imozuru

What it does

• Field mapping before digging — two web-searching subagents with different approaches collect terms, central people, and venues into a map.md, so you are not stuck with the search words you first thought of
• Chain-following search — follows citations, citing works, authors, and links inside articles (chain.py uses the OpenAlex API, no key needed), and keeps going until every claim has a source or the dig saturates
• Sources stay raw — originals are saved to sources/ without summarising (404 pages and login walls dropped after checking), and the extract page lists cut passages claim by claim, each pointing to which file and which lines; reading and judgement stay with you
• Watch mode — after a dig, subscribes via RSS to the venues it surfaced (Qiita and Zenn tags, note, YouTube, blogs, arXiv) and lists new posts

When to use it

Due diligence where the summary is not enough — auditing where a claim actually comes from, getting a footing in an unfamiliar field, or keeping a standing watch on niche venues. Ask in the style of "dig into this:" and let it surface the sources before you read them yourself.

Getting started

/plugin marketplace add MaryCache/imozuru
/plugin install imozuru@imozuru
# or: npx skills add MaryCache/imozuru

The plugin carries no version number — every commit to main counts as a new version. Turn on auto-update for imozuru in /plugin's marketplace screen, or update by hand: claude plugin marketplace update imozuru

Things to know

The skill's SKILL.md, script messages, and extract-page labels are in Japanese, and the author notes it has only been tried with Japanese requests — Claude reads the instructions in any language, but expect a Japanese-flavoured workflow. Scripts run on the Python 3.10+ standard library alone; PDF text/OCR (pdftotext, pdftoppm + tesseract with jpn data), YouTube captions (yt-dlp), and HTML-to-Markdown (ax) need optional commands, and those sources report FAIL without them.