@plumloom/cli 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/LICENSE +202 -0
- package/NOTICE +2 -0
- package/README.md +71 -0
- package/dist/actions/configured-runs.d.ts +25 -0
- package/dist/actions/configured-runs.d.ts.map +1 -0
- package/dist/actions/configured-runs.js +35 -0
- package/dist/actions/configured-runs.js.map +1 -0
- package/dist/actions/context.d.ts +5 -0
- package/dist/actions/context.d.ts.map +1 -0
- package/dist/actions/context.js +2 -0
- package/dist/actions/context.js.map +1 -0
- package/dist/actions/doctor.d.ts +46 -0
- package/dist/actions/doctor.d.ts.map +1 -0
- package/dist/actions/doctor.js +130 -0
- package/dist/actions/doctor.js.map +1 -0
- package/dist/actions/evaluations.d.ts +87 -0
- package/dist/actions/evaluations.d.ts.map +1 -0
- package/dist/actions/evaluations.js +307 -0
- package/dist/actions/evaluations.js.map +1 -0
- package/dist/actions/gate.d.ts +63 -0
- package/dist/actions/gate.d.ts.map +1 -0
- package/dist/actions/gate.js +272 -0
- package/dist/actions/gate.js.map +1 -0
- package/dist/actions/identity.d.ts +4 -0
- package/dist/actions/identity.d.ts.map +1 -0
- package/dist/actions/identity.js +4 -0
- package/dist/actions/identity.js.map +1 -0
- package/dist/actions/models.d.ts +4 -0
- package/dist/actions/models.d.ts.map +1 -0
- package/dist/actions/models.js +4 -0
- package/dist/actions/models.js.map +1 -0
- package/dist/actions/quality-standards.d.ts +30 -0
- package/dist/actions/quality-standards.d.ts.map +1 -0
- package/dist/actions/quality-standards.js +86 -0
- package/dist/actions/quality-standards.js.map +1 -0
- package/dist/actions/results.d.ts +12 -0
- package/dist/actions/results.d.ts.map +1 -0
- package/dist/actions/results.js +65 -0
- package/dist/actions/results.js.map +1 -0
- package/dist/actions/runs.d.ts +22 -0
- package/dist/actions/runs.d.ts.map +1 -0
- package/dist/actions/runs.js +59 -0
- package/dist/actions/runs.js.map +1 -0
- package/dist/actions/suite-gate.d.ts +24 -0
- package/dist/actions/suite-gate.d.ts.map +1 -0
- package/dist/actions/suite-gate.js +40 -0
- package/dist/actions/suite-gate.js.map +1 -0
- package/dist/actions/suite.d.ts +49 -0
- package/dist/actions/suite.d.ts.map +1 -0
- package/dist/actions/suite.js +68 -0
- package/dist/actions/suite.js.map +1 -0
- package/dist/actions/validation.d.ts +3 -0
- package/dist/actions/validation.d.ts.map +1 -0
- package/dist/actions/validation.js +26 -0
- package/dist/actions/validation.js.map +1 -0
- package/dist/actions/workspaces.d.ts +13 -0
- package/dist/actions/workspaces.d.ts.map +1 -0
- package/dist/actions/workspaces.js +18 -0
- package/dist/actions/workspaces.js.map +1 -0
- package/dist/api/api.d.ts +237 -0
- package/dist/api/api.d.ts.map +1 -0
- package/dist/api/api.js +396 -0
- package/dist/api/api.js.map +1 -0
- package/dist/api/client.d.ts +38 -0
- package/dist/api/client.d.ts.map +1 -0
- package/dist/api/client.js +295 -0
- package/dist/api/client.js.map +1 -0
- package/dist/api/endpoints.d.ts +31 -0
- package/dist/api/endpoints.d.ts.map +1 -0
- package/dist/api/endpoints.js +32 -0
- package/dist/api/endpoints.js.map +1 -0
- package/dist/api/errors.d.ts +22 -0
- package/dist/api/errors.d.ts.map +1 -0
- package/dist/api/errors.js +84 -0
- package/dist/api/errors.js.map +1 -0
- package/dist/api/schemas.d.ts +169 -0
- package/dist/api/schemas.d.ts.map +1 -0
- package/dist/api/schemas.js +194 -0
- package/dist/api/schemas.js.map +1 -0
- package/dist/auth/credential-store.d.ts +9 -0
- package/dist/auth/credential-store.d.ts.map +1 -0
- package/dist/auth/credential-store.js +2 -0
- package/dist/auth/credential-store.js.map +1 -0
- package/dist/auth/credentials.d.ts +27 -0
- package/dist/auth/credentials.d.ts.map +1 -0
- package/dist/auth/credentials.js +81 -0
- package/dist/auth/credentials.js.map +1 -0
- package/dist/auth/keyring-store.d.ts +9 -0
- package/dist/auth/keyring-store.d.ts.map +1 -0
- package/dist/auth/keyring-store.js +57 -0
- package/dist/auth/keyring-store.js.map +1 -0
- package/dist/auth/login.d.ts +28 -0
- package/dist/auth/login.d.ts.map +1 -0
- package/dist/auth/login.js +27 -0
- package/dist/auth/login.js.map +1 -0
- package/dist/auth/password-prompt.d.ts +5 -0
- package/dist/auth/password-prompt.d.ts.map +1 -0
- package/dist/auth/password-prompt.js +10 -0
- package/dist/auth/password-prompt.js.map +1 -0
- package/dist/auth/redact.d.ts +3 -0
- package/dist/auth/redact.d.ts.map +1 -0
- package/dist/auth/redact.js +69 -0
- package/dist/auth/redact.js.map +1 -0
- package/dist/cli.d.ts +3 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +4 -0
- package/dist/cli.js.map +1 -0
- package/dist/commands/executor.d.ts +36 -0
- package/dist/commands/executor.d.ts.map +1 -0
- package/dist/commands/executor.js +1218 -0
- package/dist/commands/executor.js.map +1 -0
- package/dist/commands/program.d.ts +21 -0
- package/dist/commands/program.d.ts.map +1 -0
- package/dist/commands/program.js +588 -0
- package/dist/commands/program.js.map +1 -0
- package/dist/commands/types.d.ts +139 -0
- package/dist/commands/types.d.ts.map +1 -0
- package/dist/commands/types.js +2 -0
- package/dist/commands/types.js.map +1 -0
- package/dist/config.d.ts +12 -0
- package/dist/config.d.ts.map +1 -0
- package/dist/config.js +57 -0
- package/dist/config.js.map +1 -0
- package/dist/configured-run/file-inputs.d.ts +8 -0
- package/dist/configured-run/file-inputs.d.ts.map +1 -0
- package/dist/configured-run/file-inputs.js +160 -0
- package/dist/configured-run/file-inputs.js.map +1 -0
- package/dist/configured-run/validation.d.ts +103 -0
- package/dist/configured-run/validation.d.ts.map +1 -0
- package/dist/configured-run/validation.js +327 -0
- package/dist/configured-run/validation.js.map +1 -0
- package/dist/domain/common.d.ts +6 -0
- package/dist/domain/common.d.ts.map +1 -0
- package/dist/domain/common.js +10 -0
- package/dist/domain/common.js.map +1 -0
- package/dist/domain/types.d.ts +117 -0
- package/dist/domain/types.d.ts.map +1 -0
- package/dist/domain/types.js +2 -0
- package/dist/domain/types.js.map +1 -0
- package/dist/errors/autoeval-error.d.ts +13 -0
- package/dist/errors/autoeval-error.d.ts.map +1 -0
- package/dist/errors/autoeval-error.js +36 -0
- package/dist/errors/autoeval-error.js.map +1 -0
- package/dist/errors/hints.d.ts +3 -0
- package/dist/errors/hints.d.ts.map +1 -0
- package/dist/errors/hints.js +43 -0
- package/dist/errors/hints.js.map +1 -0
- package/dist/gate/decision.d.ts +97 -0
- package/dist/gate/decision.d.ts.map +1 -0
- package/dist/gate/decision.js +436 -0
- package/dist/gate/decision.js.map +1 -0
- package/dist/gate/policy.d.ts +31 -0
- package/dist/gate/policy.d.ts.map +1 -0
- package/dist/gate/policy.js +26 -0
- package/dist/gate/policy.js.map +1 -0
- package/dist/guide/eval-creation-backend.d.ts +20 -0
- package/dist/guide/eval-creation-backend.d.ts.map +1 -0
- package/dist/guide/eval-creation-backend.js +16 -0
- package/dist/guide/eval-creation-backend.js.map +1 -0
- package/dist/guide/eval-guide.d.ts +24 -0
- package/dist/guide/eval-guide.d.ts.map +1 -0
- package/dist/guide/eval-guide.js +219 -0
- package/dist/guide/eval-guide.js.map +1 -0
- package/dist/mcp/context.d.ts +21 -0
- package/dist/mcp/context.d.ts.map +1 -0
- package/dist/mcp/context.js +19 -0
- package/dist/mcp/context.js.map +1 -0
- package/dist/mcp/schemas.d.ts +206 -0
- package/dist/mcp/schemas.d.ts.map +1 -0
- package/dist/mcp/schemas.js +80 -0
- package/dist/mcp/schemas.js.map +1 -0
- package/dist/mcp/server.d.ts +5 -0
- package/dist/mcp/server.d.ts.map +1 -0
- package/dist/mcp/server.js +10 -0
- package/dist/mcp/server.js.map +1 -0
- package/dist/mcp/tools.d.ts +42 -0
- package/dist/mcp/tools.d.ts.map +1 -0
- package/dist/mcp/tools.js +287 -0
- package/dist/mcp/tools.js.map +1 -0
- package/dist/mcp.d.ts +3 -0
- package/dist/mcp.d.ts.map +1 -0
- package/dist/mcp.js +32 -0
- package/dist/mcp.js.map +1 -0
