@dogfood-lab/ai-crucible 0.0.0 → 0.3.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 mcp-tool-shop
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.es.md

@@ -0,0 +1,34 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.md">English</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+Una forma sencilla de acceder a [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) mediante **npx**, sin necesidad de requisitos previos.
+Es un instrumento de medición diagnóstica que reúne a un **panel diverso de evaluadores locales de LLM** dentro de un entorno de medición aislado y evalúa los resultados en comparación con un oráculo oculto.
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## Cómo funciona
+
+Este paquete es un **lanzador ligero** (a través de [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)): en la primera ejecución, descarga el binario de la plataforma desde la versión correspondiente en [GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), verifica su **SHA-256** con el archivo `checksums-<version>.txt` de la versión, lo almacena en caché y lo ejecuta con todos los argumentos. La herramienta en sí está escrita en Python, pero **no** es necesario tener Python instalado para usarla de esta manera. Si desea utilizar la biblioteca como módulo, prefiera `pip install ai-crucible`.
+
+## Versión preliminar para investigación (v0.2.x)
+
+ai-crucible es el componente de medición de un flujo de trabajo más amplio, que se ofrece de forma transparente antes de la versión 1.0. La prueba alternativa ω del panel de evaluadores aún es un **modelo de jurado circular** hasta que se realice una ronda de etiquetado humano, por lo que los evaluadores son **provisionales** y el panel activo **se amplía a un Claude Designer** cuando no se alcanza el quórum. El repositorio contiene la puntuación completa y los comprobantes verificables.
+
+**Código fuente, documentación y comprobantes:** https://github.com/dogfood-lab/ai-crucible

package/README.fr.md

@@ -0,0 +1,38 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.md">English</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+Accès direct via **npx** (sans prérequis) à [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
+un instrument de mesure diagnostique qui réunit un **panel de juges locaux issus de différentes familles de LLM** dans un
+environnement de mesure isolé et évalue les tentatives par rapport à un oracle caché.
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## Fonctionnement
+
+Ce paquet est un **lanceur léger** (via [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)) :
+lors de la première exécution, il télécharge le binaire de la plateforme à partir de la version correspondante sur
+[GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), vérifie son **SHA-256**
+par rapport au fichier `checksums-<version>.txt` de la version, le met en cache et l’exécute avec tous les arguments. L’outil lui-même est en Python, mais vous n’avez **pas** besoin d’avoir Python installé pour l’utiliser de cette manière. Préférez `pip install ai-crucible` si vous souhaitez utiliser la bibliothèque importable.
+
+## Version préliminaire pour la recherche (v0.2.x)
+
+ai-crucible est la partie de mesure d’un pipeline plus vaste, distribué en version préliminaire avant la 1.0. Le test alternatif ω du panel de juges est encore un **modèle de bootstrap circulaire** jusqu’à ce qu’une phase d’étiquetage humain soit effectuée, de sorte que les juges présents sont **provisoires** et que le panel actif **passe à un Claude Designer** lorsque le quorum n’est pas atteint. Le dépôt contient le tableau de bord complet, sans éléments cosmétiques, et les reçus vérifiables.
+
+**Code source, documentation et reçus :** https://github.com/dogfood-lab/ai-crucible

