Converts a heterogeneous corpus of raw documents (PDF, DOCX, DOC, PPTX, IPYNB, RTF, MD, TXT, HTML, etc.) into a structured, LLM-optimized knowledge base — per-source Markdown + manifest.json + INDEX.md + AGENTS.md, ready for ingestion in a separate Claude / Codex session. USE WHEN the user asks to ingest, index, preprocess, or build a knowledge base from a folder of mixed documents; "feed files to Claude", "prepare a corpus", "build a doc index", "RAG prep", "convert documents to markdown", "ingest Jupyter notebooks". RU triggers: "обработай папку с документами", "сделай базу знаний из папки", "подготовь корпус для LLM", "извлеки markdown из файлов". Output is for AI agents, not human reading. For single-file PDF use Anthropic's `pdf` skill.
Generate Russian academic reports (.docx) formatted to GOST 7.32 — лабораторные, отчёты по практике, курсовые, ВКР, домашние задания для любого российского вуза (ИТМО, МГУ, СПбГУ, МФТИ, Бауманка, и т.д.). Use whenever the user asks for a report по ГОСТ, лабораторную, отчёт по практике, курсовой, ВКР, или любую Russian-language student paper needing proper title page, headings, page numbers, figure/table captions. Triggers even when only "лабораторная" or "отчёт" is mentioned — Russian-language context (ИТМО / МГУ / СПбГУ / университет / ГОСТ) is enough. ITMO is the default profile; other universities via UniversityProfile.
Thesis-level literature research (систематический обзор, ВКР, related work) on a topic: queries 6 open academic APIs (OpenAlex, Semantic Scholar, arXiv, Crossref, Europe PMC, CORE), downloads OA PDFs via a multi-tier cascade, builds a persistent SPECTER-embedded SQLite corpus with PaperQA2-style citation traversal, flags retractions, and emits a markdown report with DOI-grounded citations and optional BibTeX/Zotero export. MUST BE USED whenever the user asks for a literature review, related work, deep research, систематический обзор, ВКР, дипломная работа, "найди все статьи про X", thesis prep, or research synthesis. NOT for single-paper summarization (use the `pdf` skill), code documentation (use repomix/gitingest), or general web research without academic grounding.