research-literature-review

Provide a topic, an optional time range, and a tier (Premium, Standard, or Basic). The skill generates 5-25 search queries, retrieves papers from OpenAlex and other sources, scores each paper's relevance, selects references, and writes a validated review exported as PDF and Word.

It searches OpenAlex as the primary source, with automatic fallback to Semantic Scholar, Crossref, and DuckDuckGo based on a configurable provider priority list. Rate limiting, retries, and API health monitoring are built in.

Yes. The skill supports English, Chinese, Japanese, German, French, and Spanish. It translates the finished review while preserving all \cite keys and LaTeX structure, then recompiles with appropriate language packages such as ctex or babel.

Premium targets 10,000-15,000 words with 80-150 references, Standard targets 6,000-10,000 words with 50-90 references, and Basic targets 3,000-6,000 words with 30-60 references. All ranges can be overridden in the prompt.

The skill uses a fail-closed query contract: if the multi-query JSON is missing, malformed, or outside the 5-25 query range, stage 1 stops instead of silently falling back to a single query. Single-query mode requires an explicit --allow-single-query-fallback flag.

You need Python 3.9+ with pyyaml, requests, numpy, scikit-learn, and matplotlib, plus a LaTeX installation with xelatex and bibtex, and pandoc for Word conversion. Internet access is required for the academic search APIs.