Two tools that check your Claude Code plugins for problems. Open source, free to run, built while maintaining a plugin fleet in production.
A plugin is a folder of written instructions telling Claude how to do a job — and nothing checks instructions. No spell-check, no compiler, no test that goes red. Someone edits a sentence, the plugin quietly gets worse, and you find out when a user complains.
| Your problem | Use | What it costs |
|---|---|---|
| "Does my plugin still answer correctly after we edited it?" | 100xeval — testcases | ~$1–2 per run |
| "Is anything obviously wrong with my plugin?" | 100xeval — static check | Nothing. No key, no internet |
| "A fix in one plugin never reached the others." | 100xdrift-check | Claude usage per review |
100xeval is testcases for plugins. You save the questions your plugin must get right and run them after every change, so quality survives months of edits and more than one person editing them. It also ships a free static check — a quick run-free pass over the files, useful on every commit, but it cannot tell you whether the plugin still answers correctly. Only a case does that.
1. Clone this repo:
git clone https://github.com/100xopensource/100xtools.git
cd 100xtools2. Register it and install what you need:
claude plugin marketplace add ./
claude plugin install 100xeval@100xtools
claude plugin install 100xdrift-check@100xtoolsThen just ask. Open the folder your plugin is in and say what you want in plain words:
"static-check my plugin" · "why did it score 0.77?" · "what should I fix first?"
New to this? 100xeval's README walks the same path more slowly, with a troubleshooting table.
- Python 3.11+ — check with
python3 --version. Nothing else to install; these tools use only the standard library. - Claude Code or the Claude desktop app — for anything that runs your plugin.
The plugin READMEs cover how to run things. Why the pieces are shaped as they are — what a
grader is, what a design_score of 0.68 means — lives in [docs/](./docs/index.md), written
in Open Knowledge Format
v0.2: one small file per concept, cross-linked, and readable by an agent that wants a single
idea rather than a whole guide.
.claude-plugin/marketplace.json the marketplace manifest (one entry per plugin)
CHANGELOG.md releases + the scoring-version contract
docs/ OKF knowledge bundle — concepts, not how-to
plugins/
├── 100xeval/ eval engine + skill
└── 100xdrift-check/ two install skills + the reviewer and workflow they install
scripts/check_docs.py OKF bundle conformance + link check (runs in CI)
Each plugin is self-contained: copy a single directory into your own repo and it works.
See CONTRIBUTING.md. Issues and PRs welcome — especially bug reports with a failing case, since a case is the unit of work in this repo.