- package/dist/output/big-digits.d.ts +7 -0
- package/dist/output/big-digits.d.ts.map +1 -0
- package/dist/output/big-digits.js +43 -0
- package/dist/output/big-digits.js.map +1 -0
- package/dist/output/clipboard.d.ts +29 -0
- package/dist/output/clipboard.d.ts.map +1 -0
- package/dist/output/clipboard.js +82 -0
- package/dist/output/clipboard.js.map +1 -0
- package/dist/output/colors.d.ts +21 -0
- package/dist/output/colors.d.ts.map +1 -0
- package/dist/output/colors.js +21 -0
- package/dist/output/colors.js.map +1 -0
- package/dist/output/digest.d.ts +165 -0
- package/dist/output/digest.d.ts.map +1 -0
- package/dist/output/digest.js +771 -0
- package/dist/output/digest.js.map +1 -0
- package/dist/output/human.d.ts +95 -0
- package/dist/output/human.d.ts.map +1 -0
- package/dist/output/human.js +643 -0
- package/dist/output/human.js.map +1 -0
- package/dist/output/json.d.ts +2 -0
- package/dist/output/json.d.ts.map +1 -0
- package/dist/output/json.js +5 -0
- package/dist/output/json.js.map +1 -0
- package/dist/output/layout.d.ts +18 -0
- package/dist/output/layout.d.ts.map +1 -0
- package/dist/output/layout.js +64 -0
- package/dist/output/layout.js.map +1 -0
- package/dist/output/markdown.d.ts +16 -0
- package/dist/output/markdown.d.ts.map +1 -0
- package/dist/output/markdown.js +38 -0
- package/dist/output/markdown.js.map +1 -0
- package/dist/output/progress.d.ts +29 -0
- package/dist/output/progress.d.ts.map +1 -0
- package/dist/output/progress.js +136 -0
- package/dist/output/progress.js.map +1 -0
- package/dist/output/safe-text.d.ts +11 -0
- package/dist/output/safe-text.d.ts.map +1 -0
- package/dist/output/safe-text.js +44 -0
- package/dist/output/safe-text.js.map +1 -0
- package/dist/output/scorecard.d.ts +23 -0
- package/dist/output/scorecard.d.ts.map +1 -0
- package/dist/output/scorecard.js +506 -0
- package/dist/output/scorecard.js.map +1 -0
- package/dist/output/section.d.ts +15 -0
- package/dist/output/section.d.ts.map +1 -0
- package/dist/output/section.js +44 -0
- package/dist/output/section.js.map +1 -0
- package/dist/output/spinner.d.ts +24 -0
- package/dist/output/spinner.d.ts.map +1 -0
- package/dist/output/spinner.js +34 -0
- package/dist/output/spinner.js.map +1 -0
- package/dist/output/splash.d.ts +37 -0
- package/dist/output/splash.d.ts.map +1 -0
- package/dist/output/splash.js +169 -0
- package/dist/output/splash.js.map +1 -0
- package/dist/output/table.d.ts +33 -0
- package/dist/output/table.d.ts.map +1 -0
- package/dist/output/table.js +105 -0
- package/dist/output/table.js.map +1 -0
- package/dist/output/trend.d.ts +19 -0
- package/dist/output/trend.d.ts.map +1 -0
- package/dist/output/trend.js +38 -0
- package/dist/output/trend.js.map +1 -0
- package/dist/output/writer.d.ts +14 -0
- package/dist/output/writer.d.ts.map +1 -0
- package/dist/output/writer.js +33 -0
- package/dist/output/writer.js.map +1 -0
- package/dist/polling/clock.d.ts +9 -0
- package/dist/polling/clock.d.ts.map +1 -0
- package/dist/polling/clock.js +23 -0
- package/dist/polling/clock.js.map +1 -0
- package/dist/polling/wait-for-result.d.ts +16 -0
- package/dist/polling/wait-for-result.d.ts.map +1 -0
- package/dist/polling/wait-for-result.js +29 -0
- package/dist/polling/wait-for-result.js.map +1 -0
- package/dist/polling/wait-for-run.d.ts +14 -0
- package/dist/polling/wait-for-run.d.ts.map +1 -0
- package/dist/polling/wait-for-run.js +36 -0
- package/dist/polling/wait-for-run.js.map +1 -0
- package/dist/quickstart/prompt.d.ts +18 -0
- package/dist/quickstart/prompt.d.ts.map +1 -0
- package/dist/quickstart/prompt.js +37 -0
- package/dist/quickstart/prompt.js.map +1 -0
- package/dist/quickstart/sample.d.ts +28 -0
- package/dist/quickstart/sample.d.ts.map +1 -0
- package/dist/quickstart/sample.js +235 -0
- package/dist/quickstart/sample.js.map +1 -0
- package/dist/quickstart/selection.d.ts +53 -0
- package/dist/quickstart/selection.d.ts.map +1 -0
- package/dist/quickstart/selection.js +97 -0
- package/dist/quickstart/selection.js.map +1 -0
- package/dist/run.d.ts +28 -0
- package/dist/run.d.ts.map +1 -0
- package/dist/run.js +134 -0
- package/dist/run.js.map +1 -0
- package/dist/runtime/action-context.d.ts +19 -0
- package/dist/runtime/action-context.d.ts.map +1 -0
- package/dist/runtime/action-context.js +40 -0
- package/dist/runtime/action-context.js.map +1 -0
- package/dist/suite/execution.d.ts +46 -0
- package/dist/suite/execution.d.ts.map +1 -0
- package/dist/suite/execution.js +97 -0
- package/dist/suite/execution.js.map +1 -0
- package/dist/suite/manifest.d.ts +16 -0
- package/dist/suite/manifest.d.ts.map +1 -0
- package/dist/suite/manifest.js +92 -0
- package/dist/suite/manifest.js.map +1 -0
- package/dist/suite/reporting.d.ts +35 -0
- package/dist/suite/reporting.d.ts.map +1 -0
- package/dist/suite/reporting.js +131 -0
- package/dist/suite/reporting.js.map +1 -0
- package/dist/trace/deepseek-harness.d.ts +31 -0
- package/dist/trace/deepseek-harness.d.ts.map +1 -0
- package/dist/trace/deepseek-harness.js +412 -0
- package/dist/trace/deepseek-harness.js.map +1 -0
- package/dist/trace/import.d.ts +9 -0
- package/dist/trace/import.d.ts.map +1 -0
- package/dist/trace/import.js +45 -0
- package/dist/trace/import.js.map +1 -0
- package/dist/version.d.ts +3 -0
- package/dist/version.d.ts.map +1 -0
- package/dist/version.js +3 -0
- package/dist/version.js.map +1 -0
- package/package.json +82 -0
|
@@ -0,0 +1,327 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
import { AutoevalError } from '../errors/autoeval-error.js';
|
|
3
|
+
const referenceDocumentSchema = z
|
|
4
|
+
.object({
|
|
5
|
+
filename: z.string().trim().min(1),
|
|
6
|
+
content: z.string(),
|
|
7
|
+
})
|
|
8
|
+
.strict();
|
|
9
|
+
const configuredRunNormalizedConfigurationSchema = z.discriminatedUnion('contextType', [
|
|
10
|
+
z
|
|
11
|
+
.object({
|
|
12
|
+
evaluationName: z.string().trim().min(1).max(120).optional(),
|
|
13
|
+
contextName: z.string().trim().min(1),
|
|
14
|
+
contextType: z.literal('scenario'),
|
|
15
|
+
artifact: z.record(z.string(), z.json()),
|
|
16
|
+
expected: z.string().min(1),
|
|
17
|
+
referenceDocuments: z.array(referenceDocumentSchema).optional(),
|
|
18
|
+
selectedMetrics: z.array(z.string().trim().min(1)).min(1),
|
|
19
|
+
temperatureContext: z.number().optional(),
|
|
20
|
+
// Ignored: run planning is controlled by the evaluation service; retained for compatibility.
|
|
21
|
+
autoStopEnabled: z.boolean().optional(),
|
|
22
|
+
scenarioGeneration: z.record(z.string(), z.json()).optional(),
|
|
23
|
+
})
|
|
24
|
+
.strict(),
|
|
25
|
+
z
|
|
26
|
+
.object({
|
|
27
|
+
evaluationName: z.string().trim().min(1).max(120).optional(),
|
|
28
|
+
contextName: z.string().trim().min(1),
|
|
29
|
+
contextType: z.literal('conversation'),
|
|
30
|
+
artifact: z.record(z.string(), z.json()),
|
|
31
|
+
expected: z.string().min(1),
|
|
32
|
+
referenceDocuments: z.array(referenceDocumentSchema).optional(),
|
|
33
|
+
selectedMetrics: z.array(z.string().trim().min(1)).min(1),
|
|
34
|
+
temperatureContext: z.number().optional(),
|
|
35
|
+
// Ignored: run planning is controlled by the evaluation service; retained for compatibility.
|
|
36
|
+
autoStopEnabled: z.boolean().optional(),
|
|
37
|
+
})
|
|
38
|
+
.strict(),
|
|
39
|
+
z
|
|
40
|
+
.object({
|
|
41
|
+
evaluationName: z.string().trim().min(1).max(120).optional(),
|
|
42
|
+
contextName: z.string().trim().min(1),
|
|
43
|
+
contextType: z.literal('agent_trace'),
|
|
44
|
+
artifact: z.record(z.string(), z.json()),
|
|
45
|
+
expected: z.string().min(1),
|
|
46
|
+
referenceDocuments: z.array(referenceDocumentSchema).optional(),
|
|
47
|
+
selectedMetrics: z.array(z.string().trim().min(1)).min(1),
|
|
48
|
+
temperatureContext: z.number().optional(),
|
|
49
|
+
// Ignored: run planning is controlled by the evaluation service; retained for compatibility.
