taisei-ide-0123/harvest

harvest is a command line tool that bulk downloads blog articles and saves them as Markdown files, including articles behind a login or paywall. It is written in TypeScript and currently supports note.com, a Japanese blogging platform, with the README noting that adding support for other sites is straightforward through pull requests. You install it by cloning the repository, running npm install and npm build, linking the CLI package, and installing a Chromium browser through Playwright. To download from a site that requires login, you first run a login command that opens a browser window so you can sign in, and it saves that session for later use. Then a download command pulls every article from a listing page, such as a magazine or author page, and writes each one to an output folder, with options to control output location, concurrency, delay between requests, and a maximum number of articles. Internally, harvest fetches pages in one of two ways. An extension based fetcher connects to a Chrome browser extension over a WebSocket connection and reuses the browser's existing cookies, so no separate login is needed. A profile based fetcher instead launches Playwright with its own persistent browser profile and can run headless, which the README says suits automated agents and continuous integration. By default harvest tries the extension method first and falls back to the profile method automatically. Each supported site has its own adapter that knows how to find article links and parse article content into Markdown, complete with a header containing title, date, source link, author and tags. Beyond downloading, harvest ships with optional skills that let AI coding agents build a searchable knowledge base or generate interactive HTML slide decks from the downloaded articles. These are installed with a setup command and support Claude Code directly, plus a generic AGENTS.md file for tools like Codex, Cursor and Windsurf. The README shows an example workflow of downloading a set of articles, installing the skills, then using an agent to initialize a wiki, ingest the articles in batches, query the resulting knowledge base, and generate slides from a chosen topic. The project is released under the MIT license.