package/README.hi.md

@@ -0,0 +1,36 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.md">English</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+शून्य पूर्व-आवश्यकताओं वाला **npx** फ्रंट-एंड, जो [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) तक पहुंच प्रदान करता है —
+यह एक नैदानिक माप उपकरण है जो **स्थानीय एलएलएम न्यायाधीशों के एक क्रॉस-परिवार पैनल** को एक
+सीमित माप सीमा के भीतर बैठाता है और एक छिपे हुए ओरेकल के विरुद्ध प्रयासों का मूल्यांकन करता है।
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## यह कैसे काम करता है
+
+यह पैकेज एक **सरल लॉन्चर** है ([`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher) के माध्यम से):
+पहली बार चलाने पर, यह संबंधित [गिटहब रिलीज़](https://github.com/dogfood-lab/ai-crucible/releases) से प्लेटफ़ॉर्म बाइनरी डाउनलोड करता है, इसकी **SHA-256** को रिलीज़ के `checksums-<version>.txt` से सत्यापित करता है, इसे कैश करता है, और सभी तर्कों के साथ चलाता है। उपकरण स्वयं पायथन में है — लेकिन इसे इस तरह उपयोग करने के लिए आपको पायथन स्थापित करने की आवश्यकता **नहीं** है। यदि आप आयात करने योग्य लाइब्रेरी चाहते हैं, तो `pip install ai-crucible` का उपयोग करें।
+
+## अनुसंधान पूर्वावलोकन (v0.2.x)
+
+ai-crucible एक बड़े पाइपलाइन का माप भाग है, जिसे 1.0 से पहले ईमानदारी से जारी किया गया है। इसके न्यायाधीश पैनल का वैकल्पिक परीक्षण ω अभी भी एक **चक्रीय मॉडल-जूरी बूटस्ट्रैप** है, जब तक कि मानव-लेबलिंग दौर नहीं चलता, इसलिए बैठे हुए न्यायाधीश **अस्थायी** हैं और लाइव पैनल **न्यूनतम संख्या से नीचे एक क्लाउड डिज़ाइनर** तक विस्तारित होता है। रिपॉजिटरी में पूर्ण, गैर-सौंदर्य स्कोरकार्ड और सत्यापित रसीदें शामिल हैं।
+
+**स्रोत, दस्तावेज़ और रसीदें:** https://github.com/dogfood-lab/ai-crucible

package/README.it.md

@@ -0,0 +1,38 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.md">English</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+Un punto di accesso **npx** che non richiede prerequisiti per [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
+uno strumento di misurazione diagnostica che riunisce un **gruppo eterogeneo di valutatori locali di LLM** all'interno di
+un ambiente di misurazione isolato e valuta i tentativi rispetto a un oracolo nascosto.
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## Come funziona
+
+Questo pacchetto è un **launcher leggero** (tramite [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)):
+alla prima esecuzione, scarica il binario della piattaforma dalla corrispondente
+[versione su GitHub](https://github.com/dogfood-lab/ai-crucible/releases), ne verifica l'**SHA-256**
+confrontandolo con il file `checksums-<versione>.txt` della versione, lo memorizza nella cache e lo esegue passando tutti gli argomenti. Lo strumento stesso è scritto in Python, ma non è necessario avere Python installato per utilizzarlo in questo modo. Se si desidera utilizzare la libreria importabile, è preferibile utilizzare `pip install ai-crucible`.
+
+## Anteprima di ricerca (v0.2.x)
+
+ai-crucible è la parte di misurazione di una pipeline più ampia, fornita in versione pre-1.0. Il test alternativo ω del gruppo di valutatori è ancora un **modello di bootstrap circolare** fino a quando non viene eseguito un ciclo di etichettatura umana, quindi i valutatori sono **provvisori** e il gruppo attivo **si amplia fino a includere un Claude Designer** quando non si raggiunge il quorum. Il repository contiene il punteggio completo e non modificato e le ricevute verificabili.
+
+**Codice sorgente, documentazione e ricevute:** https://github.com/dogfood-lab/ai-crucible

package/README.ja.md

@@ -0,0 +1,36 @@
+<p align="center">
+  <a href="README.md">English</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+[`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) への、前提条件なしの **npx** を使用した簡単なアクセス方法 —
+これは、**異なるモデル群から選ばれたローカルLLMの審査員**を、
+密閉された測定境界内で配置し、隠された基準に対して試行を評価する診断測定ツールです。
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## 仕組み
+
+このパッケージは、**軽量ランチャー**（[`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher) を介して）です。
+初回実行時に、対応する [GitHub リリース](https://github.com/dogfood-lab/ai-crucible/releases) からプラットフォームバイナリをダウンロードし、その **SHA-256** をリリースの `checksums-<version>.txt` と照合して検証し、キャッシュし、すべての引数をそのまま渡して実行します。このツール自体は Python で記述されていますが、この方法で使用するには Python をインストールする必要はありません。インポート可能なライブラリが必要な場合は、`pip install ai-crucible` を使用してください。
+
+## リサーチプレビュー（v0.2.x）
+
+ai-crucible は、より大規模なパイプラインの一部であり、1.0 より前に正直に公開されています。その審査員パネルの代替テスト ω は、人間のラベル付けラウンドが実行されるまでは、**循環モデルによる審査員グループのブートストラップ**であり、そのため、配置された審査員は**一時的**であり、必要な人数に達するまで、審査員パネルは**Claude Designer** に切り替わります。リポジトリには、完全な、装飾のないスコアカードと検証可能な記録が含まれています。
+
+**ソースコード、ドキュメント、および記録:** https://github.com/dogfood-lab/ai-crucible