|
|
50
|
+
autoStopEnabled: z.boolean().optional(),
|
|
51
|
+
})
|
|
52
|
+
.strict(),
|
|
53
|
+
]);
|
|
54
|
+
const configuredRunLegacyScenarioConfigurationSchema = z
|
|
55
|
+
.object({
|
|
56
|
+
evaluationName: z.string().trim().min(1).max(120).optional(),
|
|
57
|
+
contextName: z.string().trim().min(1).optional(),
|
|
58
|
+
primaryModelId: z.uuid(),
|
|
59
|
+
comparisonModelIds: z.array(z.uuid()).default([]),
|
|
60
|
+
promptText: z.string().min(1),
|
|
61
|
+
scenarios: z.union([
|
|
62
|
+
z.array(z.array(z.record(z.string(), z.json())).min(1)).min(1),
|
|
63
|
+
z.array(z.record(z.string(), z.json())).min(1),
|
|
64
|
+
]),
|
|
65
|
+
referenceDocuments: z.array(referenceDocumentSchema).optional(),
|
|
66
|
+
selectedMetrics: z.array(z.string().trim().min(1)).min(1),
|
|
67
|
+
expected: z.string().min(1).optional(),
|
|
68
|
+
temperatureContext: z.number().optional(),
|
|
69
|
+
// Ignored: run planning is controlled by the evaluation service; retained for compatibility.
|
|
70
|
+
autoStopEnabled: z.boolean().optional(),
|
|
71
|
+
})
|
|
72
|
+
.strict();
|
|
73
|
+
export const configuredRunFileSchema = z
|
|
74
|
+
.object({
|
|
75
|
+
methodology: z
|
|
76
|
+
.object({
|
|
77
|
+
userSystemId: z.string().trim().min(1).optional(),
|
|
78
|
+
judgeModel: z.string().trim().min(1),
|
|
79
|
+
judgeModelId: z.uuid(),
|
|
80
|
+
runsPerScenario: z.number().int().min(1).max(10).optional(),
|
|
81
|
+
evaluatorInstructions: z.string().min(1),
|
|
82
|
+
changeLog: z.string().optional(),
|
|
83
|
+
})
|
|
84
|
+
.strict(),
|
|
85
|
+
configuration: z.union([
|
|
86
|
+
configuredRunNormalizedConfigurationSchema,
|
|
87
|
+
configuredRunLegacyScenarioConfigurationSchema,
|
|
88
|
+
]),
|
|
89
|
+
})
|
|
90
|
+
.strict();
|
|
91
|
+
const scenarioArtifactSchema = z
|
|
92
|
+
.object({
|
|
93
|
+
primaryModelId: z.uuid(),
|
|
94
|
+
comparisonModelIds: z.array(z.uuid()),
|
|
95
|
+
promptText: z.string().min(1),
|
|
96
|
+
scenarios: z.union([
|
|
97
|
+
z.array(z.array(z.record(z.string(), z.json())).min(1)).min(1),
|
|
98
|
+
z.array(z.record(z.string(), z.json())).min(1),
|
|
99
|
+
]),
|
|
100
|
+
})
|
|
101
|
+
.loose();
|
|
102
|
+
const conversationArtifactSchema = z
|
|
103
|
+
.object({
|
|
104
|
+
messages: z
|
|
105
|
+
.array(z
|
|
106
|
+
.object({
|
|
107
|
+
role: z.string().trim().min(1),
|
|
108
|
+
content: z.string(),
|
|
109
|
+
})
|
|
110
|
+
.loose())
|
|
111
|
+
.min(1),
|
|
112
|
+
})
|
|
113
|
+
.loose();
|
|
114
|
+
const agentTraceArtifactSchema = z
|
|
115
|
+
.object({
|
|
116
|
+
trace: z
|
|
117
|
+
.object({
|
|
118
|
+
resourceSpans: z.array(z.record(z.string(), z.json())).min(1),
|
|
119
|
+
})
|
|
120
|
+
.loose(),
|
|
121
|
+
})
|
|
122
|
+
.loose();
|
|
123
|
+
const PUBLIC_JUDGE_MODEL_PLACEHOLDER = '11111111-1111-4111-8111-111111111111';
|
|
124
|
+
const PUBLIC_PRIMARY_MODEL_PLACEHOLDER = '22222222-2222-4222-8222-222222222222';
|
|
125
|
+
function issuePath(path) {
|
|
126
|
+
return path.length === 0 ? 'input' : path.map(String).join('.');
|
|
127
|
+
}
|
|
128
|
+
function invalidInput(error) {
|
|
129
|
+
const details = error.issues
|
|
130
|
+
.slice(0, 8)
|
|
131
|
+
.map((issue) => `${issuePath(issue.path)}: ${issue.message}`)
|
|
132
|
+
.join('; ');
|
|
133
|
+
return new AutoevalError(`Configured run input is invalid: ${details}. Fix the listed fields and run "autoeval eval validate --input <file>" again`, {
|
|
134
|
+
kind: 'validation',
|
|
135
|
+
code: 'INVALID_CONFIGURED_RUN_SCHEMA',
|
|
136
|
+
cause: error,
|
|
137
|
+
});
|
|
138
|
+
}
|
|
139
|
+
export function parseConfiguredRunInput(parsedJson) {
|
|
140
|
+
const parsed = configuredRunFileSchema.safeParse(parsedJson);
|
|
141
|
+
if (!parsed.success) {
|
|
142
|
+
throw invalidInput(parsed.error);
|
|
143
|
+
}
|
|
144
|
+
const parsedInput = parsed.data;
|
|
145
|
+
const methodology = {
|
|
146
|
+
...(parsedInput.methodology.userSystemId !== undefined
|
|
147
|
+
? { userSystemId: parsedInput.methodology.userSystemId }
|
|
148
|
+
: {}),
|
|
149
|
+
judgeModel: parsedInput.methodology.judgeModel,
|
|
150
|
+
judgeModelId: parsedInput.methodology.judgeModelId,
|
|
151
|
+
...(parsedInput.methodology.runsPerScenario !== undefined
|
|
152
|
+
? { runsPerScenario: parsedInput.methodology.runsPerScenario }
|
|
153
|
+
: {}),
|
|
154
|
+
evaluatorInstructions: parsedInput.methodology.evaluatorInstructions,
|
|
155
|
+
...(parsedInput.methodology.changeLog !== undefined
|
|
156
|
+
? { changeLog: parsedInput.methodology.changeLog }
|
|
157
|
+
: {}),
|
|
158
|
+
};
|
|
159
|
+
let evaluationName;
|
|
160
|
+
let configuration;
|
|
161
|
+
if ('contextType' in parsedInput.configuration) {
|
|
162
|
+
evaluationName = parsedInput.configuration.evaluationName;
|
|
163
|
+
const common = {
|
|
164
|
+
contextName: parsedInput.configuration.contextName,
|
|
165
|
+
artifact: parsedInput.configuration.artifact,
|
|
166
|
+
expected: parsedInput.configuration.expected,
|
|
167
|
+
referenceDocuments: parsedInput.configuration.referenceDocuments ?? [],
|
|
168
|
+
selectedMetrics: parsedInput.configuration.selectedMetrics,
|
|
169
|
+
...(parsedInput.configuration.temperatureContext !== undefined
|
|
170
|
+
? { temperatureContext: parsedInput.configuration.temperatureContext }
|
|
171
|
+
: {}),
|
|
172
|
+
};
|
|
173
|
+
if (parsedInput.configuration.contextType === 'scenario') {
|
|
174
|
+
configuration = {
|
|
175
|
+
...common,
|
|
176
|
+
contextType: 'scenario',
|
|
177
|
+
...(parsedInput.configuration.scenarioGeneration !== undefined
|
|
178
|
+
? { scenarioGeneration: parsedInput.configuration.scenarioGeneration }
|
|
179
|
+
: {}),
|
|
180
|
+
};
|
|
181
|
+
}
|
|
182
|
+
else if (parsedInput.configuration.contextType === 'conversation') {
|
|
183
|
+
configuration = { ...common, contextType: 'conversation' };
|
|
184
|
+
}
|
|
185
|
+
else {
|
|
186
|
+
configuration = { ...common, contextType: 'agent_trace' };
|
|
187
|
+
}
|
|
188
|
+
}
|
|
189
|
+
else {
|
|
190
|
+
evaluationName = parsedInput.configuration.evaluationName;
|
|
191
|
+
const scenarioArtifact = {
|
|
192
|
+
primaryModelId: parsedInput.configuration.primaryModelId,
|
|
193
|
+
comparisonModelIds: parsedInput.configuration.comparisonModelIds,
|
|
194
|
+
promptText: parsedInput.configuration.promptText,
|
|
195
|
+
scenarios: parsedInput.configuration.scenarios,
|
|
196
|
+
};
|
|
197
|
+
configuration = {
|
|
198
|
+
contextName: parsedInput.configuration.contextName ?? 'Scenario Evaluation',
|
|
199
|
+
contextType: 'scenario',
|
|
200
|
+
artifact: scenarioArtifact,
|
|
201
|
+
expected: parsedInput.configuration.expected ?? parsedInput.configuration.promptText,
|
|
202
|
+
referenceDocuments: parsedInput.configuration.referenceDocuments ?? [],
|
|
203
|
+
selectedMetrics: parsedInput.configuration.selectedMetrics,
|
|
204
|
+
...(parsedInput.configuration.temperatureContext !== undefined
|
|
205
|
+
? { temperatureContext: parsedInput.configuration.temperatureContext }
|
|
206
|
+
: {}),
|
|
207
|
+
};
|
|
208
|
+
}
|
|
209
|
+
return {
|
|
210
|
+
methodology,
|
|
211
|
+
