@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 +1,139 @@
|
|
|
1
|
+
import type { GateThresholds } from '../actions/gate.js';
|
|
2
|
+
import type { QuickstartSampleKind } from '../quickstart/sample.js';
|
|
3
|
+
export type DeterministicCommand = {
|
|
4
|
+
kind: 'login';
|
|
5
|
+
key?: string;
|
|
6
|
+
} | {
|
|
7
|
+
kind: 'logout';
|
|
8
|
+
} | {
|
|
9
|
+
kind: 'whoami';
|
|
10
|
+
} | {
|
|
11
|
+
kind: 'workspace-list';
|
|
12
|
+
} | {
|
|
13
|
+
kind: 'workspace-create';
|
|
14
|
+
name: string;
|
|
15
|
+
description?: string;
|
|
16
|
+
} | {
|
|
17
|
+
kind: 'evaluation-list';
|
|
18
|
+
workspaceId: string;
|
|
19
|
+
} | {
|
|
20
|
+
kind: 'evaluation-create';
|
|
21
|
+
workspaceId: string;
|
|
22
|
+
name?: string;
|
|
23
|
+
} | {
|
|
24
|
+
kind: 'evaluation-update-title';
|
|
25
|
+
workspaceId: string;
|
|
26
|
+
evaluationId: string;
|
|
27
|
+
name: string;
|
|
28
|
+
userSystemId: string;
|
|
29
|
+
description?: string;
|
|
30
|
+
page?: number;
|
|
31
|
+
size?: number;
|
|
32
|
+
} | {
|
|
33
|
+
kind: 'quality-standard-create';
|
|
34
|
+
inputFile: string;
|
|
35
|
+
workspaceId?: string;
|
|
36
|
+
} | {
|
|
37
|
+
kind: 'quality-standard-update';
|
|
38
|
+
qualityStandardId: string;
|
|
39
|
+
inputFile: string;
|
|
40
|
+
} | {
|
|
41
|
+
kind: 'quality-standard-assign';
|
|
42
|
+
workspaceId: string;
|
|
43
|
+
qualityStandardId: string;
|
|
44
|
+
} | {
|
|
45
|
+
kind: 'quality-standard-workspace';
|
|
46
|
+
workspaceId: string;
|
|
47
|
+
} | {
|
|
48
|
+
kind: 'quality-standard-show';
|
|
49
|
+
qualityStandardId: string;
|
|
50
|
+
} | {
|
|
51
|
+
kind: 'evaluation-validate';
|
|
52
|
+
inputFile: string;
|
|
53
|
+
} | {
|
|
54
|
+
kind: 'evaluation-create-from';
|
|
55
|
+
workspaceId: string;
|
|
56
|
+
inputFiles: readonly string[];
|
|
57
|
+
run: boolean;
|
|
58
|
+
judgeModelId?: string;
|
|
59
|
+
primaryModelId?: string;
|
|
60
|
+
runConcurrency?: number;
|
|
61
|
+
runStartStaggerMs?: number;
|
|
62
|
+
} | {
|
|
63
|
+
kind: 'trace-import';
|
|
64
|
+
source: 'deepseek-harness';
|
|
65
|
+
sessionFile: string;
|
|
66
|
+
templateFile: string;
|
|
67
|
+
outputFile: string;
|
|
68
|
+
evaluationName?: string;
|
|
69
|
+
serviceName?: string;
|
|
70
|
+
workspaceId?: string;
|
|
71
|
+
run: boolean;
|
|
72
|
+
} | {
|
|
73
|
+
kind: 'suite-run';
|
|
74
|
+
manifestFile: string;
|
|
75
|
+
workspaceId?: string;
|
|
76
|
+
concurrency?: number;
|
|
77
|
+
staggerMs?: number;
|
|
78
|
+
judgeModelId?: string;
|
|
79
|
+
primaryModelId?: string;
|
|
80
|
+
} | {
|
|
81
|
+
kind: 'suite-gate';
|
|
82
|
+
manifestFile: string;
|
|
83
|
+
workspaceId?: string;
|
|
84
|
+
concurrency?: number;
|
|
85
|
+
staggerMs?: number;
|
|
86
|
+
judgeModelId?: string;
|
|
87
|
+
primaryModelId?: string;
|
|
88
|
+
} | {
|
|
89
|
+
kind: 'evaluation-run-configured';
|
|
90
|
+
evaluationId: string;
|
|
91
|
+
inputFile: string;
|
|
92
|
+
} | {
|
|
93
|
+
kind: 'evaluation-versions';
|
|
94
|
+
evaluationId: string;
|
|
95
|
+
} | {
|
|
96
|
+
kind: 'evaluation-show';
|
|
97
|
+
evaluationId: string;
|
|
98
|
+
version?: number;
|
|
99
|
+
} | {
|
|
100
|
+
kind: 'models';
|
|
101
|
+
} | {
|
|
102
|
+
kind: 'quickstart';
|
|
103
|
+
workspaceId?: string;
|
|
104
|
+
judgeModelId?: string;
|
|
105
|
+
primaryModelId?: string;
|
|
106
|
+
inputFile?: string;
|
|
107
|
+
sample: QuickstartSampleKind;
|
|
108
|
+
assumeYes: boolean;
|
|
109
|
+
} | {
|
|
110
|
+
kind: 'doctor';
|
|
111
|
+
workspaceId?: string;
|
|
112
|
+
manifestFile?: string;
|
|
113
|
+
inputFiles: readonly string[];
|
|
114
|
+
} | {
|
|
115
|
+
kind: 'run';
|
|
116
|
+
evaluationId: string;
|
|
117
|
+
} | {
|
|
118
|
+
kind: 'status';
|
|
119
|
+
evaluationId: string;
|
|
120
|
+
runId: string;
|
|
121
|
+
} | {
|
|
122
|
+
kind: 'results';
|
|
123
|
+
evaluationId: string;
|
|
124
|
+
runId: string;
|
|
125
|
+
showOutputs?: boolean;
|
|
126
|
+
} | {
|
|
127
|
+
kind: 'gate';
|
|
128
|
+
evaluationId: string;
|
|
129
|
+
thresholdsFile?: string;
|
|
130
|
+
thresholds: GateThresholds;