package/README.md

@@ -1,7 +1,43 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
 # @dogfood-lab/ai-crucible
 
-Name placeholder (`v0.0.0`). The real package is a thin `npx` launcher that downloads and
-runs the **ai-crucible** diagnostic measurement instrument (a Python tool); it publishes
-from CI via npm Trusted Publishing (OIDC, no tokens).
+Zero-prerequisite **npx** front door to [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
+a diagnostic measurement instrument that seats a **cross-family panel of local LLM judges** under a
+sealed measurement boundary and scores attempts against a hidden oracle.
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## How it works
+
+This package is a **thin launcher** (via [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)):
+on first run it downloads the platform binary from the matching
+[GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), verifies its **SHA-256**
+against the release's `checksums-<version>.txt`, caches it, and runs it with full argument
+passthrough. The tool itself is Python — but you do **not** need Python installed to use it this
+way. Prefer `pip install ai-crucible` if you want the importable library surface.
+
+## Research preview (v0.2.x)
+
+ai-crucible is the measurement arm of a larger pipeline, shipped honestly pre-1.0. Its judge
+panel's alt-test ω is still a **circular model-jury bootstrap** until a human-labeling round runs,
+so seated judges are **provisional** and the live panel **escalates to a Claude Designer** below
+quorum. The repository carries the full, non-cosmetic scorecard and the verifiable receipts.
 
-Source + releases: https://github.com/dogfood-lab/ai-crucible
+**Source, docs, and receipts:** https://github.com/dogfood-lab/ai-crucible

package/README.pt-BR.md

@@ -0,0 +1,39 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.zh.md">中文</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.md">English</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+Uma ferramenta de acesso via **npx** que não exige configurações prévias para [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
+um instrumento de medição e diagnóstico que reúne um **painel diversificado de avaliadores locais de LLM** em um
+ambiente de medição isolado e avalia os resultados em comparação com um critério oculto.
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## Como funciona
+
+Este pacote é um **lançador simplificado** (via [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)):
+na primeira execução, ele baixa o binário da plataforma da versão correspondente no
+[GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), verifica o seu **SHA-256**
+em relação ao arquivo `checksums-<version>.txt` da versão, armazena em cache e o executa com todos os argumentos. A ferramenta em si é em Python, mas você **não** precisa ter o Python instalado para usá-la desta forma. Se você quiser a biblioteca que pode ser importada, use `pip install ai-crucible`.
+
+## Prévia de pesquisa (v0.2.x)
+
+ai-crucible é o componente de medição de um pipeline maior, lançado de forma transparente antes da versão 1.0. O teste alternativo ω do painel de avaliadores ainda é um **modelo circular de avaliação**, até que seja realizada uma rodada de rotulagem humana,
+de modo que os avaliadores presentes são **provisórios** e o painel ativo **evolui para um Claude Designer** abaixo do quórum. O repositório contém a pontuação completa e não cosmética e os comprovantes verificáveis.
+
+**Código-fonte, documentação e comprovantes:** https://github.com/dogfood-lab/ai-crucible

