@dogfood-lab/ai-crucible 0.0.0 → 0.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (11) hide show
  1. package/LICENSE +21 -0
  2. package/README.es.md +34 -0
  3. package/README.fr.md +38 -0
  4. package/README.hi.md +36 -0
  5. package/README.it.md +38 -0
  6. package/README.ja.md +36 -0
  7. package/README.md +40 -4
  8. package/README.pt-BR.md +39 -0
  9. package/README.zh.md +35 -0
  10. package/bin/ai-crucible.js +18 -0
  11. package/package.json +32 -6
package/LICENSE ADDED
@@ -0,0 +1,21 @@
1
+ MIT License
2
+
3
+ Copyright (c) 2026 mcp-tool-shop
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
package/README.es.md ADDED
@@ -0,0 +1,34 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ Una forma sencilla de acceder a [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) mediante **npx**, sin necesidad de requisitos previos.
19
+ 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.
20
+
21
+ ```bash
22
+ npx @dogfood-lab/ai-crucible --help
23
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
24
+ ```
25
+
26
+ ## Cómo funciona
27
+
28
+ 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`.
29
+
30
+ ## Versión preliminar para investigación (v0.2.x)
31
+
32
+ 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.
33
+
34
+ **Código fuente, documentación y comprobantes:** https://github.com/dogfood-lab/ai-crucible
package/README.fr.md ADDED
@@ -0,0 +1,38 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ Accès direct via **npx** (sans prérequis) à [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
19
+ un instrument de mesure diagnostique qui réunit un **panel de juges locaux issus de différentes familles de LLM** dans un
20
+ environnement de mesure isolé et évalue les tentatives par rapport à un oracle caché.
21
+
22
+ ```bash
23
+ npx @dogfood-lab/ai-crucible --help
24
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
25
+ ```
26
+
27
+ ## Fonctionnement
28
+
29
+ Ce paquet est un **lanceur léger** (via [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)) :
30
+ lors de la première exécution, il télécharge le binaire de la plateforme à partir de la version correspondante sur
31
+ [GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), vérifie son **SHA-256**
32
+ 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.
33
+
34
+ ## Version préliminaire pour la recherche (v0.2.x)
35
+
36
+ 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.
37
+
38
+ **Code source, documentation et reçus :** https://github.com/dogfood-lab/ai-crucible
package/README.hi.md ADDED
@@ -0,0 +1,36 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ शून्य पूर्व-आवश्यकताओं वाला **npx** फ्रंट-एंड, जो [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) तक पहुंच प्रदान करता है —
19
+ यह एक नैदानिक माप उपकरण है जो **स्थानीय एलएलएम न्यायाधीशों के एक क्रॉस-परिवार पैनल** को एक
20
+ सीमित माप सीमा के भीतर बैठाता है और एक छिपे हुए ओरेकल के विरुद्ध प्रयासों का मूल्यांकन करता है।
21
+
22
+ ```bash
23
+ npx @dogfood-lab/ai-crucible --help
24
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
25
+ ```
26
+
27
+ ## यह कैसे काम करता है
28
+
29
+ यह पैकेज एक **सरल लॉन्चर** है ([`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher) के माध्यम से):
30
+ 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।
31
+
32
+ ## अनुसंधान पूर्वावलोकन (v0.2.x)
33
+
34
+ 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।
35
+
36
+ **स्रोत, दस्तावेज़ और रसीदें:** https://github.com/dogfood-lab/ai-crucible
package/README.it.md ADDED
@@ -0,0 +1,38 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ Un punto di accesso **npx** che non richiede prerequisiti per [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
19
+ uno strumento di misurazione diagnostica che riunisce un **gruppo eterogeneo di valutatori locali di LLM** all'interno di
20
+ un ambiente di misurazione isolato e valuta i tentativi rispetto a un oracolo nascosto.
21
+
22
+ ```bash
23
+ npx @dogfood-lab/ai-crucible --help
24
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
25
+ ```
26
+
27
+ ## Come funziona
28
+
29
+ Questo pacchetto è un **launcher leggero** (tramite [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)):
30
+ alla prima esecuzione, scarica il binario della piattaforma dalla corrispondente
31
+ [versione su GitHub](https://github.com/dogfood-lab/ai-crucible/releases), ne verifica l'**SHA-256**
32
+ 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`.