|
|
131
|
+
};
|
|
132
|
+
export type ExecutionOptions = {
|
|
133
|
+
json: boolean;
|
|
134
|
+
debug: boolean;
|
|
135
|
+
};
|
|
136
|
+
export interface CommandExecutor {
|
|
137
|
+
execute(command: DeterministicCommand, options: ExecutionOptions): Promise<void>;
|
|
138
|
+
}
|
|
139
|
+
//# sourceMappingURL=types.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/commands/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAEpE,MAAM,MAAM,oBAAoB,GAC5B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAA;CAAE,GAC/B;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,GAClB;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,GAClB;IAAE,IAAI,EAAE,gBAAgB,CAAA;CAAE,GAC1B;IAAE,IAAI,EAAE,kBAAkB,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAChE;IAAE,IAAI,EAAE,iBAAiB,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,IAAI,EAAE,mBAAmB,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAA;CAAE,GACjE;IACE,IAAI,EAAE,yBAAyB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;CACf,GACD;IAAE,IAAI,EAAE,yBAAyB,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAC5E;IAAE,IAAI,EAAE,yBAAyB,CAAC;IAAC,iBAAiB,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,GACjF;IACE,IAAI,EAAE,yBAAyB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,iBAAiB,EAAE,MAAM,CAAC;CAC3B,GACD;IAAE,IAAI,EAAE,4BAA4B,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,GAC3D;IAAE,IAAI,EAAE,uBAAuB,CAAC;IAAC,iBAAiB,EAAE,MAAM,CAAA;CAAE,GAC5D;IAAE,IAAI,EAAE,qBAAqB,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,GAClD;IACE,IAAI,EAAE,wBAAwB,CAAC;IAC/B,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,SAAS,MAAM,EAAE,CAAC;IAC9B,GAAG,EAAE,OAAO,CAAC;IACb,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B,GACD;IACE,IAAI,EAAE,cAAc,CAAC;IACrB,MAAM,EAAE,kBAAkB,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,GAAG,EAAE,OAAO,CAAC;CACd,GACD;IACE,IAAI,EAAE,WAAW,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,GACD;IACE,IAAI,EAAE,YAAY,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,GACD;IAAE,IAAI,EAAE,2BAA2B,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,GAC9E;IAAE,IAAI,EAAE,qBAAqB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GACrD;IAAE,IAAI,EAAE,iBAAiB,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,MAAM,CAAA;CAAE,GACnE;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,GAClB;IACE,IAAI,EAAE,YAAY,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,oBAAoB,CAAC;IAC7B,SAAS,EAAE,OAAO,CAAC;CACpB,GACD;IACE,IAAI,EAAE,QAAQ,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,SAAS,MAAM,EAAE,CAAC;CAC/B,GACD;IAAE,IAAI,EAAE,KAAK,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,GACrC;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GACvD;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,OAAO,CAAA;CAAE,GAC/E;IACE,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,cAAc,CAAC;CAC5B,CAAC;AAEN,MAAM,MAAM,gBAAgB,GAAG;IAC7B,IAAI,EAAE,OAAO,CAAC;IACd,KAAK,EAAE,OAAO,CAAC;CAChB,CAAC;AAEF,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,OAAO,EAAE,oBAAoB,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAClF"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/commands/types.ts"],"names":[],"mappings":""}
|
package/dist/config.d.ts
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
export type AutoevalConfiguration = {
|
|
2
|
+
apiBaseUrl: URL;
|
|
3
|
+
requestTimeoutMs: number;
|
|
4
|
+
maxResponseBytes: number;
|
|
5
|
+
pollIntervalMs: number;
|
|
6
|
+
pollTimeoutMs: number;
|
|
7
|
+
resultReadyIntervalMs: number;
|
|
8
|
+
resultReadyTimeoutMs: number;
|
|
9
|
+
};
|
|
10
|
+
export declare function parseApiBaseUrl(input: string): URL;
|
|
11
|
+
export declare function loadConfiguration(environment?: Readonly<Record<string, string | undefined>>): AutoevalConfiguration;
|
|
12
|
+