package/README.zh.md

@@ -0,0 +1,35 @@
+<p align="center">
+  <a href="README.ja.md">日本語</a> | <a href="README.md">English</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.hi.md">हिन्दी</a> | <a href="README.it.md">Italiano</a> | <a href="README.pt-BR.md">Português (BR)</a>
+</p>
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
+  <a href="https://www.npmjs.com/package/@dogfood-lab/ai-crucible"><img src="https://img.shields.io/npm/v/@dogfood-lab/ai-crucible" alt="npm"></a>
+  <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
+  <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
+</p>
+
+# @dogfood-lab/ai-crucible
+
+一个无需任何先决条件的 **npx** 前端，用于访问 [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible)——
+这是一种诊断测量工具，它将一个**由不同模型的本地 LLM 评委组成的评审团**置于一个封闭的测量边界内，并根据隐藏的参考标准对尝试进行评分。
+
+```bash
+npx @dogfood-lab/ai-crucible --help
+npx @dogfood-lab/ai-crucible characterize --k 3   # needs a local Ollama panel
+```
+
+## 工作原理
+
+此软件包是一个**轻量级启动器**（通过 [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)）：
+首次运行时，它会从匹配的 [GitHub 发布](https://github.com/dogfood-lab/ai-crucible/releases) 下载平台二进制文件，并将其 **SHA-256** 值与发布的 `checksums-<version>.txt` 文件进行验证，然后将其缓存，并以完全传递参数的方式运行。该工具本身是 Python 编写的——但您**不需要**安装 Python 即可以这种方式使用它。如果您需要可导入的库，请使用 `pip install ai-crucible`。
+
+## 研究预览版 (v0.2.x)
+
+ai-crucible 是一个更大流水线中的测量模块，在 1.0 版本之前发布。其评审团的替代测试 ω 仍然是一个**循环模型评审引导**，直到进行一轮人工标注，因此，评审团成员是**暂定的**，并且当评审团人数不足时，**评审团会升级到 Claude Designer**。该仓库包含完整的、非装饰性的评分表和可验证的记录。
+
+**源代码、文档和记录：**https://github.com/dogfood-lab/ai-crucible

package/bin/ai-crucible.js

@@ -0,0 +1,18 @@
+#!/usr/bin/env node
+"use strict";
+
+// Thin npm wrapper for the ai-crucible CLI. Pure JSON config — @mcptoolshop/npm-launcher derives
+// the release-asset names from convention, downloads the platform binary from the ai-crucible
+// GitHub Release, verifies its SHA256 against checksums-<version>.txt, caches it, and runs it
+// with full arg passthrough.
+//   binary:    ai-crucible-0.2.0-linux-x64
+//   checksums: checksums-0.2.0.txt
+process.env.MCPTOOLSHOP_LAUNCH_CONFIG = JSON.stringify({
+  toolName: "ai-crucible",
+  owner: "dogfood-lab",
+  repo: "ai-crucible",
+  version: "0.2.0",
+  tag: "v0.2.0",
+});
+
+require("@mcptoolshop/npm-launcher/bin/mcptoolshop-launch.js");

package/package.json

@@ -1,14 +1,40 @@
 {
   "name": "@dogfood-lab/ai-crucible",
-  "version": "0.0.0",
-  "description": "Name placeholder. The real launcher — an npx front door to the ai-crucible Python measurement instrument — publishes from CI via npm Trusted Publishing (OIDC).",
+  "version": "0.3.0",
+  "description": "ai-crucible — a diagnostic measurement instrument: a cross-family panel of local LLM judges under a sealed boundary, scoring against a hidden oracle. Zero-prerequisite npx install via a verified binary launcher. Research preview.",
+  "type": "commonjs",
   "license": "MIT",
+  "engines": {
+    "node": ">=18"
+  },
+  "bin": {
+    "ai-crucible": "bin/ai-crucible.js"
+  },
+  "files": [
+    "bin/",
+    "README.md",
+    "README.*.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "verify": "node -e \"require('./bin/ai-crucible.js')\" 2>&1 || true"
+  },
+  "keywords": [
+    "llm",
+    "evaluation",
+    "auditing-game",
+    "judge-panel",
+    "reward-hacking",
+    "inspect-ai",
+    "cli"
+  ],
+  "dependencies": {
+    "@mcptoolshop/npm-launcher": "^1.0.0"
+  },
+  "author": "mcp-tool-shop",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/dogfood-lab/ai-crucible.git"
   },
-  "homepage": "https://github.com/dogfood-lab/ai-crucible",
-  "files": [
-    "README.md"
-  ]
+  "homepage": "https://github.com/dogfood-lab/ai-crucible"
 }