@agtlantis/eval 0.1.0
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.
- package/CHANGELOG.md +198 -0
- package/LICENSE +21 -0
- package/README.md +496 -0
- package/dist/cli.js +4709 -0
- package/dist/cli.js.map +1 -0
- package/dist/index.cjs +3998 -0
- package/dist/index.cjs.map +1 -0
- package/dist/index.d.cts +2738 -0
- package/dist/index.d.ts +2738 -0
- package/dist/index.js +3868 -0
- package/dist/index.js.map +1 -0
- package/package.json +101 -0
package/package.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
1
|
+
{
|
|
2
|
+
"name": "@agtlantis/eval",
|
|
3
|
+
"version": "0.1.0",
|
|
4
|
+
"description": "LLM-as-Judge based AI Agent testing library with multi-turn conversations, AI simulated users, and statistical analysis",
|
|
5
|
+
"type": "module",
|
|
6
|
+
"main": "./dist/index.cjs",
|
|
7
|
+
"module": "./dist/index.js",
|
|
8
|
+
"types": "./dist/index.d.ts",
|
|
9
|
+
"bin": {
|
|
10
|
+
"agent-eval": "./dist/cli.js"
|
|
11
|
+
},
|
|
12
|
+
"exports": {
|
|
13
|
+
".": {
|
|
14
|
+
"types": "./dist/index.d.ts",
|
|
15
|
+
"import": "./dist/index.js",
|
|
16
|
+
"require": "./dist/index.cjs"
|
|
17
|
+
}
|
|
18
|
+
},
|
|
19
|
+
"files": [
|
|
20
|
+
"dist",
|
|
21
|
+
"README.md",
|
|
22
|
+
"LICENSE",
|
|
23
|
+
"CHANGELOG.md"
|
|
24
|
+
],
|
|
25
|
+
"scripts": {
|
|
26
|
+
"build": "tsup",
|
|
27
|
+
"test": "vitest run src/",
|
|
28
|
+
"test:e2e": "vitest run e2e/",
|
|
29
|
+
"typecheck": "tsc --noEmit",
|
|
30
|
+
"prepublishOnly": "pnpm run build && pnpm run test run",
|
|
31
|
+
"example:qa": "tsx --env-file=.env examples/eval-qa-agent.ts",
|
|
32
|
+
"example:multi-turn": "tsx --env-file=.env examples/eval-multi-turn.ts",
|
|
33
|
+
"example:ai-user": "tsx --env-file=.env examples/eval-ai-user.ts",
|
|
34
|
+
"example:full-pipeline": "tsx --env-file=.env examples/eval-full-pipeline.ts"
|
|
35
|
+
},
|
|
36
|
+
"keywords": [
|
|
37
|
+
"ai",
|
|
38
|
+
"agent",
|
|
39
|
+
"evaluation",
|
|
40
|
+
"testing",
|
|
41
|
+
"llm",
|
|
42
|
+
"llm-as-judge",
|
|
43
|
+
"openai",
|
|
44
|
+
"gpt",
|
|
45
|
+
"gemini",
|
|
46
|
+
"multi-turn",
|
|
47
|
+
"conversation",
|
|
48
|
+
"prompt-engineering",
|
|
49
|
+
"ai-testing",
|
|
50
|
+
"agent-testing",
|
|
51
|
+
"typescript",
|
|
52
|
+
"vercel-ai-sdk"
|
|
53
|
+
],
|
|
54
|
+
"author": "hakzzong",
|
|
55
|
+
"license": "MIT",
|
|
56
|
+
"homepage": "https://github.com/hakzzong/agent-eval#readme",
|
|
57
|
+
"repository": {
|
|
58
|
+
"type": "git",
|
|
59
|
+
"url": "git+https://github.com/hakzzong/agent-eval.git"
|
|
60
|
+
},
|
|
61
|
+
"bugs": {
|
|
62
|
+
"url": "https://github.com/hakzzong/agent-eval/issues"
|
|
63
|
+
},
|
|
64
|
+
"engines": {
|
|
65
|
+
"node": ">=18.0.0"
|
|
66
|
+
},
|
|
67
|
+
"publishConfig": {
|
|
68
|
+
"access": "public"
|
|
69
|
+
},
|
|
70
|
+
"devDependencies": {
|
|
71
|
+
"@types/better-sqlite3": "^7.6.13",
|
|
72
|
+
"@types/node": "^22.0.0",
|
|
73
|
+
"dotenv": "^17.2.3",
|
|
74
|
+
"execa": "^9.6.1",
|
|
75
|
+
"tsup": "^8.5.1",
|
|
76
|
+
"tsx": "^4.7.0",
|
|
77
|
+
"typescript": "^5.9.3",
|
|
78
|
+
"vitest": "^4.0.16",
|
|
79
|
+
"zod": "^3.23.8"
|
|
80
|
+
},
|
|
81
|
+
"peerDependencies": {
|
|
82
|
+
"zod": "^3.22.0"
|
|
83
|
+
},
|
|
84
|
+
"peerDependenciesMeta": {
|
|
85
|
+
"zod": {
|
|
86
|
+
"optional": false
|
|
87
|
+
}
|
|
88
|
+
},
|
|
89
|
+
"dependencies": {
|
|
90
|
+
"@agtlantis/core": "workspace:*",
|
|
91
|
+
"@ai-sdk/google": "^3.0.2",
|
|
92
|
+
"@ai-sdk/openai": "^3.0.4",
|
|
93
|
+
"ai": "^6.0.11",
|
|
94
|
+
"better-sqlite3": "^12.5.0",
|
|
95
|
+
"bundle-require": "^5.1.0",
|
|
96
|
+
"cac": "^6.7.14",
|
|
97
|
+
"fast-glob": "^3.3.3",
|
|
98
|
+
"handlebars": "^4.7.8",
|
|
99
|
+
"yaml": "^2.8.2"
|
|
100
|
+
}
|
|
101
|
+
}
|