What is deep-research?

This repository is a plugin, called a "skill," for the OpenCode AI coding tool. Its purpose is to run a structured research pipeline that produces a long, cited report on any topic you give it. You trigger it with a single slash command and the process runs automatically for about six minutes without further input. The output is a Markdown file containing roughly 15,000 Chinese characters, organized into eight or more chapters, each with numbered paragraphs, data tables, cited sources, and a section presenting counterarguments. The pipeline works in four stages. First it analyzes your topic and generates a search plan. Second it runs web searches using a service called Exa as the primary engine, falling back to DuckDuckGo, Bing, Semantic Scholar, and a set of Chinese-language sources including Zhihu, Baidu Baike, industry databases, and national statistics sites. A Python library called Scrapling fetches the full text of pages rather than just summaries. Third, multiple chapters are written in parallel by passing the collected facts directly into the language model prompt. Fourth, a validation and assembly step checks each paragraph, converts citation formats, and runs a quality pass before saving the file. Three depth modes are available: a quick mode that caps output at about 8,000 characters and takes five to eight minutes, a standard mode targeting 15,000 characters in six to ten minutes, and a deep mode going up to 25,000 characters in twelve to eighteen minutes. All processing stays on your local machine. No data is sent to external services beyond the search queries and language model API calls you already use. Although it was built for OpenCode, the README notes that the same approach works with Claude Code, Codex CLI, Cursor, Windsurf, and other AI coding tools with some adaptation. Installation can be handed to the AI tool itself by pasting a prompt that tells it to read the project documentation and set everything up.