agentv 5.0.0-next.1 → 5.2.0-next.1

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 (108) hide show
  1. package/README.md +239 -53
  2. package/dist/{agentv-provider-AYXH7WLW-NJRC6UQX.js → agentv-provider-AYXH7WLW-6YZKZ7PS.js} +2 -2
  3. package/dist/{artifact-writer-QD52FC56.js → artifact-writer-SN6UNL5G.js} +15 -8
  4. package/dist/{chunk-GZJMODLW.js → chunk-3S6A2RKR.js} +45880 -30542
  5. package/dist/chunk-3S6A2RKR.js.map +1 -0
  6. package/dist/{chunk-6Q56CEXV.js → chunk-3X357HS4.js} +7900 -8985
  7. package/dist/chunk-3X357HS4.js.map +1 -0
  8. package/dist/{chunk-6JVM4GJI.js → chunk-46K2OET3.js} +187 -80
  9. package/dist/chunk-46K2OET3.js.map +1 -0
  10. package/dist/{chunk-IJEZFZJR.js → chunk-7BGERE6L.js} +2 -2
  11. package/dist/chunk-7BGERE6L.js.map +1 -0
  12. package/dist/{chunk-UG4JHZYM.js → chunk-ALVZQUZP.js} +670 -1427
  13. package/dist/chunk-ALVZQUZP.js.map +1 -0
  14. package/dist/{chunk-RCIEHFQ2.js → chunk-PEUTJS7B.js} +3 -3
  15. package/dist/{chunk-5H446C7X.js → chunk-QMRVH5ZP.js} +2 -8
  16. package/dist/{chunk-IBI7A5RP.js → chunk-T4AD3H3Y.js} +3444 -2251
  17. package/dist/chunk-T4AD3H3Y.js.map +1 -0
  18. package/dist/cli.js +8 -9
  19. package/dist/cli.js.map +1 -1
  20. package/dist/dashboard/assets/{index-D5_8DE_B.js → index-DNgf3qJ2.js} +1 -1
  21. package/dist/dashboard/assets/index-D_bokML8.css +1 -0
  22. package/dist/dashboard/assets/index-r_jSJmlw.js +121 -0
  23. package/dist/dashboard/index.html +2 -2
  24. package/dist/{dist-KJ6PELQY.js → dist-DEPJOMCA.js} +74 -60
  25. package/dist/{docker-workspace-ZJ2ESVBI-6UDETQII.js → docker-workspace-NNK4Z6FR-QRTIG6TM.js} +3 -3
  26. package/dist/{exec-MYPAJJP7-C5HD5WG5.js → exec-4HFFRQ3B-AXDMJNLJ.js} +3 -3
  27. package/dist/index.js +8 -9
  28. package/dist/{interactive-2WMVAASS.js → interactive-36ZNMQB2.js} +13 -29
  29. package/dist/interactive-36ZNMQB2.js.map +1 -0
  30. package/dist/skills/agentv-bench/SKILL.md +38 -34
  31. package/dist/skills/agentv-bench/agents/analyzer.md +6 -6
  32. package/dist/skills/agentv-bench/agents/comparator.md +4 -5
  33. package/dist/skills/agentv-bench/agents/grader.md +4 -4
  34. package/dist/skills/agentv-bench/references/autoresearch.md +3 -3
  35. package/dist/skills/agentv-bench/references/description-optimization.md +2 -2
  36. package/dist/skills/agentv-bench/references/environment-adaptation.md +10 -10
  37. package/dist/skills/agentv-bench/references/eval-yaml-spec.md +51 -30
  38. package/dist/skills/agentv-bench/references/migrating-from-skill-creator.md +4 -4
  39. package/dist/skills/agentv-bench/references/subagent-pipeline.md +13 -12
  40. package/dist/skills/agentv-eval-migrations/SKILL.md +33 -0
  41. package/dist/skills/agentv-eval-migrations/references/breaking-changes.md +1012 -0
  42. package/dist/skills/agentv-eval-review/SKILL.md +6 -3
  43. package/dist/skills/agentv-eval-review/scripts/lint_eval.py +1 -1
  44. package/dist/skills/agentv-eval-writer/SKILL.md +380 -246
