Claude skills for Research

4 skills in the bank are filed under Research. They are grouped here because they compete for the same jobs — which is the reason an agent reaching for one of them sometimes gets another.

• AI Research ReproductionlllllllamaFaithfully re-runs a published deep-learning repository to verify its documented results — reading the README first, picking the smallest honest inference or evaluation target, executing it under strict no-silent-edits patch rules, and logging evidence, deviations, and human sign-off points into a standardized outputs bundle. It is about REPRODUCING someone else's already-published numbers, not summarizing a paper, gathering fresh sources, or drawing business conclusions.
• Market Researchaffaan-mReads the competitive and market landscape to inform a business decision — TAM/SAM/SOM sizing, competitor and product comparisons, investor and fund diligence, and technology scans — with every claim sourced, contrarian evidence included, and the output ending in a recommendation rather than a summary. It ANALYZES the market you are entering or funding, not reproducing research results, synthesizing user interviews, or organizing your own research files.
• Research PapersfirecrawlRuns a literature review: discovers and synthesizes published scholarly work on a topic — journal papers, preprints, whitepapers, and technical reports — through semantic search over a paper index (PubMed, bioRxiv, medRxiv, arXiv), expanding from seed papers into the surrounding citation family and verifying claims inside individual PDFs. It is about SURVEYING what the published literature already says on a subject, not re-running one paper's code to check its numbers, compressing your own interviews and tickets into themes, or sizing a commercial market.
• Synthesize ResearchanthropicsTurns a pile of already-collected qualitative input — interview notes, survey responses, support tickets, sales-call feedback — into one structured read: themes grouped and counted, findings ranked by frequency and impact, and recommendations tied back to the evidence. It is about COMPRESSING many scattered sources into a single coherent synthesis, not running an experiment, sizing a market, or filing and circulating the raw materials.