imozuru: research an unfamiliar field down to the original sources
Added NewCopyable prompt below
Published Last verified Released
Our verdict
Use it when a quick A.I. summary is not enough and you need to check what a claim rests on. Instead of reading the top few search results, imozuru first learns the field's terms, key people and venues, then follows references, citing papers, authors and blog links until every point of your question has a source. It saves the sources and tells you which file and which lines support each point. Builders use it to fact-check viral claims, find papers they had missed in a neighbouring field, and dig through interview articles; one ported it to an MCP server for Hermes Agent and another adapted it for Codex. Two costs: it is slow and token-hungry (one builder measured 360k tokens and 22 minutes against 178k and 6 minutes for a plain request, with 9 of 9 spot-checked citations holding up), and its instructions, script messages and dashboard labels are in Japanese, though it accepts requests in other languages.
- Who it's for
- Founders, analysts, writers and builders using Claude Code who need to get up to speed on an unfamiliar topic and want every claim traceable to a source
- Leverage
- A saved, line-referenced source pack and a layered explainer for a new field, instead of a summary you cannot check
- Effort
- Medium
- Date added
What it does
imozuru (MaryCache/imozuru, MIT) is a Claude Code skill and plugin based on information-seeking research from library science (Bates's berrypicking, Ellis, Kuhlthau). It writes the question and 3 to 6 points needed to answer it, sends two web-search subagents with different approaches to map terms, people and venues into map.md, then follows citations (via the free OpenAlex API, no key), links and authors. Sources are saved to sources/ with a label (paper, first-hand, observation, roundup, official), and 404s and login walls are dropped. It stops when every point has a source or when 70% of new finds are already on the map. Version 2.0 (25 October 2026) adds a detailed report at three reading levels, where a script checks every quoted sentence against the saved sources before building the page, plus a dashboard of all your investigations that you can open on a phone over Tailscale. A watch mode (/imozuru watch <topic>) collects new posts from the venues a dig found via RSS. The scripts need only Python 3.10+; pdftotext, tesseract, yt-dlp and ax are used when installed.
How builders use it
- Install it as a Claude Code plugin (/plugin marketplace add MaryCache/imozuru, then /plugin install imozuru@imozuru) or with npx skills add MaryCache/imozuru.
- Ask it to dig into a topic you need to decide on (a technique, a market, a standard) and read the extract page before trusting the explainer.
- Open the sources and line numbers it points to for the claims that matter, rather than re-searching.
- Run /imozuru watch <topic> afterwards to follow new posts from the venues it found.
- Budget for it: run it on questions that deserve a deep dig, not quick lookups.
What to expect
A folder per topic with the saved sources, a map of the field, a searchable log of every query, an extract page and a three-level explainer whose quotes have been checked against the sources.
Copy this: Install + first dig
# Official source: github.com/MaryCache/imozuru (MIT). Needs Python 3.10+.
# In Claude Code:
/plugin marketplace add MaryCache/imozuru
/plugin install imozuru@imozuru
# or: npx skills add MaryCache/imozuru
# Optional helpers (used when installed): poppler (pdftotext), tesseract, yt-dlp
# Ask (the README's own example, in English):
Dig into this: I keep hearing about Gaussian Splatting but have no idea what it is. Dig through the papers and explainers.
# Results: ~/research/<topic>/ (map.md, sources/, report.html, digest.html)
# Dashboard of every investigation:
python3 skills/imozuru/scripts/dashboard.py ~/research
# Follow new posts from the venues it found:
/imozuru watch <topic>Related discoveries
See the newest picks →- ToolResearch & Automation
Open Deep Research (LangChain)
An open-source, configurable deep research agent you can run and modify.
Effort: High#Open source
- ToolResearch & Automation
GPT Researcher
An autonomous agent that researches any topic and writes a report, with any LLM.
Effort: Medium#Open source
- ToolResearch & Automation
Gemini Deep Research
Hand off a research question and get back a multi-source report.
Effort: LowPrompt
- ToolResearch & Automation
Perplexity API Platform
Real-time, web-grounded search and answers via API for your own products.
Effort: Medium#API
Want more like this?
Practical discoveries, without the noise.