  45. package/dist/skills/agentv-eval-writer/references/{config-schema.json → config.schema.json} +6 -5
  46. package/dist/skills/agentv-eval-writer/references/custom-evaluators.md +23 -18
  47. package/dist/skills/agentv-eval-writer/references/eval.schema.json +9852 -0
  48. package/dist/skills/agentv-eval-writer/references/python-helpers.md +8 -8
  49. package/dist/skills/agentv-eval-writer/references/rubric-evaluator.md +21 -22
  50. package/dist/skills/agentv-trace-analyst/SKILL.md +4 -4
  51. package/dist/templates/.agentv/targets.yaml +41 -40
  52. package/dist/{ts-eval-loader-RE43OAGX-AMRCUKBZ.js → ts-eval-loader-OMG2JBHP-OP5RR7A3.js} +5 -4
  53. package/package.json +19 -4
  54. package/dist/chunk-6JVM4GJI.js.map +0 -1
  55. package/dist/chunk-6Q56CEXV.js.map +0 -1
  56. package/dist/chunk-76FOHROU.js +0 -96
  57. package/dist/chunk-76FOHROU.js.map +0 -1
  58. package/dist/chunk-GZJMODLW.js.map +0 -1
  59. package/dist/chunk-HSHTTN6C.js +0 -565
  60. package/dist/chunk-HSHTTN6C.js.map +0 -1
  61. package/dist/chunk-IBI7A5RP.js.map +0 -1
  62. package/dist/chunk-IJEZFZJR.js.map +0 -1
  63. package/dist/chunk-LRULMAAA.js +0 -1711
  64. package/dist/chunk-LRULMAAA.js.map +0 -1
  65. package/dist/chunk-R3CXYNTN.js +0 -1238
  66. package/dist/chunk-R3CXYNTN.js.map +0 -1
  67. package/dist/chunk-UG4JHZYM.js.map +0 -1
  68. package/dist/chunk-VQ2ZO7XJ.js +0 -2098
  69. package/dist/chunk-VQ2ZO7XJ.js.map +0 -1
  70. package/dist/chunk-XALGXSKB.js +0 -21
  71. package/dist/chunk-XALGXSKB.js.map +0 -1
  72. package/dist/dashboard/assets/index-BVMrdN8a.js +0 -119
  73. package/dist/dashboard/assets/index-DZWXfWsv.css +0 -1
  74. package/dist/esm-NNMQYIKT.js +0 -32
  75. package/dist/esm-R77SNOF5.js +0 -65
  76. package/dist/esm-RVQPUGWH.js +0 -1207
  77. package/dist/esm-RVQPUGWH.js.map +0 -1
  78. package/dist/esm-TTEMJEVV.js +0 -933
  79. package/dist/esm-TTEMJEVV.js.map +0 -1
  80. package/dist/esm-ZADQ4XQH-QUFGGZL4.js +0 -924
  81. package/dist/esm-ZADQ4XQH-QUFGGZL4.js.map +0 -1
  82. package/dist/exec-MYPAJJP7-C5HD5WG5.js.map +0 -1
  83. package/dist/getMachineId-bsd-HSK5LZMG.js +0 -41
  84. package/dist/getMachineId-bsd-HSK5LZMG.js.map +0 -1
  85. package/dist/getMachineId-darwin-4DP6CCJV.js +0 -41
  86. package/dist/getMachineId-darwin-4DP6CCJV.js.map +0 -1
  87. package/dist/getMachineId-linux-44LJ5UJB.js +0 -33
  88. package/dist/getMachineId-linux-44LJ5UJB.js.map +0 -1
  89. package/dist/getMachineId-unsupported-NVK6IATM.js +0 -24
  90. package/dist/getMachineId-unsupported-NVK6IATM.js.map +0 -1
  91. package/dist/getMachineId-win-YZ36S7VA.js +0 -43
  92. package/dist/getMachineId-win-YZ36S7VA.js.map +0 -1
  93. package/dist/interactive-2WMVAASS.js.map +0 -1
  94. package/dist/otlp-json-file-exporter-RY63S3IG-PZBQPVYY.js +0 -9
  95. package/dist/otlp-json-file-exporter-RY63S3IG-PZBQPVYY.js.map +0 -1
  96. package/dist/skills/agentv-eval-writer/references/eval-schema.json +0 -17295
  97. package/dist/skills/agentv-eval-writer/references/experiment-schema.json +0 -278