//# sourceMappingURL=config.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"config.d.ts","sourceRoot":"","sources":["../src/config.ts"],"names":[],"mappings":"AAEA,MAAM,MAAM,qBAAqB,GAAG;IAClC,UAAU,EAAE,GAAG,CAAC;IAChB,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,cAAc,EAAE,MAAM,CAAC;IACvB,aAAa,EAAE,MAAM,CAAC;IACtB,qBAAqB,EAAE,MAAM,CAAC;IAC9B,oBAAoB,EAAE,MAAM,CAAC;CAC9B,CAAC;AAMF,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,GAAG,CA4ClD;AAED,wBAAgB,iBAAiB,CAC/B,WAAW,GAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAe,GACtE,qBAAqB,CAqBvB"}
|
package/dist/config.js
ADDED
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
import { AutoevalError } from './errors/autoeval-error.js';
|
|
2
|
+
function isLocalhost(hostname) {
|
|
3
|
+
return hostname === 'localhost' || hostname === '127.0.0.1' || hostname === '[::1]';
|
|
4
|
+
}
|
|
5
|
+
export function parseApiBaseUrl(input) {
|
|
6
|
+
let parsed;
|
|
7
|
+
try {
|
|
8
|
+
parsed = new URL(input);
|
|
9
|
+
}
|
|
10
|
+
catch (error) {
|
|
11
|
+
throw new AutoevalError('AUTOEVAL_API_BASE_URL must be a valid URL.', {
|
|
12
|
+
kind: 'validation',
|
|
13
|
+
code: 'INVALID_BASE_URL',
|
|
14
|
+
cause: error,
|
|
15
|
+
});
|
|
16
|
+
}
|
|
17
|
+
if (parsed.protocol !== 'https:' &&
|
|
18
|
+
!(parsed.protocol === 'http:' && isLocalhost(parsed.hostname))) {
|
|
19
|
+
throw new AutoevalError('The Autoeval API must use HTTPS; HTTP is allowed only for localhost development.', {
|
|
20
|
+
kind: 'validation',
|
|
21
|
+
code: 'INSECURE_BASE_URL',
|
|
22
|
+
});
|
|
23
|
+
}
|
|
24
|
+
if (parsed.username || parsed.password || parsed.search || parsed.hash) {
|
|
25
|
+
throw new AutoevalError('AUTOEVAL_API_BASE_URL must not contain credentials, query parameters, or fragments.', {
|
|
26
|
+
kind: 'validation',
|
|
27
|
+
code: 'INVALID_BASE_URL',
|
|
28
|
+
});
|
|
29
|
+
}
|
|
30
|
+
if (parsed.pathname !== '/' && parsed.pathname !== '') {
|
|
31
|
+
throw new AutoevalError('AUTOEVAL_API_BASE_URL must be an origin without a path.', {
|
|
32
|
+
kind: 'validation',
|
|
33
|
+
code: 'INVALID_BASE_URL',
|
|
34
|
+
});
|
|
35
|
+
}
|
|
36
|
+
parsed.pathname = '/';
|
|
37
|
+
return parsed;
|
|
38
|
+
}
|
|
39
|
+
export function loadConfiguration(environment = process.env) {
|
|
40
|
+
const apiBaseUrl = environment.AUTOEVAL_API_BASE_URL?.trim();
|
|
41
|
+
if (!apiBaseUrl) {
|
|
42
|
+
throw new AutoevalError('AUTOEVAL_API_BASE_URL is not defined. Set it to the Autoeval API origin to continue.', {
|
|
43
|
+
kind: 'validation',
|
|
44
|
+
code: 'MISSING_API_BASE_URL',
|
|
45
|
+
});
|
|
46
|
+
}
|
|
47
|
+
return {
|
|
48
|
+
apiBaseUrl: parseApiBaseUrl(apiBaseUrl),
|
|
49
|
+
requestTimeoutMs: 30_000,
|
|
50
|
+
maxResponseBytes: 2 * 1024 * 1024,
|
|
51
|
+
pollIntervalMs: 2_000,
|
|
52
|
+
pollTimeoutMs: 10 * 60_000,
|
|
53
|
+
resultReadyIntervalMs: 2_000,
|
|
54
|
+
resultReadyTimeoutMs: 60_000,
|
|
55
|
+
};
|
|
56
|
+
}
|
|
57
|
+
//# sourceMappingURL=config.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"config.js","sourceRoot":"","sources":["../src/config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAY3D,SAAS,WAAW,CAAC,QAAgB;IACnC,OAAO,QAAQ,KAAK,WAAW,IAAI,QAAQ,KAAK,WAAW,IAAI,QAAQ,KAAK,OAAO,CAAC;AACtF,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,KAAa;IAC3C,IAAI,MAAW,CAAC;IAChB,IAAI,CAAC;QACH,MAAM,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC;IAC1B,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,IAAI,aAAa,CAAC,4CAA4C,EAAE;YACpE,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,kBAAkB;YACxB,KAAK,EAAE,KAAK;SACb,CAAC,CAAC;IACL,CAAC;IAED,IACE,MAAM,CAAC,QAAQ,KAAK,QAAQ;QAC5B,CAAC,CAAC,MAAM,CAAC,QAAQ,KAAK,OAAO,IAAI,WAAW,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,EAC9D,CAAC;QACD,MAAM,IAAI,aAAa,CACrB,kFAAkF,EAClF;YACE,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,mBAAmB;SAC1B,CACF,CAAC;IACJ,CAAC;IAED,IAAI,MAAM,CAAC,QAAQ,IAAI,MAAM,CAAC,QAAQ,IAAI,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,IAAI,EAAE,CAAC;QACvE,MAAM,IAAI,aAAa,CACrB,qFAAqF,EACrF;YACE,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,kBAAkB;SACzB,CACF,CAAC;IACJ,CAAC;IAED,IAAI,MAAM,CAAC,QAAQ,KAAK,GAAG,IAAI,MAAM,CAAC,QAAQ,KAAK,EAAE,EAAE,CAAC;QACtD,MAAM,IAAI,aAAa,CAAC,yDAAyD,EAAE;YACjF,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,kBAAkB;SACzB,CAAC,