GregorStocksCommunity·10 Agent Skills Included

mage-bench

Benchmark language models through competitive Magic: The Gathering matches

Benchmarks language models by having them play full Magic: The Gathering duels and Commander games against each other on a real rules engine. Records every game with video, logs, and decision data, then ranks models on a public leaderboard. Includes analysis workflows that review gameplay quality, detect platform bugs, and improve error messages without manual log digging.

64|8

↗

npx skills add GregorStocks/mage-bench --all -g -y

Available:

Gives the agent strict operating rules for this repo, including build and test commands, fail-fast coding style, CI flake handling, and pull request procedures.

All Skills in This Repository (10)

Pure Emerald Level Indicators

📦 In Repo

GregorStocks

blunder-backfill

Backfill outdated blunder analyses for recent MageBench games using Sonnet 4.5.

fast-analysis

Analyze exported game files to assess gameplay, model decisions, and likely bugs.

golden-test

Automate creation and maintenance of stable golden prompt tests for Puppeteer-based tests.

card-finder

Find Magic: The Gathering cards for compact golden-test scenarios.

new-export-version

Create a new game export schema version, migration module, and tests.

solve-issue-local

Resolve mage-bench issues with a standardized build, test, and PR workflow.

deep-analysis

Trace mage-bench game log bugs to source code and file issue summaries.

investigate-golden-flake

Diagnose failing golden tests by analyzing CI logs and diffs.

pr

Validate the full diff and checks, then open a pull request for the current branch.

find-improvable-errors

Analyze MCP tool error messages in magebench game exports to identify least helpful ones.

Community

Advanced

Frequently Asked Questions

FAQPage Schema

How to install mage-bench?▼

Run `npx skills add GregorStocks/mage-bench --all -g -y` in your terminal to install all skills in this suite globally.

What does mage-bench measure?▼

It measures how well language models play Magic: The Gathering by running full games on the XMage rules engine and ranking models on a public leaderboard.

How do models play Magic games in mage-bench?▼

Models connect through MCP tool calls exposed by a Java bridge, seeing the board state and choosing actions while the engine enforces all 28,000+ card rules.

Can I analyze recorded games for bugs?▼

Yes. The fast-analysis and deep-analysis skills review exported game files and raw logs to assess decision quality and trace platform bugs to source code.

Do I need API keys to run mage-bench?▼

No. The default configuration runs a two-CPU Jumpstart duel locally with no API keys; keys are only needed to benchmark hosted models.