  98. package/dist/src-76ISS3LH.js +0 -1733
  99. package/dist/src-76ISS3LH.js.map +0 -1
  100. package/dist/ts-eval-loader-RE43OAGX-AMRCUKBZ.js.map +0 -1
  101. /package/dist/{agentv-provider-AYXH7WLW-NJRC6UQX.js.map → agentv-provider-AYXH7WLW-6YZKZ7PS.js.map} +0 -0
  102. /package/dist/{artifact-writer-QD52FC56.js.map → artifact-writer-SN6UNL5G.js.map} +0 -0
  103. /package/dist/{chunk-RCIEHFQ2.js.map → chunk-PEUTJS7B.js.map} +0 -0
  104. /package/dist/{chunk-5H446C7X.js.map → chunk-QMRVH5ZP.js.map} +0 -0
  105. /package/dist/{dist-KJ6PELQY.js.map → dist-DEPJOMCA.js.map} +0 -0
  106. /package/dist/{docker-workspace-ZJ2ESVBI-6UDETQII.js.map → docker-workspace-NNK4Z6FR-QRTIG6TM.js.map} +0 -0
  107. /package/dist/{esm-NNMQYIKT.js.map → exec-4HFFRQ3B-AXDMJNLJ.js.map} +0 -0
  108. /package/dist/{esm-R77SNOF5.js.map → ts-eval-loader-OMG2JBHP-OP5RR7A3.js.map} +0 -0
package/dist/skills/agentv-eval-writer/references/experiment-schema.json DELETED
@@ -1,278 +0,0 @@
1
- {
2
- "$schema": "https://json-schema.org/draft/2019-09/schema#",
3
- "title": "AgentV Experiment File",
4
- "description": "Schema for AgentV experiment YAML files (experiments/*.yaml)",
5
- "$ref": "#/definitions/ExperimentFile",
6
- "definitions": {
7
- "ExperimentFile": {
8
- "type": "object",
9
- "properties": {
10
- "name": {
11
- "type": "string",
12
- "minLength": 1
13
- },
14
- "agent": {
15
- "type": "string",
16
- "minLength": 1
17
- },
18
- "target": {
19
- "type": "string",
20
- "minLength": 1
21
- },
22
- "targets": {
23
- "type": "array",
24
- "items": {
25
- "anyOf": [
26
- {
27
- "type": "string",
28
- "minLength": 1
29
- },
30
- {
31
- "type": "object",
32
- "properties": {
33
- "name": {
34
- "type": "string",
35
- "minLength": 1
36
- },
37
- "use_target": {
38
- "type": "string",
39
- "minLength": 1
40
- },
41
- "hooks": {
42
- "type": "object",
43
- "properties": {},
44
- "additionalProperties": {}
45
- }
46
- },
47
- "required": ["name"],
48
- "additionalProperties": false
49
- }
50
- ]
51
- },
52
- "minItems": 1
53
- },
54
- "model": {
55
- "type": "string",
56
- "minLength": 1
57
- },
58
- "agent_options": {
59
- "type": "object",
60
- "properties": {},
61
- "additionalProperties": {}
62
- },
63
- "suites": {
64
- "type": "array",
65
- "items": {
66
- "type": "object",
67
- "properties": {
68
- "ref": {
69
- "type": "string",
70
- "minLength": 1
71
- },
72
- "select": {
73
- "type": "object",
74
- "properties": {
75
- "test_ids": {
76
- "type": "array",
77
- "items": {
78
- "type": "string",
79
- "minLength": 1
80
- },
81
- "minItems": 1
82
- }
83
- },
84
- "required": ["test_ids"],
85
- "additionalProperties": false
86
- }
87
- },
88
- "required": ["ref"],
89
- "additionalProperties": false
90
- },
91
- "minItems": 1
92
- },
93
- "scripts": {
94
- "type": "array",
95
- "items": {
96
- "anyOf": [
97
- {
98
- "type": "string",
99
- "minLength": 1
100
- },
101
- {
102
- "type": "object",
103
- "properties": {
104
- "command": {
105
- "anyOf": [
106
- {
107
- "type": "string",
108