33
+
34
+ ## Anteprima di ricerca (v0.2.x)
35
+
36
+ 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.
37
+
38
+ **Codice sorgente, documentazione e ricevute:** https://github.com/dogfood-lab/ai-crucible
package/README.ja.md ADDED
@@ -0,0 +1,36 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ [`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 —
19
+ 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、
20
+ 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。
21
+
22
+ ```bash
23
+ npx @dogfood-lab/ai-crucible --help
24
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
25
+ ```
26
+
27
+ ## 仕組み
28
+
29
+ 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。
30
+ 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。
31
+
32
+ ## リサーチプレビュー（v0.2.x）
33
+
34
+ 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。
35
+
36
+ **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 CHANGED
@@ -1,7 +1,43 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
1
16
  # @dogfood-lab/ai-crucible
2
17
 
3
- Name placeholder (`v0.0.0`). The real package is a thin `npx` launcher that downloads and
4
- runs the **ai-crucible** diagnostic measurement instrument (a Python tool); it publishes
5
- from CI via npm Trusted Publishing (OIDC, no tokens).
18
+ Zero-prerequisite **npx** front door to [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
19
+ a diagnostic measurement instrument that seats a **cross-family panel of local LLM judges** under a
20
+ sealed measurement boundary and scores attempts against a hidden oracle.
21
+
22
+ ```bash
23
+ npx @dogfood-lab/ai-crucible --help
24
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
25
+ ```
26
+
27
+ ## How it works
28
+
29
+ This package is a **thin launcher** (via [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)):
30
+ on first run it downloads the platform binary from the matching
31
+ [GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), verifies its **SHA-256**
32
+ against the release's `checksums-<version>.txt`, caches it, and runs it with full argument
33
+ passthrough. The tool itself is Python — but you do **not** need Python installed to use it this
34
+ way. Prefer `pip install ai-crucible` if you want the importable library surface.
35
+
36
+ ## Research preview (v0.2.x)
37
+
38
+ ai-crucible is the measurement arm of a larger pipeline, shipped honestly pre-1.0. Its judge
39
+ panel's alt-test ω is still a **circular model-jury bootstrap** until a human-labeling round runs,
40
+ so seated judges are **provisional** and the live panel **escalates to a Claude Designer** below
41
+ quorum. The repository carries the full, non-cosmetic scorecard and the verifiable receipts.
6
42
 
7
- Source + releases: https://github.com/dogfood-lab/ai-crucible
43
+ **Source, docs, and receipts:** https://github.com/dogfood-lab/ai-crucible
package/README.pt-BR.md ADDED
@@ -0,0 +1,39 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ Uma ferramenta de acesso via **npx** que não exige configurações prévias para [`ai-crucible`](https://github.com/dogfood-lab/ai-crucible) —
19
+ um instrumento de medição e diagnóstico que reúne um **painel diversificado de avaliadores locais de LLM** em um
20
+ ambiente de medição isolado e avalia os resultados em comparação com um critério oculto.
21
+
22
+ ```bash
23
+ npx @dogfood-lab/ai-crucible --help
24
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
25
+ ```
26
+
27
+ ## Como funciona
28
+
29
+ Este pacote é um **lançador simplificado** (via [`@mcptoolshop/npm-launcher`](https://www.npmjs.com/package/@mcptoolshop/npm-launcher)):
30
+ na primeira execução, ele baixa o binário da plataforma da versão correspondente no
31
+ [GitHub Release](https://github.com/dogfood-lab/ai-crucible/releases), verifica o seu **SHA-256**
32
+ 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`.
33
+
34
+ ## Prévia de pesquisa (v0.2.x)
35
+
36
+ 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,
37
+ 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.