CAAC;IACL,CAAC;IAED,MAAM,CAAC,QAAQ,GAAG,GAAG,CAAC;IACtB,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,iBAAiB,CAC/B,cAA4D,OAAO,CAAC,GAAG;IAEvE,MAAM,UAAU,GAAG,WAAW,CAAC,qBAAqB,EAAE,IAAI,EAAE,CAAC;IAC7D,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,MAAM,IAAI,aAAa,CACrB,sFAAsF,EACtF;YACE,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,sBAAsB;SAC7B,CACF,CAAC;IACJ,CAAC;IAED,OAAO;QACL,UAAU,EAAE,eAAe,CAAC,UAAU,CAAC;QACvC,gBAAgB,EAAE,MAAM;QACxB,gBAAgB,EAAE,CAAC,GAAG,IAAI,GAAG,IAAI;QACjC,cAAc,EAAE,KAAK;QACrB,aAAa,EAAE,EAAE,GAAG,MAAM;QAC1B,qBAAqB,EAAE,KAAK;QAC5B,oBAAoB,EAAE,MAAM;KAC7B,CAAC;AACJ,CAAC"}
|
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Expands `artifactFile` and `referenceDocumentFiles` in an already-parsed eval
|
|
3
|
+
* file. `baseDirectory` is the directory holding the eval file.
|
|
4
|
+
*/
|
|
5
|
+
export declare function resolveEvalFileInputs(parsedJson: unknown, baseDirectory: string): Promise<unknown>;
|
|
6
|
+
/** Reads an eval file from disk and expands any file references it declares. */
|
|
7
|
+
export declare function readEvalFile(path: string): Promise<unknown>;
|
|
8
|
+
//# sourceMappingURL=file-inputs.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"file-inputs.d.ts","sourceRoot":"","sources":["../../src/configured-run/file-inputs.ts"],"names":[],"mappings":"AAsGA;;;GAGG;AACH,wBAAsB,qBAAqB,CACzC,UAAU,EAAE,OAAO,EACnB,aAAa,EAAE,MAAM,GACpB,OAAO,CAAC,OAAO,CAAC,CAuDlB;AAED,gFAAgF;AAChF,wBAAsB,YAAY,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAkBjE"}
|
|
@@ -0,0 +1,160 @@
|
|
|
1
|
+
import { readFile } from 'node:fs/promises';
|
|
2
|
+
import { basename, dirname, isAbsolute, resolve } from 'node:path';
|
|
3
|
+
import { AutoevalError } from '../errors/autoeval-error.js';
|
|
4
|
+
/**
|
|
5
|
+
* Reading an eval file that points at its own inputs.
|
|
6
|
+
*
|
|
7
|
+
* A configured-run file is the contract between the developer and Autoeval, and
|
|
8
|
+
* it is committed alongside the code. Inlining a whole transcript, trace, or
|
|
9
|
+
* policy document into it makes that file unreadable in review, so the file may
|
|
10
|
+
* instead reference them by path:
|
|
11
|
+
*
|
|
12
|
+
* "artifactFile": "./data/conversation.json"
|
|
13
|
+
* "referenceDocumentFiles": ["./policy/escalation.md"]
|
|
14
|
+
*
|
|
15
|
+
* Paths resolve relative to the eval file, so a checkout can live anywhere. The
|
|
16
|
+
* references are expanded here, before validation, into exactly the inline
|
|
17
|
+
* shape the existing schema and run path already accept — nothing downstream
|
|
18
|
+
* knows the difference.
|
|
19
|
+
*/
|
|
20
|
+
const MAX_ARTIFACT_CHARS = 2_000_000;
|
|
21
|
+
const MAX_REFERENCE_DOCUMENT_CHARS = 200_000;
|
|
22
|
+
function readFailure(path, error) {
|
|
23
|
+
return new AutoevalError(`File was not found or could not be read: ${path}`, {
|
|
24
|
+
kind: 'usage',
|
|
25
|
+
code: 'EVAL_FILE_READ_FAILED',
|
|
26
|
+
cause: error,
|
|
27
|
+
});
|
|
28
|
+
}
|
|
29
|
+
async function readBounded(path, limit) {
|
|
30
|
+
let content;
|
|
31
|
+
try {
|
|
32
|
+
content = await readFile(path, 'utf8');
|
|
33
|
+
}
|
|
34
|
+
catch (error) {
|
|
35
|
+
throw readFailure(path, error);
|
|
36
|
+
}
|
|
37
|
+
if (content.length > limit) {
|
|
38
|
+
throw new AutoevalError(`${path} is larger than ${limit} characters.`, {
|
|
39
|
+
kind: 'validation',
|
|
40
|
+
code: 'EVAL_FILE_TOO_LARGE',
|
|
41
|
+
});
|
|
42
|
+
}
|
|
43
|
+
return content;
|
|
44
|
+
}
|
|
45
|
+
function resolveAgainst(baseDirectory, path) {
|
|
46
|
+
return isAbsolute(path) ? path : resolve(baseDirectory, path);
|
|
47
|
+
}
|
|
48
|
+
function asRecord(value) {
|
|
49
|
+
return typeof value === 'object' && value !== null && !Array.isArray(value)
|
|
50
|
+
? value
|
|
51
|
+
: undefined;
|
|
52
|
+
}
|
|
53
|
+
/**
|
|
54
|
+
* An agent trace file is usually the raw OTLP export, whose top level is
|
|
55
|
+
* `resourceSpans`. The configuration schema expects it under `trace`, so the
|
|
56
|
+
* raw export is wrapped rather than rejected.