- "minLength": 1
109
- },
110
- {
111
- "type": "array",
112
- "items": {
113
- "type": "string",
114
- "minLength": 1
115
- },
116
- "minItems": 1
117
- }
118
- ]
119
- },
120
- "script": {
121
- "anyOf": [
122
- {
123
- "type": "string",
124
- "minLength": 1
125
- },
126
- {
127
- "type": "array",
128
- "items": {
129
- "type": "string",
130
- "minLength": 1
131
- },
132
- "minItems": 1
133
- }
134
- ]
135
- },
136
- "timeout_seconds": {
137
- "type": "number",
138
- "exclusiveMinimum": true,
139
- "minimum": 0
140
- },
141
- "cwd": {
142
- "type": "string",
143
- "minLength": 1
144
- },
145
- "env": {
146
- "type": "object",
147
- "additionalProperties": {
148
- "type": "string"
149
- }
150
- }
151
- },
152
- "additionalProperties": false
153
- }
154
- ]
155
- }
156
- },
157
- "repeat": {
158
- "type": "object",
159
- "properties": {
160
- "count": {
161
- "type": "integer",
162
- "minimum": 1
163
- },
164
- "strategy": {
165
- "type": "string",
166
- "enum": ["pass_at_k", "mean", "confidence_interval"]
167
- },
168
- "cost_limit_usd": {
169
- "type": "number",
170
- "minimum": 0
171
- },
172
- "costLimitUsd": {
173
- "type": "number",
174
- "minimum": 0
175
- }
176
- },
177
- "required": ["count"],
178
- "additionalProperties": false
179
- },
180
- "runs": {
181
- "type": "integer",
182
- "minimum": 1
183
- },
184
- "early_exit": {
185
- "type": "boolean"
186
- },
187
- "timeout_seconds": {
188
- "type": "number",
189
- "exclusiveMinimum": true,
190
- "minimum": 0
191
- },
192
- "workers": {
193
- "type": "integer",
194
- "minimum": 1
195
- },
196
- "budget_usd": {
197
- "type": "number",
198
- "exclusiveMinimum": true,
199
- "minimum": 0
200
- },
201
- "sandbox": {
202
- "type": "string",
203
- "enum": ["auto", "docker", "vercel"]
204
- },
205
- "workspace": {
206
- "type": "object",
207
- "properties": {},
208
- "additionalProperties": {}
209
- },
210
- "setup": {
211
- "type": "array",
212
- "items": {
213
- "anyOf": [
214
- {
215
- "type": "string",
216
- "minLength": 1
217
- },
218
- {
219
- "type": "object",
220
- "properties": {
221
- "command": {
222
- "anyOf": [
223
- {
224
- "type": "string",
225
- "minLength": 1
226
- },
227
- {
228
- "type": "array",
229
- "items": {
230
- "type": "string",
231
- "minLength": 1
232
- },
233
- "minItems": 1
234
- }
235
- ]
236
- },
237
- "script": {
238
- "anyOf": [
239
- {
240
- "type": "string",
241
- "minLength": 1
242
- },
243
- {
244
- "type": "array",
245
- "items": {
246
- "type": "string",
247
- "minLength": 1
248
- },
249
- "minItems": 1
250
- }
251
- ]
252
- },
253
- "timeout_seconds": {
254
- "type": "number",
255
- "exclusiveMinimum": true,
256
- "minimum": 0
257
- },
258
- "cwd": {
259
- "type": "string",
260
- "minLength": 1
261
- },
262
- "env": {
263
- "type": "object",
264
- "additionalProperties": {
265
- "type": "string"
266
- }
267
- }
268
- },
269
- "additionalProperties": false
270
- }
271
- ]
272
- }
273
- }
274
- },
275
- "additionalProperties": false
276
- }
277
- }
278
- }