Paper Search Pro

The problem it addresses

Literature searching is rarely about finding one paper; it is about narrowing a broad topic into a defensible set of candidate studies. For thesis proposals, coursework, systematic reviews, scoping reviews, or meta-analyses, the main pain points are usually keyword coverage, incomplete source coverage, and manual relevance screening. Abstracts often need to be compared against PICO, SPIDER, MeSH, RCT, or other research designs, and a single database can miss important studies. paper-search-pro frames the task as an orchestrated search workflow instead of a one-off query or a PDF download job.

How the skill works

• Multi-source search: it organizes retrieval across five data sources, which helps with academic topics that require broader coverage.
• Adjustable depth: four tiers allow switching between quick topic scanning and deeper research passes.
• Scripted deterministic work: Python helpers handle stable engineering steps such as parsing, organizing, or repeatable operations.
• Delegated classification: LLM classification is delegated to parallel Inline SubAgents to judge candidate relevance.
• No external API key: the documented workflow does not depend on an external API key, which fits environments where the platform or local capabilities are already available.

The skill acts more like a literature-search orchestrator: the main agent interprets the question, chooses sources and depth, then coordinates retrieval and classification. That structure is especially useful when the query is academic and needs a defensible shortlist, rather than a single known article.

Where it does not fit

Use it for finding papers and scoping research, not for reading a specific paper, summarizing a single known paper, downloading PDFs from DOIs, or writing a review from an existing literature set. Concept explanations such as "what is prospect theory" should be answered directly without invoking a search workflow. In practice, specify the topic, time range, study design, and preferred sources when possible; otherwise results will depend heavily on keyword phrasing and source coverage. The output is a candidate literature set with relevance signals, not a finished conclusion.