...(evaluationName !== undefined ? { evaluationName } : {}),
|
|
212
|
+
configuration,
|
|
213
|
+
};
|
|
214
|
+
}
|
|
215
|
+
function availableModelById(models) {
|
|
216
|
+
return new Map(models
|
|
217
|
+
.filter((model) => !model.isDeprecated && !model.isLocked)
|
|
218
|
+
.map((model) => [model.id.toLowerCase(), model]));
|
|
219
|
+
}
|
|
220
|
+
function requireEnabledModel(modelId, role, modelsById) {
|
|
221
|
+
const parsedId = z.uuid().safeParse(modelId);
|
|
222
|
+
if (!parsedId.success) {
|
|
223
|
+
throw new AutoevalError(`${role} model ID "${modelId}" must be a UUID. Replace it with an ID from "autoeval models"`, { kind: 'validation', code: 'INVALID_MODEL_ID', cause: parsedId.error });
|
|
224
|
+
}
|
|
225
|
+
const model = modelsById.get(parsedId.data.toLowerCase());
|
|
226
|
+
if (!model) {
|
|
227
|
+
throw new AutoevalError(`${role} model ID ${parsedId.data} is not enabled for this account. Run "autoeval models" and choose an enabled model ID`, { kind: 'validation', code: 'MODEL_NOT_ENABLED' });
|
|
228
|
+
}
|
|
229
|
+
return model;
|
|
230
|
+
}
|
|
231
|
+
function modelRoleError(message) {
|
|
232
|
+
throw new AutoevalError(`${message}. Choose a different model ID for that role`, {
|
|
233
|
+
kind: 'validation',
|
|
234
|
+
code: 'MODEL_ROLE_CONFLICT',
|
|
235
|
+
});
|
|
236
|
+
}
|
|
237
|
+
export function validateConfiguredRunInput(input, enabledModels) {
|
|
238
|
+
const modelsById = availableModelById(enabledModels);
|
|
239
|
+
const judge = requireEnabledModel(input.methodology.judgeModelId, 'Judge', modelsById);
|
|
240
|
+
const judgeId = judge.id.toLowerCase();
|
|
241
|
+
if (input.configuration.contextType === 'conversation') {
|
|
242
|
+
const artifact = conversationArtifactSchema.safeParse(input.configuration.artifact);
|
|
243
|
+
if (!artifact.success)
|
|
244
|
+
throw invalidInput(artifact.error);
|
|
245
|
+
return {
|
|
246
|
+
valid: true,
|
|
247
|
+
contextType: 'conversation',
|
|
248
|
+
artifactCount: artifact.data.messages.length,
|
|
249
|
+
models: { judge, comparisons: [] },
|
|
250
|
+
};
|
|
251
|
+
}
|
|
252
|
+
if (input.configuration.contextType === 'agent_trace') {
|
|
253
|
+
const artifact = agentTraceArtifactSchema.safeParse(input.configuration.artifact);
|
|
254
|
+
if (!artifact.success)
|
|
255
|
+
throw invalidInput(artifact.error);
|
|
256
|
+
return {
|
|
257
|
+
valid: true,
|
|
258
|
+
contextType: 'agent_trace',
|
|
259
|
+
artifactCount: artifact.data.trace.resourceSpans.length,
|
|
260
|
+
models: { judge, comparisons: [] },
|
|
261
|
+
};
|
|
262
|
+
}
|
|
263
|
+
const artifact = scenarioArtifactSchema.safeParse(input.configuration.artifact);
|
|
264
|
+
if (!artifact.success)
|
|
265
|
+
throw invalidInput(artifact.error);
|
|
266
|
+
const primary = requireEnabledModel(artifact.data.primaryModelId, 'Primary', modelsById);
|
|
267
|
+
const primaryId = primary.id.toLowerCase();
|
|
268
|
+
if (primaryId === judgeId) {
|
|
269
|
+
modelRoleError('The judge model cannot also be the primary model');
|
|
270
|
+
}
|
|
271
|
+
const seenComparisonIds = new Set();
|
|
272
|
+
const comparisons = artifact.data.comparisonModelIds.map((modelId, index) => {
|
|
273
|
+
const comparison = requireEnabledModel(modelId, `Comparison ${index + 1}`, modelsById);
|
|
274
|
+
const comparisonId = comparison.id.toLowerCase();
|
|
275
|
+
if (comparisonId === judgeId) {
|
|
276
|
+
modelRoleError('The judge model cannot also be a comparison model');
|
|
277
|
+
}
|
|
278
|
+
if (comparisonId === primaryId) {
|
|
279
|
+
modelRoleError('The primary model cannot also be a comparison model');
|
|
280
|
+
}
|
|
281
|
+
if (seenComparisonIds.has(comparisonId)) {
|
|
282
|
+
throw new AutoevalError(`Comparison model ID ${comparison.id} is duplicated. Remove duplicate comparison models and validate again`, { kind: 'validation', code: 'DUPLICATE_COMPARISON_MODEL' });
|
|
283
|
+
}
|
|
284
|
+
seenComparisonIds.add(comparisonId);
|
|
285
|
+
return comparison;
|
|
286
|
+
});
|
|
287
|
+
const scenarios = artifact.data.scenarios;
|
|
288
|
+
const artifactCount = Array.isArray(scenarios[0]) ? scenarios[0].length : scenarios.length;
|
|
289
|
+
return {
|
|
290
|
+
valid: true,
|
|
291
|
+
contextType: 'scenario',
|
|
292
|
+
artifactCount,
|
|
293
|
+
models: { judge, primary, comparisons },
|
|
294
|
+
};
|
|
295
|
+
}
|
|
296
|
+
/** Applies explicit CLI model UUIDs without mutating the parsed file input. */
|
|
297
|
+
export function applyModelOverrides(input, overrides) {
|
|
298
|
+
const methodology = {
|
|
299
|
+
...input.methodology,
|
|
300
|
+
...(overrides.judgeModelId === undefined ? {} : { judgeModelId: overrides.judgeModelId }),
|
|
301
|
+
};
|
|
302
|
+
if (input.configuration.contextType !== 'scenario' || overrides.primaryModelId === undefined) {
|
|
303
|
+
return { ...input, methodology };
|
|
304
|
+
}
|
|
305
|
+
return {
|
|
306
|
+
...input,
|
|
307
|
+
methodology,
|
|
308
|
+
configuration: {
|
|
309
|
+
...input.configuration,
|
|
310
|
+
artifact: {
|
|
311
|
+
...input.configuration.artifact,
|
|
312
|
+
primaryModelId: overrides.primaryModelId,
|
|
313
|
+
},
|
|
314
|
+
},
|
|
315
|
+
};
|
|
316
|
+
}
|
|
317
|
+
/** Fails fast when a sanitized public fixture was not given account model UUIDs. */
|
|
318
|
+
export function requireModelOverridesForPublicPlaceholders(input) {
|
|
319
|
+
if (input.methodology.judgeModelId === PUBLIC_JUDGE_MODEL_PLACEHOLDER) {
|
|
320
|
+
throw new AutoevalError('This eval file contains a synthetic judge model UUID. Run "autoeval models" and pass --judge-model-id <uuid>', { kind: 'validation', code: 'MODEL_OVERRIDE_REQUIRED' });
|
|
321
|
+
}
|
|
322
|
+
if (input.configuration.contextType === 'scenario' &&
|
|
323
|
+
input.configuration.artifact.primaryModelId === PUBLIC_PRIMARY_MODEL_PLACEHOLDER) {
|
|
324
|
+
throw new AutoevalError('This eval file contains a synthetic primary model UUID. Run "autoeval models" and pass --primary-model-id <uuid>', { kind: 'validation', code: 'MODEL_OVERRIDE_REQUIRED' });
|
|
325
|
+
}
|
|
326
|
+
}
|
|
327
|
+
