Resources

Only one thing on this page is required: an AI coding agent, installed and logged in before you start the Day 1 workflow. Everything else here is optional equipment: tools and skills that can make your agent more capable during the week.

// AI Coding Agent (required)

An AI coding agent is a chat assistant that runs in your terminal and can read files, write code, and run programs. The harness assumes you have one and can talk to it. Install one of:

To verify it works, start the tool in any folder and ask it to read a file.

// Learning Resources

• The Complete Guide to Building Skills for Claude (PDF) — what skills are and how to write your own.
• Agentic Coding — MIT Open Course (Video) — how to work effectively with a coding agent.

// MCP Servers & CLI Tools

Two ways to give your agent extra abilities:

• An MCP server (Model Context Protocol) plugs an outside service into your agent through a standard interface — for example, letting it search arXiv or drive a Jupyter notebook.
• A CLI tool is an ordinary command-line program; once installed, agents are good at calling these on their own.

Below is a subset curated for many-body physics simulation; GitHub star counts are a rough popularity signal. Install what your project needs — none of these are required to start. For an exhaustive list across research domains, see the HKUST-Got-Skills resources page.

Literature & Papers

• arxiv-mcp-server — MCP: search and analyze arXiv papers (2.4k stars)
• paper-search-mcp — MCP: multi-source search (arXiv, Semantic Scholar, Crossref, OpenAlex) (900+ stars)
• arxiv-latex-mcp — MCP: fetch arXiv LaTeX source so the agent reads equations exactly (110+ stars)
• doi2bib — CLI: convert DOI/arXiv ID to BibTeX (66+ stars)
• arxiv-dl — CLI: download papers by URL or arXiv ID (55+ stars)

Reference Management

• zotero-mcp — MCP: search, summarize, export BibTeX from Zotero (2.1k stars)
• zotero-cli — CLI: Zotero web API from the terminal (310+ stars)

Jupyter & Notebooks

• jupyter-mcp-server — MCP: full notebook control (960+ stars)
• nbconvert — CLI: convert notebooks to PDF, HTML, LaTeX
• nbstripout — CLI: strip notebook output for clean git diffs

Scientific Computing

• mcp.science — MCP: Materials Project, GPAW DFT, sandboxed Python (120+ stars)
• mcp-wolframalpha — MCP: Wolfram Alpha (75+ stars)
• Wolfram MCP Service — MCP: official Wolfram cloud service

Plotting & Visualization

• mcp-server-chart — MCP: 25+ chart types via AntV (3.9k stars)
• gnuplot — CLI: scientific plotting from the terminal
• ffmpeg — CLI: encode media, build animations from frame sequences
• imagemagick — CLI: convert, resize, annotate images

SSH, Remote & HPC

HPC means high-performance computing — shared clusters for jobs too large for a laptop. Useful if your track’s challenge needs one.

Compute for the week runs on SCNet 超算互联网, China’s national supercomputing platform. Accounts are handed out at the on-site help desk — register there to get login-node, scratch, and queue access. Compute-heavy tracks such as CTMRG and QMC should do this on Day 1.

• ssh-mcp — MCP: run commands on remote machines over SSH (350+ stars)
• sbatch / squeue / scancel — CLI: submit and manage jobs on SLURM clusters
• rsync / scp — CLI: efficient file transfer
• tmux — CLI: persistent terminal sessions (survive SSH disconnects)
• GNU parallel — CLI: run commands in parallel across cores/nodes
• rclone — CLI: sync to/from S3, Drive, Dropbox, 70+ providers

Git, GitHub & Data Versioning

• github-mcp-server — MCP: official GitHub server — code search, PRs, issues, CI (28k stars)
• gh — CLI: official GitHub tool — issues, PRs, actions, releases
• git-lfs — CLI: version large files in Git
• dvc — CLI: version control for datasets and pipelines

Build & Dev Utilities

• uv — CLI: fast Python package/project manager (82k stars)
• ripgrep (rg) — CLI: fast recursive text search (62k stars)
• fd — CLI: fast find alternative (42k stars)
• hyperfine — CLI: statistical command-line benchmarking (28k stars)
• ruff — CLI: fast Python linter + formatter (47k stars)
• make / cmake / ninja — CLI: build automation
• duckdb — CLI: analytical SQL on CSV/Parquet/JSON (37k stars)
• jq — CLI: JSON slice/filter/transform (34k stars)

// Skills for Researchers

A skill is a folder with a SKILL.md instruction file that teaches an AI agent a workflow — the harness commands like /beginner-training are skills. Beyond the ones the harness ships, you can install more: drop a skill folder into .claude/skills/ inside your project and Claude Code picks it up.

Install a Skill

# Option 1: clone into your project's .claude/skills/
git clone https://github.com/anthropics/skills.git /tmp/skills
cp -r /tmp/skills/skills/pdf .claude/skills/pdf

# Option 2: skills CLI (12k stars)
npx skills install anthropics/skills/pdf

# Option 3: skillkit (680+ stars)
npx skillkit@latest init

Recommended Collections

Especially Relevant for Many-Body Simulation

• test-driven-development (superpowers) — turn every numerical claim into a passing test before you trust it.
• systematic-debugging (superpowers) — find the root cause when a simulation misbehaves, instead of guessing.
• pdf (anthropics) — read the reference paper and your reproduction target.
• qiskit / cirq / pennylane / qutip (K-Dense-AI) — quantum computing toolkits.
• matplotlib / scientific-visualization (K-Dense-AI) — publication-quality plots.

Skill Tutorials

• Claude Code Skills Docs — official documentation
• Anthropic Guide to Building Skills (PDF)
• Claude Skills Deep Dive

// Background Reading

Each track’s tracks/<track>/README.md in quantum.harness names its reference paper and reproduction targets; rendered copies of the key literature live under .knowledge/literature/ in the same repository. More material lands as track leads finish their harnesses.