AI skills for Research

Explores AI research repos and papers systematically before reproduction or training runs.

Prepares a conservative environment and assets bootstrap for README-first machine-learning reproduction.

This skill uses the Parallel FindAll API to list every company, person, product, or other item that matches a plain-language description. You can ask for extra fields on each match, and every result comes with evidence. It returns a list, not web pages or a narrative report. The Parallel CLI is required.

Fine-tunes language and vision models with TRL or Unsloth on Hugging Face Jobs.

Creates and edits Jupyter notebooks for analysis, demos, and reproducible research workflows.

This skill combines results from several search sources into a single answer with duplicates removed. Each claim keeps a citation, confidence depends on how fresh and authoritative the source is, and conflicting information is called out. It changes how it summarizes depending on how many results come back.

Captures and normalizes a minimal successful research run for auditability.

Scans a research repository and drafts a README-first reproduction plan before coding.

Runs structured engineering research—options, tradeoffs, and evidence—before committing to an approach.

This skill starts a deep-research job through the Parallel CLI, waits for it to finish, and returns the report with citations. It is only for requests that explicitly ask for exhaustive research, because it takes longer and is pricier than an ordinary Parallel web search. Follow-up questions can reuse an earlier task.

This skill pulls the content of a known URL with the Firecrawl CLI, as Markdown, links, or a direct answer to a question about the page. It can wait for JavaScript, keep only the main content, and process several URLs at once. It can also run a Firecrawl data tool once one is chosen.

This skill runs web searches with the Firecrawl CLI, giving back matching excerpts, with full page text when needed. It can limit results to recent news or certain domains and look for datasets, listings, and transcripts. It can also find and call Firecrawl's data tools by category.

This skill uses the Defuddle CLI to pull the readable part of a web page and drop menus, ads, and other clutter. It outputs Markdown, JSON, or HTML and can return single fields such as title, description, or domain.

This skill uses Firecrawl to search and read the web on a complex topic, then writes a formal report with sources. The report has an executive summary, key findings, detailed analysis, contrarian views and risks, and open questions. It is meant for hard scientific, technical, policy, or market questions, not quick lookups or product picks.

This skill searches Firecrawl's paper index and writes a literature review on a topic. The index leans toward biomedical and life-science work, with PubMed, bioRxiv, and medRxiv as sources, and it also holds arXiv papers in computing, physics, and mathematics. It expands to related papers and checks claims in the paper text. The review covers key papers, themes, debates, trends, and sources.