smithy-org/bishop-and-barrister

A Claude skill that answers Magic: The Gathering rules questions from the official Comprehensive Rules and
current Oracle card text instead of from memory, and cites the rules it relies on.
## Install
Clone it into your Claude Code skills folder:
bash
git clone https://github.com/smithy-org/bishop-and-barrister ~/.claude/skills/bishop-and-barrister
Then fetch the rules documents once: the Comprehensive Rules, the Magic Tournament Rules, the Infraction
Procedure Guide and Judging at Regular REL. They are not bundled, because they belong to Wizards of the Coast
and change several times a year. The three tournament documents are PDFs, read with pypdf:
bash
pip install pypdf
python ~/.claude/skills/bishop-and-barrister/scripts/sources.py update
Optionally, build the rulings record now; otherwise the skill does it the first time it needs it. This downloads
Scryfall's official rulings (about 5 MB) into ~/.claude/bishop-and-barrister/, a git repository that keeps
every ruling the judge makes, with its history:
bash
python ~/.claude/skills/bishop-and-barrister/scripts/judge.py build
Needs Python 3.10 or later, git, and pypdf for the tournament documents.
### Challenges (optional)
Everyday answering needs nothing more. To let a user challenge a ruling or argue for a new rule, the skill uses a
small program, bab-challenge, built for Linux, macOS and Windows. The first time a challenge runs, the skill
downloads the one for your machine from this repository's latest release and refuses it unless its SHA-256
matches the one recorded in challenge/binaries.json. If there is none for your machine, or no network,
challenges say so and everything else keeps working. A challenge uses a few million tokens at most, so the skill
tells you the cost and asks first. See CHALLENGE.md.
### On claude.ai
Zip this repository and add it as a custom skill in claude.ai's skills settings. This has not been tested
there yet. The skill fetches card data from scryfall.com and the rules documents from media.wizards.com, and a
challenge downloads its program from GitHub; if the sandbox blocks any of them, the script that needed it
reports the failure and the rest still works. Without independent subagents, claude.ai can open and refute a
challenge as a single-agent panel but cannot introduce a rule.
### Optional: sync across your devices
By default everything stays on your machine. To share your rulings between devices, point the skill at a
RustFS you control, on your own network or a remote server (1.0.4 or later; any S3-compatible store should
work). The bucket must be private. Put the keys in the environment, never in a file the skill writes:
bash
export BISHOPANDBARRISTERS3ACCESSKEY=... # or a profile in ~/.aws/credentials (--profile)
export BISHOPANDBARRISTERS3SECRETKEY=...
python ~/.claude/skills/bishop-and-barrister/scripts/judge.py sync --setup --endpoint https://rustfs.example.net --bucket bishop-and-barrister --create
python ~/.claude/skills/bishop-and-barrister/scripts/judge.py sync
After setup it syncs by itself: searches pull other devices' rulings in (at most every 10 minutes) and every