|
|
57
|
+
*/
|
|
58
|
+
function normalizeArtifact(contextType, artifact, path) {
|
|
59
|
+
if (contextType !== 'agent_trace')
|
|
60
|
+
return artifact;
|
|
61
|
+
const record = asRecord(artifact);
|
|
62
|
+
if (!record) {
|
|
63
|
+
throw new AutoevalError(`${path} must contain a JSON object.`, {
|
|
64
|
+
kind: 'validation',
|
|
65
|
+
code: 'ARTIFACT_FILE_INVALID_SHAPE',
|
|
66
|
+
});
|
|
67
|
+
}
|
|
68
|
+
return 'trace' in record ? record : { trace: record };
|
|
69
|
+
}
|
|
70
|
+
async function readArtifactFile(path) {
|
|
71
|
+
const content = await readBounded(path, MAX_ARTIFACT_CHARS);
|
|
72
|
+
try {
|
|
73
|
+
return JSON.parse(content);
|
|
74
|
+
}
|
|
75
|
+
catch (error) {
|
|
76
|
+
throw new AutoevalError(`${path} is not valid JSON.`, {
|
|
77
|
+
kind: 'validation',
|
|
78
|
+
code: 'ARTIFACT_FILE_INVALID_JSON',
|
|
79
|
+
cause: error,
|
|
80
|
+
});
|
|
81
|
+
}
|
|
82
|
+
}
|
|
83
|
+
async function readReferenceDocument(path) {
|
|
84
|
+
const content = await readBounded(path, MAX_REFERENCE_DOCUMENT_CHARS);
|
|
85
|
+
if (content.trim() === '') {
|
|
86
|
+
throw new AutoevalError(`${path} is empty.`, {
|
|
87
|
+
kind: 'validation',
|
|
88
|
+
code: 'REFERENCE_DOCUMENT_EMPTY',
|
|
89
|
+
});
|
|
90
|
+
}
|
|
91
|
+
return { filename: basename(path), content };
|
|
92
|
+
}
|
|
93
|
+
/**
|
|
94
|
+
* Expands `artifactFile` and `referenceDocumentFiles` in an already-parsed eval
|
|
95
|
+
* file. `baseDirectory` is the directory holding the eval file.
|
|
96
|
+
*/
|
|
97
|
+
export async function resolveEvalFileInputs(parsedJson, baseDirectory) {
|
|
98
|
+
const root = asRecord(parsedJson);
|
|
99
|
+
const configuration = root === undefined ? undefined : asRecord(root.configuration);
|
|
100
|
+
if (root === undefined || configuration === undefined)
|
|
101
|
+
return parsedJson;
|
|
102
|
+
const artifactFile = configuration.artifactFile;
|
|
103
|
+
const referenceDocumentFiles = configuration.referenceDocumentFiles;
|
|
104
|
+
if (artifactFile === undefined && referenceDocumentFiles === undefined)
|
|
105
|
+
return parsedJson;
|
|
106
|
+
const resolved = { ...configuration };
|
|
107
|
+
delete resolved.artifactFile;
|
|
108
|
+
delete resolved.referenceDocumentFiles;
|
|
109
|
+
if (artifactFile !== undefined) {
|
|
110
|
+
if (typeof artifactFile !== 'string' || artifactFile.trim() === '') {
|
|
111
|
+
throw new AutoevalError('configuration.artifactFile must be a file path.', {
|
|
112
|
+
kind: 'validation',
|
|
113
|
+
code: 'ARTIFACT_FILE_INVALID_PATH',
|
|
114
|
+
});
|
|
115
|
+
}
|
|
116
|
+
if (configuration.artifact !== undefined) {
|
|
117
|
+
throw new AutoevalError('configuration has both artifact and artifactFile. Keep one of them.', { kind: 'validation', code: 'ARTIFACT_FILE_CONFLICT' });
|
|
118
|
+
}
|
|
119
|
+
const path = resolveAgainst(baseDirectory, artifactFile);
|
|
120
|
+
resolved.artifact = normalizeArtifact(configuration.contextType, await readArtifactFile(path), path);
|
|
121
|
+
}
|
|
122
|
+
if (referenceDocumentFiles !== undefined) {
|
|
123
|
+
if (!Array.isArray(referenceDocumentFiles) ||
|
|
124
|
+
referenceDocumentFiles.some((entry) => typeof entry !== 'string' || entry.trim() === '')) {
|
|
125
|
+
throw new AutoevalError('configuration.referenceDocumentFiles must be an array of file paths.', { kind: 'validation', code: 'REFERENCE_DOCUMENT_INVALID_PATH' });
|
|
126
|
+
}
|
|
127
|
+
const existing = Array.isArray(configuration.referenceDocuments)
|
|
128
|
+
? configuration.referenceDocuments