38
+
39
+ **Código-fonte, documentação e comprovantes:** https://github.com/dogfood-lab/ai-crucible
package/README.zh.md ADDED
@@ -0,0 +1,35 @@
1
+ <p align="center">
2
+ <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>
3
+ </p>
4
+
5
+ <p align="center">
6
+ <img src="https://raw.githubusercontent.com/dogfood-lab/ai-crucible/main/assets/logo.png" alt="ai-crucible" width="420">
7
+ </p>
8
+
9
+ <p align="center">
10
+ <a href="https://pypi.org/project/ai-crucible/"><img src="https://img.shields.io/pypi/v/ai-crucible" alt="PyPI"></a>
11
+ <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>
12
+ <a href="https://github.com/dogfood-lab/ai-crucible"><img src="https://img.shields.io/badge/source-GitHub-blue" alt="source"></a>
13
+ <a href="https://dogfood-lab.github.io/ai-crucible/"><img src="https://img.shields.io/badge/docs-handbook-orange" alt="docs"></a>
14
+ </p>
15
+
16
+ # @dogfood-lab/ai-crucible
17
+
18
+ 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)——
19
+ 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。
20
+
21
+ ```bash
22
+ npx @dogfood-lab/ai-crucible --help
23
+ npx @dogfood-lab/ai-crucible characterize --k 3 # needs a local Ollama panel
24
+ ```
25
+
26
+ ## 工作原理
27
+
28
+ 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)）：
29
+ 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`。
30
+
31
+ ## 研究预览版 (v0.2.x)
32
+
33
+ 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。
34
+
35
+ **源代码、文档和记录：**https://github.com/dogfood-lab/ai-crucible
package/bin/ai-crucible.js ADDED
@@ -0,0 +1,18 @@
1
+ #!/usr/bin/env node
2
+ "use strict";
3
+
4
+ // Thin npm wrapper for the ai-crucible CLI. Pure JSON config — @mcptoolshop/npm-launcher derives
5
+ // the release-asset names from convention, downloads the platform binary from the ai-crucible
6
+ // GitHub Release, verifies its SHA256 against checksums-<version>.txt, caches it, and runs it
7
+ // with full arg passthrough.
8
+ // binary: ai-crucible-0.2.0-linux-x64
9
+ // checksums: checksums-0.2.0.txt
10
+ process.env.MCPTOOLSHOP_LAUNCH_CONFIG = JSON.stringify({
11
+ toolName: "ai-crucible",
12
+ owner: "dogfood-lab",
13
+ repo: "ai-crucible",
14
+ version: "0.2.0",
15
+ tag: "v0.2.0",
16
+ });
17
+
18
+ require("@mcptoolshop/npm-launcher/bin/mcptoolshop-launch.js");
package/package.json CHANGED
@@ -1,14 +1,40 @@
1
1
  {
2
2
  "name": "@dogfood-lab/ai-crucible",
3
- "version": "0.0.0",
4
- "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).",
3
+ "version": "0.2.0",
4
+ "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.",
5
+ "type": "commonjs",
5
6
  "license": "MIT",
7
+ "engines": {
8
+ "node": ">=18"
9
+ },
10
+ "bin": {
11
+ "ai-crucible": "bin/ai-crucible.js"
12
+ },
13
+ "files": [
14
+ "bin/",
15
+ "README.md",
16
+ "README.*.md",
17
+ "LICENSE"
18
+ ],
19
+ "scripts": {
20
+ "verify": "node -e \"require('./bin/ai-crucible.js')\" 2>&1 || true"
21
+ },
22
+ "keywords": [
23
+ "llm",
24
+ "evaluation",
25
+ "auditing-game",
26
+ "judge-panel",
27
+ "reward-hacking",
28
+ "inspect-ai",
29
+ "cli"
30
+ ],
31
+ "dependencies": {
32
+ "@mcptoolshop/npm-launcher": "^1.0.0"
33
+ },
34
+ "author": "mcp-tool-shop",
6
35
  "repository": {
7
36
  "type": "git",
8
37
  "url": "git+https://github.com/dogfood-lab/ai-crucible.git"
9
38
  },
10
- "homepage": "https://github.com/dogfood-lab/ai-crucible",
11
- "files": [
12
- "README.md"
13
- ]
39
+ "homepage": "https://github.com/dogfood-lab/ai-crucible"
14
40
  }