kengo006

Alexandria — Data skill for Claude Code

Data community

A citation-integrity-first academic writing system for Claude Code and Obsidian — six roles, a verbatim-to-source quote pipeline, and the governance that keeps it honest.

How to install Alexandria

This entry records only its repository, not the path inside it, so there is no exact command to give. Open kengo006/alexandria and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Alexandria does

A citation-integrity-first academic writing system for Claude Code and Obsidian — six roles, a verbatim-to-source quote pipeline, and the governance that keeps it honest. 為 Claude Code 與 Obsidian 設計、以引文完整性為先的學術寫作系統：六個角色、逐字回到原始文獻的引文管線，以及讓它保持誠實的治理層。

Alternatives in Data

Portaljs Add Dataset — Add a dataset (CSV, TSV, JSON, or GeoJSON) to an existing PortalJS portal 2.3k ★
Autonovel — An autonomous novel writing pipeline, by Hermes Agent 1.5k ★
Web Agent — 🔥 Open-source web data agent optimized for structured web research 1.2k ★

README

Alexandria

**A citation-integrity-first academic writing system for Claude Code and Obsidian.**

[中文簡介](#中文簡介)

Large language models fabricate citations. A recent [cross-model audit](https://arxiv.org/abs/2603.03299) of ten models measured reference-fabrication rates between 11.4% and 56.8%, and most tooling attacks the problem *after* the text is written, by detecting hallucinated references. Alexandria attacks it *before*: it is a six-role writing system whose workflow makes fabrication structurally difficult. Every verbatim quote must be read back from the source PDF at a real page number, pass four verification layers, survive a blind review, and be audited line-by-line before a draft is allowed to call itself done.

Alexandria is not a library or a server. It is a set of role definitions, methods, and governance files you drop into [Claude Code](https://claude.com/claude-code), pointed at your own [Obsidian](https://obsidian.md) vault.

**In thirty seconds.** **You need**: Claude Code, and a folder with your PDFs in it. No server, no database, no key beyond the one you already have; Obsidian and the Python scripts are for later tiers. **You do**: copy two role files into `.claude/skills/` and two agent files into `.claude/agents/`, keep the evidence handbook where they can read it, point them at `sources/` and `notes/`, and ask the Writer for a paragraph. That is Tier A, and the whole citation pipeline runs at it — [GETTING-STARTED.md](GETTING-STARTED.md) has the exact steps. **You see**: [`examples/`](examples/) — a dispatched Searcher's report, a blind review, and the audit that refuses to sign off, all fabricated for illustration. **Everything below this line is why it is built this way**; if you would rather find out by running it, the two links above are the whole path in.

**Where it comes from.** I am a graduate student in Taiwan working at the intersection of political philosophy and AI ethics. Alexandria is