|
|
129
|
+
: [];
|
|
130
|
+
const documents = [...existing];
|
|
131
|
+
for (const entry of referenceDocumentFiles) {
|
|
132
|
+
documents.push(await readReferenceDocument(resolveAgainst(baseDirectory, entry)));
|
|
133
|
+
}
|
|
134
|
+
resolved.referenceDocuments = documents;
|
|
135
|
+
}
|
|
136
|
+
return { ...root, configuration: resolved };
|
|
137
|
+
}
|
|
138
|
+
/** Reads an eval file from disk and expands any file references it declares. */
|
|
139
|
+
export async function readEvalFile(path) {
|
|
140
|
+
let content;
|
|
141
|
+
try {
|
|
142
|
+
content = await readFile(path, 'utf8');
|
|
143
|
+
}
|
|
144
|
+
catch (error) {
|
|
145
|
+
throw readFailure(path, error);
|
|
146
|
+
}
|
|
147
|
+
let parsed;
|
|
148
|
+
try {
|
|
149
|
+
parsed = JSON.parse(content);
|
|
150
|
+
}
|
|
151
|
+
catch (error) {
|
|
152
|
+
throw new AutoevalError(`${path} must contain valid JSON.`, {
|
|
153
|
+
kind: 'usage',
|
|
154
|
+
code: 'INVALID_CONFIGURED_RUN_JSON',
|
|
155
|
+
cause: error,
|
|
156
|
+
});
|
|
157
|
+
}
|
|
158
|
+
return resolveEvalFileInputs(parsed, dirname(resolve(path)));
|
|
159
|
+
}
|
|
160
|
+
//# sourceMappingURL=file-inputs.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"file-inputs.js","sourceRoot":"","sources":["../../src/configured-run/file-inputs.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEnE,OAAO,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAC;AAE5D;;;;;;;;;;;;;;;GAeG;AAEH,MAAM,kBAAkB,GAAG,SAAS,CAAC;AACrC,MAAM,4BAA4B,GAAG,OAAO,CAAC;AAI7C,SAAS,WAAW,CAAC,IAAY,EAAE,KAAc;IAC/C,OAAO,IAAI,aAAa,CAAC,4CAA4C,IAAI,EAAE,EAAE;QAC3E,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,uBAAuB;QAC7B,KAAK,EAAE,KAAK;KACb,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,IAAY,EAAE,KAAa;IACpD,IAAI,OAAe,CAAC;IACpB,IAAI,CAAC;QACH,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACzC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACjC,CAAC;IACD,IAAI,OAAO,CAAC,MAAM,GAAG,KAAK,EAAE,CAAC;QAC3B,MAAM,IAAI,aAAa,CAAC,GAAG,IAAI,mBAAmB,KAAK,cAAc,EAAE;YACrE,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,qBAAqB;SAC5B,CAAC,CAAC;IACL,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,cAAc,CAAC,aAAqB,EAAE,IAAY;IACzD,OAAO,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;AAChE,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QACzE,CAAC,CAAE,KAA6B;QAChC,CAAC,CAAC,SAAS,CAAC;AAChB,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,WAAoB,EAAE,QAAiB,EAAE,IAAY;IAC9E,IAAI,WAAW,KAAK,aAAa;QAAE,OAAO,QAAQ,CAAC;IACnD,MAAM,MAAM,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAClC,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,IAAI,aAAa,CAAC,GAAG,IAAI,8BAA8B,EAAE;YAC7D,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,6BAA6B;SACpC,CAAC,CAAC;IACL,CAAC;IACD,OAAO,OAAO,IAAI,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AACxD,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,IAAY;IAC1C,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,IAAI,EAAE,kBAAkB,CAAC,CAAC;IAC5D,IAAI,CAAC;QACH,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAY,CAAC;IACxC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,IAAI,aAAa,CAAC,GAAG,IAAI,qBAAqB,EAAE;YACpD,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,4BAA4B;YAClC,KAAK,EAAE,KAAK;SACb,CAAC,CAAC;IACL,CAAC;AACH,CAAC;AAED,KAAK,UAAU,qBAAqB,CAAC,IAAY;IAC/C,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,IAAI,EAAE,4BAA4B,CAAC,CAAC;IACtE,IAAI,OAAO,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;QAC1B,MAAM,IAAI,aAAa,CAAC,GAAG,IAAI,YAAY,EAAE;YAC3C,IAAI,EAAE,YAAY;YAClB,IAAI,EAAE,0BAA0B;SACjC,CAAC,CAAC;IACL,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,QAAQ,CAAC,IAAI,CAAC,EAAE,OAAO,EAAE,CAAC;AAC/C,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,UAAmB,EACnB,aAAqB;IAErB,MAAM,IAAI,GAAG,QAAQ,CAAC,UAAU,CAAC,CAAC;IAClC,MAAM,aAAa,GAAG,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IACpF,IAAI,IAAI,KAAK,SAAS,IAAI,aAAa,KAAK,SAAS;QAAE,OAAO,UAAU,CAAC;IAEzE,MAAM,YAAY,GAAG,aAAa,CAAC,YAAY,CAAC;IAChD,MAAM,sBAAsB,GAAG,aAAa,CAAC,sBAAsB,CAAC;IACpE,IAAI,YAAY,KAAK,SAAS,IAAI,sBAAsB,KAAK,SAAS;QAAE,OAAO,UAAU,CAAC;IAE1F,MAAM,QAAQ,GAAwB,EAAE,GAAG,aAAa,EAAE,CAAC;IAC3D,OAAO,QAAQ,CAAC,YAAY,CAAC;IAC7B,OAAO,QAAQ,CAAC,sBAAsB,CAAC;IAEvC,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;QAC/B,IAAI,OAAO,YAAY,KAAK,QAAQ,IAAI,YAAY,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACnE,MAAM,IAAI,aAAa,CAAC,iDAAiD,EAAE;gBACzE,IAAI,EAAE,YAAY;gBAClB,IAAI,EAAE,4BAA4B;aACnC,CAAC,CAAC;QACL,CAAC;QACD,IAAI,aAAa,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;YACzC,MAAM,IAAI,aAAa,CACrB,qEAAqE,EACrE,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,wBAAwB,EAAE,CACvD,CAAC;QACJ,CAAC;QACD,MAAM,IAAI,GAAG,cAAc,CAAC,aAAa,EAAE,YAAY,CAAC,CAAC;QACzD,QAAQ,CAAC,QAAQ,GAAG,iBAAiB,CACnC,aAAa,CAAC,WAAW,EACzB,MAAM,gBAAgB,CAAC,IAAI,CAAC,EAC5B,IAAI,CACL,CAAC;IACJ,CAAC;IAED,IAAI,sBAAsB,KAAK,SAAS,EAAE,CAAC;QACzC,IACE,CAAC,KAAK,CAAC,OAAO,CAAC,sBAAsB,CAAC;YACtC,sBAAsB,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EACxF,CAAC;YACD,MAAM,IAAI,aAAa,CACrB,sEAAsE,EACtE,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,iCAAiC,EAAE,CAChE,CAAC;QACJ,CAAC;QACD,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,kBAAkB,CAAC;YAC9D,CAAC,CAAE,aAAa,CAAC,kBAAgC;YACjD,CAAC,CAAC,EAAE,CAAC;QACP,MAAM,SAAS,GAAc,CAAC,GAAG,QAAQ,CAAC,CAAC;QAC3C,KAAK,MAAM,KAAK,IAAI,sBAA2C,EAAE,CAAC;YAChE,SAAS,CAAC,IAAI,CAAC,MAAM,qBAAqB,CAAC,cAAc,CAAC,aAAa,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;QACpF,CAAC;QACD,QAAQ,CAAC,kBAAkB,GAAG,SAAS,CAAC;IAC1C,CAAC;IAED,OAAO,EAAE,GAAG,IAAI,EAAE,aAAa,EAAE,QAAQ,EAAE,CAAC;AAC9C,CAAC;AAED,gFAAgF;AAChF,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,IAAY;IAC7C,IAAI,OAAe,CAAC;IACpB,IAAI,CAAC;QACH,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACzC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACjC,CAAC;IACD,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAY,CAAC;IAC1C,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,IAAI,aAAa,CAAC,GAAG,IAAI,2BAA2B,EAAE;YAC1D,IAAI,EAAE,OAAO;YACb,IAAI,EAAE,6BAA6B;YACnC,KAAK,EAAE,KAAK;SACb,CAAC,CAAC;IACL,CAAC;IACD,OAAO,qBAAqB,CAAC,MAAM,EAAE,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAC/D,CAAC"}
|
|
@@ -0,0 +1,103 @@
|
|
|
1
|
+
import { z } from 'zod';
|
|
2
|
+
import type { EvaluationConfigInput } from '../actions/evaluations.js';
|
|
3
|
+
import type { SupportedModel } from '../domain/types.js';
|
|
4
|
+
export declare const configuredRunFileSchema: z.ZodObject<{
|
|
5
|
+
methodology: z.ZodObject<{
|
|
6
|
+
userSystemId: z.ZodOptional<z.ZodString>;
|
|
7
|
+
judgeModel: z.ZodString;
|
|
8
|
+
judgeModelId: z.ZodUUID;
|
|
9
|
+
runsPerScenario: z.ZodOptional<z.ZodNumber>;
|
|
10
|
+
evaluatorInstructions: z.ZodString;
|
|
11
|
+
changeLog: z.ZodOptional<z.ZodString>;
|
|
12
|
+
}, z.core.$strict>;
|
|
13
|
+
configuration: z.ZodUnion<readonly [z.ZodDiscriminatedUnion<[z.ZodObject<{
|
|
14
|
+
evaluationName: z.ZodOptional<z.ZodString>;
|
|
15
|
+
contextName: z.ZodString;
|
|
16
|
+
contextType: z.ZodLiteral<"scenario">;
|
|
17
|
+
artifact: z.ZodRecord<z.ZodString, z.ZodJSONSchema>;
