduplicate and Merge

After collecting results from all databases:

1. Normalize titles: lowercase, remove punctuation, strip extra whitespace
2. Match by DOI: exact DOI match → merge
3. Match by arXiv ID: exact arXiv ID match → merge
4. Match by title similarity: if title edit distance < 10% of length → likely same paper
5. Merge data: for matched papers, combine metadata from all sources, preferring:

• DOI from Semantic Scholar or CrossRef
• Citation count from Semantic Scholar
• Abstract from arXiv (usually most complete)
• Venue info from DBLP (usually most accurate)

Step 4: Venue Tier Annotation

For each paper, check its venue (conference or journal) against the tier database below. Assign ALL applicable tiers.

Tier Badge Format: 🏆CCF-A ⭐SCI-Q1 📊CAS-Q1

Tier Database

AI / Machine Learning

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | NeurIPS | Conference | A | — | — | | ICML | Conference | A | — | — | | AAAI | Conference | A | — | — | | IJCAI | Conference | A | — | — | | ICLR | Conference | A* | — | — | | UAI | Conference | B | — | — | | AISTATS | Conference | B | — | — | | ECAI | Conference | B | — | — | | ACML | Conference | C | — | — | | TPAMI | Journal | A | Q1 | Q1 | | JMLR | Journal | A | Q1 | Q1 | | AIJ | Journal | A | Q1 | Q1 | | Neural Networks | Journal | B | Q1 | Q1 | | MLJ | Journal | B | Q2 | Q2 | | Neurocomputing | Journal | C | Q1 | Q2 | | IJON | Journal | C | Q1 | Q2 | | Pattern Recognition | Journal | B | Q1 | Q1 |

Computer Vision

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | CVPR | Conference | A | — | — | | ICCV | Conference | A | — | — | | ECCV | Conference | B | — | — | | BMVC | Conference | C | — | — | | ACCV | Conference | C | — | — | | IJCV | Journal | A | Q1 | Q1 | | TIP | Journal | A | Q1 | Q1 | | CVIU | Journal | B | Q2 | Q2 | | PR Letters | Journal | C | Q3 | Q3 | | IET-CV | Journal | C | Q4 | Q4 |

Natural Language Processing

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | ACL | Conference | A | — | — | | EMNLP | Conference | A | — | — | | NAACL | Conference | B | — | — | | COLING | Conference | B | — | — | | CoNLL | Conference | B | — | — | | EACL | Conference | C | — | — | | CL | Journal | A | Q1 | Q1 | | TACL | Journal | A | Q1 | Q1 | | NLP Engineering | Journal | B | Q2 | Q2 | | ACM-TALLIP | Journal | C | Q4 | Q4 |

Computer Systems / Architecture

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | OSDI | Conference | A | — | — | | SOSP | Conference | A | — | — | | ASPLOS | Conference | A | — | — | | EuroSys | Conference | A | — | — | | USENIX ATC | Conference | A | — | — | | FAST | Conference | A | — | — | | ISCA | Conference | A | — | — | | MICRO | Conference | A | — | — | | HPCA | Conference | A | — | — | | SC | Conference | A | — | — | | PPoPP | Conference | A | — | — | | PODC | Conference | B | — | — | | ICDCS | Conference | B | — | — | | Middleware | Conference | B | — | — | | TOCS | Journal | A | Q1 | Q1 | | TPDS | Journal | A | Q1 | Q1 | | TC | Journal | A | Q1 | Q1 | | TACO | Journal | A | Q1 | Q1 | | JPDC | Journal | B | Q2 | Q2 | | FGCS | Journal | C | Q1 | Q2 |

Computer Networks

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | SIGCOMM | Conference | A | — | — | | NSDI | Conference | A | — | — | | MobiCom | Conference | A | — | — | | INFOCOM | Conference | A | — | — | | SenSys | Conference | B | — | — | | ICNP | Conference | B | — | — | | CoNEXT | Conference | B | — | — | | MobiSys | Conference | B | — | — | | TON | Journal | A | Q1 | Q1 | | JSAC | Journal | A | Q1 | Q1 | | TMC | Journal | A | Q1 | Q1 | | TWC | Journal | B | Q1 | Q1 | | ComNet | Journal | B | Q2 | Q2 |

Databases

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | SIGMOD | Conference | A | — | — | | VLDB | Conference | A | — | — | | ICDE | Conference | A | — | — | | PODS | Conference | B | — | — | | CIKM | Conference | B | — | — | | EDBT | Conference | B | — | — | | DASFAA | Conference | B | — | — | | TKDE | Journal | A | Q1 | Q1 | | VLDBJ | Journal | A | Q1 | Q1 | | ISJ | Journal | B | Q1 | Q1 | | DKE | Journal | B | Q2 | Q2 | | WWWJ | Journal | B | Q3 | Q3 |

Security / Privacy

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | IEEE S&P | Conference | A | — | — | | ACM CCS | Conference | A | — | — | | USENIX Security | Conference | A | — | — | | NDSS | Conference | A | — | — | | ESORICS | Conference | B | — | — | | RAID | Conference | B | — | — | | ACSAC | Conference | B | — | — | | ASIACRYPT | Conference | B | — | — | | TDSC | Journal | A | Q1 | Q1 | | TIFS | Journal | A | Q1 | Q1 | | Computers & Security | Journal | B | Q1 | Q2 | | JCS | Journal | B | Q2 | Q2 |

Software Engineering

| Venue | Type | CCF | SCI | CAS-JCR | |-------|------|-----|-----|---------| | ICSE | Conference | A | — | — | | FSE/ESEC | Conference | A | — | — | | ASE | Conference | A | — | — | | ISSTA | Conference | A | — | — | | OOPSLA/SPLASH | Conference | A | — | — | | ICSME | Conference | B | — | — | | SANER | Conference | B | — | — | | MSR | Conference | B | — | — | | TSE | Journal | A | Q1 | Q1 | | TOSEM | Journal | A | Q1 | Q1 | | EMSE | Journal | B | Q1 | Q1 | | JSS | Journal | B | Q1 | Q1 | | IST | Journal | B |

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: lyq777-Xing
• Source: lyq777-Xing/yqpaperquery_skill
• License: MIT

Install and usage instructions live in the source repository linked above.