//# sourceMappingURL=validation.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"validation.js","sourceRoot":"","sources":["../../src/configured-run/validation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB,OAAO,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAC;AAE5D,MAAM,uBAAuB,GAAG,CAAC;KAC9B,MAAM,CAAC;IACN,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAClC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;CACpB,CAAC;KACD,MAAM,EAAE,CAAC;AAEZ,MAAM,0CAA0C,GAAG,CAAC,CAAC,kBAAkB,CAAC,aAAa,EAAE;IACrF,CAAC;SACE,MAAM,CAAC;QACN,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE;QAC5D,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrC,WAAW,EAAE,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC;QAClC,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC3B,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAC,QAAQ,EAAE;QAC/D,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACzD,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACzC,6FAA6F;QAC7F,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;QACvC,kBAAkB,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,QAAQ,EAAE;KAC9D,CAAC;SACD,MAAM,EAAE;IACX,CAAC;SACE,MAAM,CAAC;QACN,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE;QAC5D,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrC,WAAW,EAAE,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC;QACtC,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC3B,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAC,QAAQ,EAAE;QAC/D,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACzD,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACzC,6FAA6F;QAC7F,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;KACxC,CAAC;SACD,MAAM,EAAE;IACX,CAAC;SACE,MAAM,CAAC;QACN,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE;QAC5D,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACrC,WAAW,EAAE,CAAC,CAAC,OAAO,CAAC,aAAa,CAAC;QACrC,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC3B,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAC,QAAQ,EAAE;QAC/D,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACzD,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QACzC,6FAA6F;QAC7F,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;KACxC,CAAC;SACD,MAAM,EAAE;CACZ,CAAC,CAAC;AAEH,MAAM,8CAA8C,GAAG,CAAC;KACrD,MAAM,CAAC;IACN,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE;IAC5D,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IAChD,cAAc,EAAE,CAAC,CAAC,IAAI,EAAE;IACxB,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IACjD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC;QACjB,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC9D,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;KAC/C,CAAC;IACF,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,uBAAuB,CAAC,CAAC,QAAQ,EAAE;IAC/D,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACzD,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IACtC,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACzC,6FAA6F;IAC7F,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;CACxC,CAAC;KACD,MAAM,EAAE,CAAC;AAEZ,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC;KACrC,MAAM,CAAC;IACN,WAAW,EAAE,CAAC;SACX,MAAM,CAAC;QACN,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;QACjD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACpC,YAAY,EAAE,CAAC,CAAC,IAAI,EAAE;QACtB,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,QAAQ,EAAE;QAC3D,qBAAqB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACxC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;KACjC,CAAC;SACD,MAAM,EAAE;IACX,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC;QACrB,0CAA0C;QAC1C,8CAA8C;KAC/C,CAAC;CACH,CAAC;KACD,MAAM,EAAE,CAAC;AAIZ,MAAM,sBAAsB,GAAG,CAAC;KAC7B,MAAM,CAAC;IACN,cAAc,EAAE,CAAC,CAAC,IAAI,EAAE;IACxB,kBAAkB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACrC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC;QACjB,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QAC9D,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;KAC/C,CAAC;CACH,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,0BAA0B,GAAG,CAAC;KACjC,MAAM,CAAC;IACN,QAAQ,EAAE,CAAC;SACR,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QAC9B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;KACpB,CAAC;SACD,KAAK,EAAE,CACX;SACA,GAAG,CAAC,CAAC,CAAC;CACV,CAAC;KACD,KAAK,EAAE,CAAC;AAEX,MAAM,wBAAwB,GAAG,CAAC;KAC/B,MAAM,CAAC;IACN,KAAK,EAAE,CAAC;SACL,MAAM,CAAC;QACN,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;KAC9D,CAAC;SACD,KAAK,EAAE;CACX,CAAC;KACD,KAAK,EAAE,CAAC;AA+BX,MAAM,8BAA8B,GAAG,sCAAsC,CAAC;AAC9E,MAAM,gCAAgC,GAAG,sCAAsC,CAAC;AAEhF,SAAS,SAAS,CAAC,IAAmB;IACpC,OAAO,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,YAAY,CAAC,KAAiB;IACrC,MAAM,OAAO,GAAG,KAAK,CAAC,MAAM;SACzB,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;SACX,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,GAAG,SAAS,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC;SAC5D,IAAI,CAAC,IAAI,CAAC,CAAC;IACd,OAAO,IAAI,aAAa,CACtB,oCAAoC,OAAO,+EAA+E,EAC1H;QACE,IAAI,EAAE,YAAY;QAClB,IAAI,EAAE,+BAA+B;QACrC,KAAK,EAAE,KAAK;KACb,CACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,UAAmB;IACzD,MAAM,MAAM,GAAG,uBAAuB,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;IAC7D,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACnC,CAAC;IACD,MAAM,WAAW,GAAG,MAAM,CAAC,IAAI,CAAC;IAEhC,MAAM,WAAW,GAAG;QAClB,GAAG,CAAC,WAAW,CAAC,WAAW,CAAC,YAAY,KAAK,SAAS;YACpD,CAAC,CAAC,EAAE,YAAY,EAAE,WAAW,CAAC,WAAW,CAAC,YAAY,EAAE;YACxD,CAAC,CAAC,EAAE,CAAC;QACP,UAAU,EAAE,WAAW,CAAC,WAAW,CAAC,UAAU;QAC9C,YAAY,EAAE,WAAW,CAAC,WAAW,CAAC,YAAY;QAClD,GAAG,CAAC,WAAW,CAAC,WAAW,CAAC,eAAe,KAAK,SAAS;YACvD,CAAC,CAAC,EAAE,eAAe,EAAE,WAAW,CAAC,WAAW,CAAC,eAAe,EAAE;YAC9D,CAAC,CAAC,EAAE,CAAC;QACP,qBAAqB,EAAE,WAAW,CAAC,WAAW,CAAC,qBAAqB;QACpE,GAAG,CAAC,WAAW,CAAC,WAAW,CAAC,SAAS,KAAK,SAAS;YACjD,CAAC,CAAC,EAAE,SAAS,EAAE,WAAW,CAAC,WAAW,CAAC,SAAS,EAAE;YAClD,CAAC,CAAC,EAAE,CAAC;KACR,CAAC;IAEF,IAAI,cAAkC,CAAC;IACvC,IAAI,aAAoC,CAAC;IACzC,IAAI,aAAa,IAAI,WAAW,CAAC,aAAa,EAAE,CAAC;QAC/C,cAAc,GAAG,WAAW,CAAC,aAAa,CAAC,cAAc,CAAC;QAC1D,MAAM,MAAM,GAAG;YACb,WAAW,EAAE,WAAW,CAAC,aAAa,CAAC,WAAW;YAClD,QAAQ,EAAE,WAAW,CAAC,aAAa,CAAC,QAAQ;YAC5C,QAAQ,EAAE,WAAW,CAAC,aAAa,CAAC,QAAQ;YAC5C,kBAAkB,EAAE,WAAW,CAAC,aAAa,CAAC,kBAAkB,IAAI,EAAE;YACtE,eAAe,EAAE,WAAW,CAAC,aAAa,CAAC,eAAe;YAC1D,GAAG,CAAC,WAAW,CAAC,aAAa,CAAC,kBAAkB,KAAK,SAAS;gBAC5D,CAAC,CAAC,EAAE,kBAAkB,EAAE,WAAW,CAAC,aAAa,CAAC,kBAAkB,EAAE;gBACtE,CAAC,CAAC,EAAE,CAAC;SACR,CAAC;QACF,IAAI,WAAW,CAAC,aAAa,CAAC,WAAW,KAAK,UAAU,EAAE,CAAC;YACzD,aAAa,GAAG;gBACd,GAAG,MAAM;gBACT,WAAW,EAAE,UAAU;gBACvB,GAAG,CAAC,WAAW,CAAC,aAAa,CAAC,kBAAkB,KAAK,SAAS;oBAC5D,CAAC,CAAC,EAAE,kBAAkB,EAAE,WAAW,CAAC,aAAa,CAAC,kBAAkB,EAAE;oBACtE,CAAC,CAAC,EAAE,CAAC;aACR,CAAC;QACJ,CAAC;aAAM,IAAI,WAAW,CAAC,aAAa,CAAC,WAAW,KAAK,cAAc,EAAE,CAAC;YACpE,aAAa,GAAG,EAAE,GAAG,MAAM,EAAE,WAAW,EAAE,cAAc,EAAE,CAAC;QAC7D,CAAC;aAAM,CAAC;YACN,aAAa,GAAG,EAAE,GAAG,MAAM,EAAE,WAAW,EAAE,aAAa,EAAE,CAAC;QAC5D,CAAC;IACH,CAAC;SAAM,CAAC;QACN,cAAc,GAAG,WAAW,CAAC,aAAa,CAAC,cAAc,CAAC;QAC1D,MAAM,gBAAgB,GAAG;YACvB,cAAc,EAAE,WAAW,CAAC,aAAa,CAAC,cAAc;YACxD,kBAAkB,EAAE,WAAW,CAAC,aAAa,CAAC,kBAAkB;YAChE,UAAU,EAAE,WAAW,CAAC,aAAa,CAAC,UAAU;YAChD,SAAS,EAAE,WAAW,CAAC,aAAa,CAAC,SAAS;SACE,CAAC;QAEnD,aAAa,GAAG;YACd,WAAW,EAAE,WAAW,CAAC,aAAa,CAAC,WAAW,IAAI,qBAAqB;YAC3E,WAAW,EAAE,UAAU;YACvB,QAAQ,EAAE,gBAAgB;YAC1B,QAAQ,EAAE,WAAW,CAAC,aAAa,CAAC,QAAQ,IAAI,WAAW,CAAC,aAAa,CAAC,UAAU;YACpF,kBAAkB,EAAE,WAAW,CAAC,aAAa,CAAC,kBAAkB,IAAI,EAAE;YACtE,eAAe,EAAE,WAAW,CAAC,aAAa,CAAC,eAAe;YAC1D,GAAG,CAAC,WAAW,CAAC,aAAa,CAAC,kBAAkB,KAAK,SAAS;gBAC5D,CAAC,CAAC,EAAE,kBAAkB,EAAE,WAAW,CAAC,aAAa,CAAC,kBAAkB,EAAE;gBACtE,CAAC,CAAC,EAAE,CAAC;SACR,CAAC;IACJ,CAAC;IAED,OAAO;QACL,WAAW;QACX,GAAG,CAAC,cAAc,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3D,aAAa;KACd,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CAAC,MAAiC;IAC3D,OAAO,IAAI,GAAG,CACZ,MAAM;SACH,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC;SACzD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,EAAE,KAAK,CAAC,CAAC,CACnD,CAAC;AACJ,CAAC;AAED,SAAS,mBAAmB,CAC1B,OAAe,EACf,IAAY,EACZ,UAA+C;IAE/C,MAAM,QAAQ,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;IAC7C,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC;QACtB,MAAM,IAAI,aAAa,CACrB,GAAG,IAAI,cAAc,OAAO,gEAAgE,EAC5F,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,kBAAkB,EAAE,KAAK,EAAE,QAAQ,CAAC,KAAK,EAAE,CACxE,CAAC;IACJ,CAAC;IACD,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;IAC1D,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,IAAI,aAAa,CACrB,GAAG,IAAI,aAAa,QAAQ,CAAC,IAAI,wFAAwF,EACzH,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,mBAAmB,EAAE,CAClD,CAAC;IACJ,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,cAAc,CAAC,OAAe;IACrC,MAAM,IAAI,aAAa,CAAC,GAAG,OAAO,6CAA6C,EAAE;QAC/E,IAAI,EAAE,YAAY;QAClB,IAAI,EAAE,qBAAqB;KAC5B,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,0BAA0B,CACxC,KAA+B,EAC/B,aAAwC;IAExC,MAAM,UAAU,GAAG,kBAAkB,CAAC,aAAa,CAAC,CAAC;IACrD,MAAM,KAAK,GAAG,mBAAmB,CAAC,KAAK,CAAC,WAAW,CAAC,YAAY,EAAE,OAAO,EAAE,UAAU,CAAC,CAAC;IACvF,MAAM,OAAO,GAAG,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC;IAEvC,IAAI,KAAK,CAAC,aAAa,CAAC,WAAW,KAAK,cAAc,EAAE,CAAC;QACvD,MAAM,QAAQ,GAAG,0BAA0B,CAAC,SAAS,CAAC,KAAK,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC;QACpF,IAAI,CAAC,QAAQ,CAAC,OAAO;YAAE,MAAM,YAAY,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAC1D,OAAO;YACL,KAAK,EAAE,IAAI;YACX,WAAW,EAAE,cAAc;YAC3B,aAAa,EAAE,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM;YAC5C,MAAM,EAAE,EAAE,KAAK,EAAE,WAAW,EAAE,EAAE,EAAE;SACnC,CAAC;IACJ,CAAC;IAED,IAAI,KAAK,CAAC,aAAa,CAAC,WAAW,KAAK,aAAa,EAAE,CAAC;QACtD,MAAM,QAAQ,GAAG,wBAAwB,CAAC,SAAS,CAAC,KAAK,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC;QAClF,IAAI,CAAC,QAAQ,CAAC,OAAO;YAAE,MAAM,YAAY,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAC1D,OAAO;YACL,KAAK,EAAE,IAAI;YACX,WAAW,EAAE,aAAa;YAC1B,aAAa,EAAE,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,aAAa,CAAC,MAAM;YACvD,MAAM,EAAE,EAAE,KAAK,EAAE,WAAW,EAAE,EAAE,EAAE;SACnC,CAAC;IACJ,CAAC;IAED,MAAM,QAAQ,GAAG,sBAAsB,CAAC,SAAS,CAAC,KAAK,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC;IAChF,IAAI,CAAC,QAAQ,CAAC,OAAO;QAAE,MAAM,YAAY,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;IAE1D,MAAM,OAAO,GAAG,mBAAmB,CAAC,QAAQ,CAAC,IAAI,CAAC,cAAc,EAAE,SAAS,EAAE,UAAU,CAAC,CAAC;IACzF,MAAM,SAAS,GAAG,OAAO,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,IAAI,SAAS,KAAK,OAAO,EAAE,CAAC;QAC1B,cAAc,CAAC,kDAAkD,CAAC,CAAC;IACrE,CAAC;IAED,MAAM,iBAAiB,GAAG,IAAI,GAAG,EAAU,CAAC;IAC5C,MAAM,WAAW,GAAG,QAAQ,CAAC,IAAI,CAAC,kBAAkB,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE;QAC1E,MAAM,UAAU,GAAG,mBAAmB,CAAC,OAAO,EAAE,cAAc,KAAK,GAAG,CAAC,EAAE,EAAE,UAAU,CAAC,CAAC;QACvF,MAAM,YAAY,GAAG,UAAU,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC;QACjD,IAAI,YAAY,KAAK,OAAO,EAAE,CAAC;YAC7B,cAAc,CAAC,mDAAmD,CAAC,CAAC;QACtE,CAAC;QACD,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;YAC/B,cAAc,CAAC,qDAAqD,CAAC,CAAC;QACxE,CAAC;QACD,IAAI,iBAAiB,CAAC,GAAG,CAAC,YAAY,CAAC,EAAE,CAAC;YACxC,MAAM,IAAI,aAAa,CACrB,uBAAuB,UAAU,CAAC,EAAE,uEAAuE,EAC3G,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,4BAA4B,EAAE,CAC3D,CAAC;QACJ,CAAC;QACD,iBAAiB,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;QACpC,OAAO,UAAU,CAAC;IACpB,CAAC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC;IAC1C,MAAM,aAAa,GAAG,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC;IAC3F,OAAO;QACL,KAAK,EAAE,IAAI;QACX,WAAW,EAAE,UAAU;QACvB,aAAa;QACb,MAAM,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,WAAW,EAAE;KACxC,CAAC;AACJ,CAAC;AAED,+EAA+E;AAC/E,MAAM,UAAU,mBAAmB,CACjC,KAA+B,EAC/B,SAAsC;IAEtC,MAAM,WAAW,GAAG;QAClB,GAAG,KAAK,CAAC,WAAW;QACpB,GAAG,CAAC,SAAS,CAAC,YAAY,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,SAAS,CAAC,YAAY,EAAE,CAAC;KAC1F,CAAC;IAEF,IAAI,KAAK,CAAC,aAAa,CAAC,WAAW,KAAK,UAAU,IAAI,SAAS,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;QAC7F,OAAO,EAAE,GAAG,KAAK,EAAE,WAAW,EAAE,CAAC;IACnC,CAAC;IAED,OAAO;QACL,GAAG,KAAK;QACR,WAAW;QACX,aAAa,EAAE;YACb,GAAG,KAAK,CAAC,aAAa;YACtB,QAAQ,EAAE;gBACR,GAAG,KAAK,CAAC,aAAa,CAAC,QAAQ;gBAC/B,cAAc,EAAE,SAAS,CAAC,cAAc;aACzC;SACF;KACF,CAAC;AACJ,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,0CAA0C,CAAC,KAA+B;IACxF,IAAI,KAAK,CAAC,WAAW,CAAC,YAAY,KAAK,8BAA8B,EAAE,CAAC;QACtE,MAAM,IAAI,aAAa,CACrB,8GAA8G,EAC9G,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,yBAAyB,EAAE,CACxD,CAAC;IACJ,CAAC;IAED,IACE,KAAK,CAAC,aAAa,CAAC,WAAW,KAAK,UAAU;QAC9C,KAAK,CAAC,aAAa,CAAC,QAAQ,CAAC,cAAc,KAAK,gCAAgC,EAChF,CAAC;QACD,MAAM,IAAI,aAAa,CACrB,kHAAkH,EAClH,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,yBAAyB,EAAE,CACxD,CAAC;IACJ,CAAC;AACH,CAAC"}
|
|
@@ -0,0 +1,6 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
export declare const uuidSchema: z.ZodUUID;
|
|
3
|
+
export type JsonValue = z.infer<ReturnType<typeof z.json>>;
|
|
4
|
+
export type JsonObject = Record<string, JsonValue>;
|
|
5
|
+
export declare function parseUuid(input: string, label: string): string;
|
|
6
|
+
//# sourceMappingURL=common.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"common.d.ts","sourceRoot":"","sources":["../../src/domain/common.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,eAAO,MAAM,UAAU,WAAW,CAAC;AACnC,MAAM,MAAM,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;AAC3D,MAAM,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;AAEnD,wBAAgB,SAAS,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAM9D"}
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
export const uuidSchema = z.uuid();
|
|
3
|
+
export function parseUuid(input, label) {
|
|
4
|
+
const parsed = uuidSchema.safeParse(input);
|
|
5
|
+
if (!parsed.success) {
|
|
6
|
+
throw new Error(`${label} must be a valid UUID.`);
|
|
7
|
+
}
|
|
8
|
+
return parsed.data;
|
|
9
|
+
}
|
|
10
|
+
//# sourceMappingURL=common.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"common.js","sourceRoot":"","sources":["../../src/domain/common.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,UAAU,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;AAInC,MAAM,UAAU,SAAS,CAAC,KAAa,EAAE,KAAa;IACpD,MAAM,MAAM,GAAG,UAAU,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IAC3C,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,IAAI,KAAK,CAAC,GAAG,KAAK,wBAAwB,CAAC,CAAC;IACpD,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC"}
|
|
@@ -0,0 +1,117 @@
|
|
|
1
|
+
import type { JsonObject } from './common.js';
|
|
2
|
+
export type UserIdentity = {
|
|
3
|
+
id: string;
|
|
4
|
+
userSystemId: string;
|
|
5
|
+
email: string;
|
|
6
|
+
name?: string;
|
|
7
|
+
};
|
|
8
|
+
export type Workspace = {
|
|
9
|
+
id: string;
|
|
10
|
+
name: string;
|
|
11
|
+
description?: string;
|
|
12
|
+
evaluationCount: number;
|
|
13
|
+
updatedAt?: string;
|
|
14
|
+
qualityStandardId?: string;
|
|
15
|
+
};
|
|
16
|
+
export type WorkspacePage = {
|
|
17
|
+
items: Workspace[];
|
|
18
|
+
total: number;
|
|
19
|
+
page: number;
|
|
20
|
+
size: number;
|
|
21
|
+
totalPages: number;
|
|
22
|
+
};
|
|
23
|
+
export type EvaluationSummary = {
|
|
24
|
+
id: string;
|
|
25
|
+
name: string;
|
|
26
|
+
workspaceId?: string;
|
|
27
|
+
updatedAt?: string;
|
|
28
|
+
};
|
|
29
|
+
export type EvaluationPage = {
|
|
30
|
+
items: EvaluationSummary[];
|
|
31
|
+
total: number;
|
|
32
|
+
page: number;
|
|
33
|
+
