Browse / Research & science / Teaching & study / ai-research-claude-tools
ai-research-claude-tools
ShuyangDen
My personal claude code workflow, from an Econ PhD student in Binghamton University
Open ShuyangDen/ai-research-claude-tools → Checked against the four rules.
Works with any agent
ships AGENTS.md, the cross-tool instruction file.
In its own words
A local-first research workflow for finding papers, reading them, preserving source-grounded knowledge, and developing research ideas without mixing personal data into the public tools repository.
Components
The repository contains tools, schemas, templates, synthetic fixtures, and packaging metadata only. Personal ideas, paper notes, profiles, queues, feedback, source records, and credentials remain in the machine paths configured by the user.
Closed loop
Deterministic state, hashes, retries, and writes belong to Research Core. LLM workflows interpret research content but do not own transaction state. Ordinary idea chat remains single-agent; multi-agent S2/Challenge behavior is reserved for a later A/B evaluation.
Installation and upgrade
Clone the repository, then follow INSTALL.md. The installer/sync tool reads a private ~/.claude/machine_paths.md, backs up system targets, protects personal-data paths, and records source/destination hashes.
Run doctor and sync in check/dry-run mode before applying an upgrade. Never copy package skeleton data over a personal vault.
Requirements
- Python 3.9+
- Claude Code and/or Codex
- Package-specific Python requirements for PDF/email/tracker features
- Zotero and GitHub are optional integrations
Quoting 48% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.