science-plugins
A personal Claude Code plugin marketplace for mathematical computing, scientific research, and developer productivity. Add the marketplace once, then install any plugin with a single command.
What this is
Claude Code can be extended with plugins — bundles of agents, commands, skills, and hooks that add new capabilities to your sessions. This repository is a marketplace catalog: it tells Claude Code where to find plugins and what they contain. Point Claude Code at this catalog once, and from then on you can install any listed plugin with /plugin install.
The plugins here are focused on quantitative research and developer productivity, and they're designed to work together. wolfram-hart handles symbolic math. compound-science handles econometric workflows. claude-codex routes mechanical tasks to a cheaper model. ralpha-team drives long-running jobs to completion without babysitting.
How plugins work
Claude Code plugins have four types of components:
- Skills are passive triggers. When your prompt matches a skill's description, Claude loads that skill's instructions and follows them without any explicit invocation from you.
- Commands are slash commands you call directly (e.g.
/estimate, /eval). They run a structured prompt with defined tools and argument handling.
- Agents are subprocesses Claude spawns to handle a focused task (research, review, implementation), each with its own context window.
- Hooks run in response to Claude Code lifecycle events (session start, prompt submit, tool call) without being invoked at all.
Most real-world behavior comes from combining these. A hook detects context, a skill loads additional instructions, a command handles the interface for the heavy lifting.
Add this marketplace
/plugin marketplace add James-Traina/science-plugins
This downloads the catalog and registers it locally. You don't need to reinstall the marketplace when plugins are updated.
Plugins
Gives Claude a real computer algebra system. Without this, Claude approximates symbolic math: it reasons about it rather than computing it. This plugin connects Claude to wolframscript, the command-line interface to Wolfram Language (the engine behind Mathematica and Wolfram Alpha).
When a math or science question comes up, the included skill triggers. Claude translates the problem to Wolfram Language, runs it through wolframscript, and shows you the computed result — the actual output, not a reconstruction from training data. Plots are saved as PNGs and shown inline. You never need to write any Wolfram Language yourself.
Coverage includes symbolic algebra, calculus, linear algebra, statistics, differential equations, number theory, combinatorics, and plotting. The three commands give you direct control: /eval runs Wolfram Language code you supply; /check verifies a mathematical claim step-by-step; /patterns lists the problem types the skill handles. The wolfram-reviewer agent audits a derivation or proof, running each step through the engine to check it independently.
Components: 1 skill · 3 commands (/eval, /check, /patterns) · 1 agent (wolfram-reviewer)
Prerequisites: A free Wolfram Engine (local, ~1 GB) or wolframscript for cloud evaluation. Requires a free Wolfram ID.
/plugin install wolfram-hart@science-plugins
A research workflow toolkit for structural econometrics, causal inference, game theory, applied micro, identification arguments, Monte Carlo studies, and reproducible pipelines. Inspired by compound-engineering: the same plan → work → review → compound loop, adapted for quantitative social science.
The organizing idea is that solved problems should stay solved. Every methodological fix — a convergence workaround, an identification argument that held up, a numerical issue and how it was handled — gets documented and made searchable in docs/solutions/. The next project starts where the last one left off. Over time you build a searchable record of hard-won solutions, and Claude can draw on it when facing a similar problem.
James-Traina