size: number;
|
|
34
|
+
totalPages: number;
|
|
35
|
+
};
|
|
36
|
+
export type EvaluationContextType = 'scenario' | 'conversation' | 'agent_trace';
|
|
37
|
+
export type EvaluationVersionSummary = {
|
|
38
|
+
version: number;
|
|
39
|
+
methodologyVersionId: string;
|
|
40
|
+
configVersionId: string;
|
|
41
|
+
isCurrent: boolean;
|
|
42
|
+
createdAt?: string;
|
|
43
|
+
};
|
|
44
|
+
export type EvaluationVersionHistory = {
|
|
45
|
+
evaluationId: string;
|
|
46
|
+
contextType: EvaluationContextType;
|
|
47
|
+
currentVersion?: number;
|
|
48
|
+
versions: EvaluationVersionSummary[];
|
|
49
|
+
raw: JsonObject;
|
|
50
|
+
};
|
|
51
|
+
export type EvaluationVersion = {
|
|
52
|
+
version?: number;
|
|
53
|
+
evaluationId: string;
|
|
54
|
+
methodologyVersionId: string;
|
|
55
|
+
configVersionId: string;
|
|
56
|
+
isCurrent?: boolean;
|
|
57
|
+
createdAt?: string;
|
|
58
|
+
contextType: EvaluationContextType;
|
|
59
|
+
raw: JsonObject;
|
|
60
|
+
};
|
|
61
|
+
export type Evaluation = EvaluationVersion & {
|
|
62
|
+
configuration: JsonObject;
|
|
63
|
+
};
|
|
64
|
+
export type SupportedModel = {
|
|
65
|
+
id: string;
|
|
66
|
+
provider: string;
|
|
67
|
+
displayName: string;
|
|
68
|
+
apiModelId: string;
|
|
69
|
+
isDeprecated: boolean;
|
|
70
|
+
isLocked: boolean;
|
|
71
|
+
modelKey?: string;
|
|
72
|
+
};
|
|
73
|
+
export type RunHandle = {
|
|
74
|
+
evaluationId: string;
|
|
75
|
+
runId: string;
|
|
76
|
+
status: string;
|
|
77
|
+
configVersionId: string;
|
|
78
|
+
methodologyVersionId: string;
|
|
79
|
+
};
|
|
80
|
+
export type RunStatus = {
|
|
81
|
+
evaluationId: string;
|
|
82
|
+
runId: string;
|
|
83
|
+
state: string;
|
|
84
|
+
progress?: {
|
|
85
|
+
percentage?: number;
|
|
86
|
+
runsCompleted?: number;
|
|
87
|
+
totalRuns?: number;
|
|
88
|
+
};
|
|
89
|
+
errorMessage?: string;
|
|
90
|
+
failureCode?: string;
|
|
91
|
+
raw: JsonObject;
|
|
92
|
+
};
|
|
93
|
+
export type PollOutcome = {
|
|
94
|
+
status: RunStatus;
|
|
95
|
+
elapsedMs: number;
|
|
96
|
+
};
|
|
97
|
+
export type RunExecution = {
|
|
98
|
+
run: RunHandle;
|
|
99
|
+
outcome: PollOutcome;
|
|
100
|
+
};
|
|
101
|
+
export type ScenarioResults = {
|
|
102
|
+
contextType: 'scenario';
|
|
103
|
+
modelPerformance: JsonObject;
|
|
104
|
+
scenarioComparison: JsonObject;
|
|
105
|
+
modelResponses: JsonObject;
|
|
106
|
+
};
|
|
107
|
+
export type ConversationResults = {
|
|
108
|
+
contextType: 'conversation';
|
|
109
|
+
conversation: JsonObject;
|
|
110
|
+
};
|
|
111
|
+
export type AgentTraceResults = {
|
|
112
|
+
contextType: 'agent_trace';
|
|
113
|
+
agentTrace: JsonObject;
|
|
114
|
+
trajectory?: JsonObject;
|
|
115
|
+
};
|
|
116
|
+
export type EvaluationResults = ScenarioResults | ConversationResults | AgentTraceResults;
|
|
117
|
+
//# sourceMappingURL=types.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/domain/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAE9C,MAAM,MAAM,YAAY,GAAG;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;CACf,CAAC;AAEF,MAAM,MAAM,SAAS,GAAG;IACtB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B,CAAC;AAEF,MAAM,MAAM,aAAa,GAAG;IAC1B,KAAK,EAAE,SAAS,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,iBAAiB,GAAG;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,cAAc,GAAG;IAC3B,KAAK,EAAE,iBAAiB,EAAE,CAAC;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,qBAAqB,GAAG,UAAU,GAAG,cAAc,GAAG,aAAa,CAAC;AAEhF,MAAM,MAAM,wBAAwB,GAAG;IACrC,OAAO,EAAE,MAAM,CAAC;IAChB,oBAAoB,EAAE,MAAM,CAAC;IAC7B,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,OAAO,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,wBAAwB,GAAG;IACrC,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,qBAAqB,CAAC;IACnC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,EAAE,wBAAwB,EAAE,CAAC;IACrC,GAAG,EAAE,UAAU,CAAC;CACjB,CAAC;AAEF,MAAM,MAAM,iBAAiB,GAAG;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,oBAAoB,EAAE,MAAM,CAAC;IAC7B,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,qBAAqB,CAAC;IACnC,GAAG,EAAE,UAAU,CAAC;CACjB,CAAC;AAEF,MAAM,MAAM,UAAU,GAAG,iBAAiB,GAAG;IAC3C,aAAa,EAAE,UAAU,CAAC;CAC3B,CAAC;AAEF,MAAM,MAAM,cAAc,GAAG;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,OAAO,CAAC;IACtB,QAAQ,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB,CAAC;AAEF,MAAM,MAAM,SAAS,GAAG;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,eAAe,EAAE,MAAM,CAAC;IACxB,oBAAoB,EAAE,MAAM,CAAC;CAC9B,CAAC;AAEF,MAAM,MAAM,SAAS,GAAG;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE;QACT,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,aAAa,CAAC,EAAE,MAAM,CAAC;QACvB,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB,CAAC;IACF,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,GAAG,EAAE,UAAU,CAAC;CACjB,CAAC;AAEF,MAAM,MAAM,WAAW,GAAG;IACxB,MAAM,EAAE,SAAS,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB,CAAC;AAEF,MAAM,MAAM,YAAY,GAAG;IACzB,GAAG,EAAE,SAAS,CAAC;IACf,OAAO,EAAE,WAAW,CAAC;CACtB,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,WAAW,EAAE,UAAU,CAAC;IACxB,gBAAgB,EAAE,UAAU,CAAC;IAC7B,kBAAkB,EAAE,UAAU,CAAC;IAC/B,cAAc,EAAE,UAAU,CAAC;CAC5B,CAAC;AAEF,MAAM,MAAM,mBAAmB,GAAG;IAChC,WAAW,EAAE,cAAc,CAAC;IAC5B,YAAY,EAAE,UAAU,CAAC;CAC1B,CAAC;AAEF,MAAM,MAAM,iBAAiB,GAAG;IAC9B,WAAW,EAAE,aAAa,CAAC;IAC3B,UAAU,EAAE,UAAU,CAAC;IACvB,UAAU,CAAC,EAAE,UAAU,CAAC;CACzB,CAAC;AAEF,MAAM,MAAM,iBAAiB,GAAG,eAAe,GAAG,mBAAmB,GAAG,iBAAiB,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/domain/types.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
export type AutoevalErrorKind = 'usage' | 'authentication' | 'authorization' | 'plan_restriction' | 'validation' | 'network' | 'upstream' | 'run_failed' | 'gate_failed' | 'timeout' | 'unsupported';
|
|
2
|
+
export declare class AutoevalError extends Error {
|
|
3
|
+
readonly kind: AutoevalErrorKind;
|
|
4
|
+
readonly code: string | undefined;
|
|
5
|
+
readonly exitCode: number;
|
|
6
|
+
constructor(message: string, options: {
|
|
7
|
+
kind: AutoevalErrorKind;
|
|
8
|
+
code?: string;
|
|
9
|
+
cause?: unknown;
|
|
10
|
+
});
|
|
11
|
+
}
|
|
12
|
+
export declare function asAutoevalError(error: unknown): AutoevalError;
|
|
13
|
+
//# sourceMappingURL=autoeval-error.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"autoeval-error.d.ts","sourceRoot":"","sources":["../../src/errors/autoeval-error.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,iBAAiB,GACzB,OAAO,GACP,gBAAgB,GAChB,eAAe,GACf,kBAAkB,GAClB,YAAY,GACZ,SAAS,GACT,UAAU,GACV,YAAY,GACZ,aAAa,GACb,SAAS,GACT,aAAa,CAAC;AAgBlB,qBAAa,aAAc,SAAQ,KAAK;IACtC,QAAQ,CAAC,IAAI,EAAE,iBAAiB,CAAC;IACjC,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,CAAC;IAClC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;gBAGxB,OAAO,EAAE,MAAM,EACf,OAAO,EAAE;QACP,IAAI,EAAE,iBAAiB,CAAC;QACxB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,KAAK,CAAC,EAAE,OAAO,CAAC;KACjB;CAQJ;AAED,wBAAgB,eAAe,CAAC,KAAK,EAAE,OAAO,GAAG,aAAa,CAU7D"}
|
|
@@ -0,0 +1,36 @@
|
|
|
1
|
+
const EXIT_CODES = {
|
|
2
|
+
usage: 2,
|
|
3
|
+
authentication: 3,
|
|
4
|
+
authorization: 3,
|
|
5
|
+
plan_restriction: 3,
|
|
6
|
+
validation: 2,
|
|
7
|
+
network: 4,
|
|
8
|
+
upstream: 4,
|
|
9
|
+
run_failed: 5,
|
|
10
|
+
gate_failed: 1,
|
|
11
|
+
timeout: 4,
|
|
12
|
+
unsupported: 2,
|
|
13
|
+
};
|
|
14
|
+
export class AutoevalError extends Error {
|
|
15
|
+
kind;
|
|
16
|
+
code;
|
|
17
|
+
exitCode;
|
|
18
|
+
constructor(message, options) {
|
|
19
|
+
super(message, { cause: options.cause });
|
|
20
|
+
this.name = 'AutoevalError';
|
|
21
|
+
this.kind = options.kind;
|
|
22
|
+
this.code = options.code;
|
|
23
|
+
this.exitCode = EXIT_CODES[options.kind];
|
|
24
|
+
}
|
|
25
|
+
}
|
|
26
|
+
export function asAutoevalError(error) {
|
|
27
|
+
if (error instanceof AutoevalError) {
|
|
28
|
+
return error;
|
|
29
|
+
}
|
|
30
|
+
return new AutoevalError('Autoeval could not complete the request.', {
|
|
31
|
+
kind: 'upstream',
|
|
32
|
+
code: 'UNEXPECTED_ERROR',
|
|
33
|
+
cause: error,
|
|
34
|
+
});
|
|
35
|
+
}
|
|
36
|
+
//# sourceMappingURL=autoeval-error.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"autoeval-error.js","sourceRoot":"","sources":["../../src/errors/autoeval-error.ts"],"names":[],"mappings":"AAaA,MAAM,UAAU,GAAgD;IAC9D,KAAK,EAAE,CAAC;IACR,cAAc,EAAE,CAAC;IACjB,aAAa,EAAE,CAAC;IAChB,gBAAgB,EAAE,CAAC;IACnB,UAAU,EAAE,CAAC;IACb,OAAO,EAAE,CAAC;IACV,QAAQ,EAAE,CAAC;IACX,UAAU,EAAE,CAAC;IACb,WAAW,EAAE,CAAC;IACd,OAAO,EAAE,CAAC;IACV,WAAW,EAAE,CAAC;CACf,CAAC;AAEF,MAAM,OAAO,aAAc,SAAQ,KAAK;IAC7B,IAAI,CAAoB;IACxB,IAAI,CAAqB;IACzB,QAAQ,CAAS;IAE1B,YACE,OAAe,EACf,OAIC;QAED,KAAK,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC;QACzC,IAAI,CAAC,IAAI,GAAG,eAAe,CAAC;QAC5B,IAAI,CAAC,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;QACzB,IAAI,CAAC,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;QACzB,IAAI,CAAC,QAAQ,GAAG,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC;CACF;AAED,MAAM,UAAU,eAAe,CAAC,KAAc;IAC5C,IAAI,KAAK,YAAY,aAAa,EAAE,CAAC;QACnC,OAAO,KAAK,CAAC;IACf,CAAC;IAED,OAAO,IAAI,aAAa,CAAC,0CAA0C,EAAE;QACnE,IAAI,EAAE,UAAU;QAChB,IAAI,EAAE,kBAAkB;QACxB,KAAK,EAAE,KAAK;KACb,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"hints.d.ts","sourceRoot":"","sources":["../../src/errors/hints.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAqB,MAAM,qBAAqB,CAAC;AAkD5E,wBAAgB,YAAY,CAAC,KAAK,EAAE,IAAI,CAAC,aAAa,EAAE,MAAM,GAAG,MAAM,CAAC,GAAG,MAAM,GAAG,SAAS,CAG5F"}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Remediation hints are advisory only. They never contain credentials, request
|
|
3
|
+
* payloads, or upstream implementation details.
|
|
4
|
+
*/
|
|
5
|
+
const HINTS_BY_CODE = {
|
|
6
|
+
CREDENTIAL_REQUIRED: 'Run `autoeval login` (it will prompt for a key), pass `autoeval login --key <pl_sk_...>`, or set AUTOEVAL_API_KEY',
|
|
7
|
+
INVALID_CLI_KEY_FORMAT: 'Create a current CLI key at https://app.plumloom.ai (it starts with pl_sk_), then run `autoeval login` or `autoeval login --key <key>`',
|
|
8
|
+
MISSING_API_BASE_URL: 'Set AUTOEVAL_API_BASE_URL to the Autoeval API origin',
|
|
9
|
+
INVALID_BASE_URL: 'Set AUTOEVAL_API_BASE_URL to an https origin without a path or query',
|
|
10
|
+
INSECURE_BASE_URL: 'Use an https origin; http is allowed only for localhost',
|
|
11
|
+
INVALID_ENVIRONMENT: 'Check the Autoeval environment variables listed in .env.example',
|
|
12
|
+
RUN_POLL_TIMEOUT: 'Check progress later with `autoeval status <evaluation-id> <run-id>`; the run keeps going server-side',
|
|
13
|
+
RESULT_POLL_TIMEOUT: 'Results are still being prepared; retry `autoeval results <evaluation-id> <run-id>` shortly',
|
|
14
|
+
REQUEST_ABORTED: 'The command was interrupted; rerun it when ready',
|
|
15
|
+
INVALID_QUALITY_STANDARD_SCHEMA: 'Compare your file with the canonical schema in docs/command-reference.md',
|
|
16
|
+
INVALID_QUALITY_STANDARD_JSON: 'Validate the file with a JSON linter, then retry',
|
|
17
|
+
QUALITY_STANDARD_INPUT_READ_FAILED: 'Check the --input path relative to your current directory',
|
|
18
|
+
CONFIGURED_RUN_INPUT_READ_FAILED: 'Check the --input path relative to your current directory',
|
|
19
|
+
INVALID_CONFIGURED_RUN_JSON: 'Validate the file with a JSON linter, then retry',
|
|
20
|
+
GATE_THRESHOLD_NOT_MET: 'Review the threshold table above, then fix the regression or update the gate thresholds',
|
|
21
|
+
INVALID_GATE_THRESHOLD: 'Thresholds must be numbers, for example --min-overall 4.0',
|
|
22
|
+
NO_GATE_THRESHOLD: 'Set at least one threshold, for example --min-overall 4.0, or pass --thresholds <file>',
|
|
23
|
+
GATE_THRESHOLDS_READ_FAILED: 'Check the --thresholds path relative to your current directory',
|
|
24
|
+
INVALID_GATE_THRESHOLDS_JSON: 'Validate the thresholds file with a JSON linter, then retry',
|
|
25
|
+
};
|
|
26
|
+
const HINTS_BY_KIND = {
|
|
27
|
+
authentication: 'Run `autoeval login`, or set AUTOEVAL_API_KEY in your environment',
|
|
28
|
+
authorization: 'Confirm your account has access to this workspace or evaluation',
|
|
29
|
+
plan_restriction: 'This action needs a different Plumloom plan; contact your workspace owner',
|
|
30
|
+
network: 'Check your connection and AUTOEVAL_API_BASE_URL, then retry',
|
|
31
|
+
timeout: 'The request timed out; retry, or poll status separately for long runs',
|
|
32
|
+
usage: 'Run the command with --help to see required options and examples',
|
|
33
|
+
validation: 'Run the command with --help to see required options and examples',
|
|
34
|
+
run_failed: 'Inspect the run with `autoeval status <evaluation-id> <run-id>`',
|
|
35
|
+
gate_failed: 'Inspect the failing metric with `autoeval results <evaluation-id> <run-id>`, then adjust the evaluation or the thresholds',
|
|
36
|
+
upstream: 'Retry shortly; rerun with --debug to capture redacted request diagnostics',
|
|
37
|
+
unsupported: 'Run `autoeval --help` to see the supported commands',
|
|
38
|
+
};
|
|
39
|
+
export function hintForError(error) {
|
|
40
|
+
const byCode = error.code === undefined ? undefined : HINTS_BY_CODE[error.code];
|
|
41
|
+
return byCode ?? HINTS_BY_KIND[error.kind];
|
|
42
|
+
}
|
|
43
|
+
//# sourceMappingURL=hints.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"hints.js","sourceRoot":"","sources":["../../src/errors/hints.ts"],"names":[],"mappings":"AAEA;;;GAGG;AACH,MAAM,aAAa,GAAqC;IACtD,mBAAmB,EACjB,mHAAmH;IACrH,sBAAsB,EACpB,wIAAwI;IAC1I,oBAAoB,EAAE,sDAAsD;IAC5E,gBAAgB,EAAE,sEAAsE;IACxF,iBAAiB,EAAE,yDAAyD;IAC5E,mBAAmB,EAAE,iEAAiE;IACtF,gBAAgB,EACd,uGAAuG;IACzG,mBAAmB,EACjB,6FAA6F;IAC/F,eAAe,EAAE,kDAAkD;IACnE,+BAA+B,EAC7B,0EAA0E;IAC5E,6BAA6B,EAAE,kDAAkD;IACjF,kCAAkC,EAAE,2DAA2D;IAC/F,gCAAgC,EAAE,2DAA2D;IAC7F,2BAA2B,EAAE,kDAAkD;IAC/E,sBAAsB,EACpB,yFAAyF;IAC3F,sBAAsB,EAAE,2DAA2D;IACnF,iBAAiB,EACf,wFAAwF;IAC1F,2BAA2B,EAAE,gEAAgE;IAC7F,4BAA4B,EAAE,6DAA6D;CAC5F,CAAC;AAEF,MAAM,aAAa,GAAyD;IAC1E,cAAc,EAAE,mEAAmE;IACnF,aAAa,EAAE,iEAAiE;IAChF,gBAAgB,EAAE,2EAA2E;IAC7F,OAAO,EAAE,6DAA6D;IACtE,OAAO,EAAE,uEAAuE;IAChF,KAAK,EAAE,kEAAkE;IACzE,UAAU,EAAE,kEAAkE;IAC9E,UAAU,EAAE,iEAAiE;IAC7E,WAAW,EACT,2HAA2H;IAC7H,QAAQ,EAAE,2EAA2E;IACrF,WAAW,EAAE,qDAAqD;CACnE,CAAC;AAEF,MAAM,UAAU,YAAY,CAAC,KAA2C;IACtE,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChF,OAAO,MAAM,IAAI,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;AAC7C,CAAC"}
|