|
|
18
|
+
expected: z.ZodString;
|
|
19
|
+
referenceDocuments: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
20
|
+
filename: z.ZodString;
|
|
21
|
+
content: z.ZodString;
|
|
22
|
+
}, z.core.$strict>>>;
|
|
23
|
+
selectedMetrics: z.ZodArray<z.ZodString>;
|
|
24
|
+
temperatureContext: z.ZodOptional<z.ZodNumber>;
|
|
25
|
+
autoStopEnabled: z.ZodOptional<z.ZodBoolean>;
|
|
26
|
+
scenarioGeneration: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodJSONSchema>>;
|
|
27
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
28
|
+
evaluationName: z.ZodOptional<z.ZodString>;
|
|
29
|
+
contextName: z.ZodString;
|
|
30
|
+
contextType: z.ZodLiteral<"conversation">;
|
|
31
|
+
artifact: z.ZodRecord<z.ZodString, z.ZodJSONSchema>;
|
|
32
|
+
expected: z.ZodString;
|
|
33
|
+
referenceDocuments: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
34
|
+
filename: z.ZodString;
|
|
35
|
+
content: z.ZodString;
|
|
36
|
+
}, z.core.$strict>>>;
|
|
37
|
+
selectedMetrics: z.ZodArray<z.ZodString>;
|
|
38
|
+
temperatureContext: z.ZodOptional<z.ZodNumber>;
|
|
39
|
+
autoStopEnabled: z.ZodOptional<z.ZodBoolean>;
|
|
40
|
+
}, z.core.$strict>, z.ZodObject<{
|
|
41
|
+
evaluationName: z.ZodOptional<z.ZodString>;
|
|
42
|
+
contextName: z.ZodString;
|
|
43
|
+
contextType: z.ZodLiteral<"agent_trace">;
|
|
44
|
+
artifact: z.ZodRecord<z.ZodString, z.ZodJSONSchema>;
|
|
45
|
+
expected: z.ZodString;
|
|
46
|
+
referenceDocuments: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
47
|
+
filename: z.ZodString;
|
|
48
|
+
content: z.ZodString;
|
|
49
|
+
}, z.core.$strict>>>;
|
|
50
|
+
selectedMetrics: z.ZodArray<z.ZodString>;
|
|
51
|
+
temperatureContext: z.ZodOptional<z.ZodNumber>;
|
|
52
|
+
autoStopEnabled: z.ZodOptional<z.ZodBoolean>;
|
|
53
|
+
}, z.core.$strict>], "contextType">, z.ZodObject<{
|
|
54
|
+
evaluationName: z.ZodOptional<z.ZodString>;
|
|
55
|
+
contextName: z.ZodOptional<z.ZodString>;
|
|
56
|
+
primaryModelId: z.ZodUUID;
|
|
57
|
+
comparisonModelIds: z.ZodDefault<z.ZodArray<z.ZodUUID>>;
|
|
58
|
+
promptText: z.ZodString;
|
|
59
|
+
scenarios: z.ZodUnion<readonly [z.ZodArray<z.ZodArray<z.ZodRecord<z.ZodString, z.ZodJSONSchema>>>, z.ZodArray<z.ZodRecord<z.ZodString, z.ZodJSONSchema>>]>;
|
|
60
|
+
referenceDocuments: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
61
|
+
filename: z.ZodString;
|
|
62
|
+
content: z.ZodString;
|
|
63
|
+
}, z.core.$strict>>>;
|
|
64
|
+
selectedMetrics: z.ZodArray<z.ZodString>;
|
|
65
|
+
expected: z.ZodOptional<z.ZodString>;
|
|
66
|
+
temperatureContext: z.ZodOptional<z.ZodNumber>;
|
|
67
|
+
autoStopEnabled: z.ZodOptional<z.ZodBoolean>;
|
|
68
|
+
}, z.core.$strict>]>;
|
|
69
|
+
}, z.core.$strict>;
|
|
70
|
+
export type ConfiguredRunFileInput = z.input<typeof configuredRunFileSchema>;
|
|
71
|
+
export type ParsedConfiguredRunInput = {
|
|
72
|
+
methodology: {
|
|
73
|
+
userSystemId?: string;
|
|
74
|
+
judgeModel: string;
|
|
75
|
+
judgeModelId: string;
|
|
76
|
+
runsPerScenario?: number;
|
|
77
|
+
evaluatorInstructions: string;
|
|
78
|
+
changeLog?: string;
|
|
79
|
+
};
|
|
80
|
+
evaluationName?: string;
|
|
81
|
+
configuration: EvaluationConfigInput;
|
|
82
|
+
};
|
|
83
|
+
export type ConfiguredRunValidationResult = {
|
|
84
|
+
valid: true;
|
|
85
|
+
contextType: EvaluationConfigInput['contextType'];
|
|
86
|
+
artifactCount: number;
|
|
87
|
+
models: {
|
|
88
|
+
judge: SupportedModel;
|
|
89
|
+
primary?: SupportedModel;
|
|
90
|
+
comparisons: SupportedModel[];
|
|
91
|
+
};
|
|
92
|
+
};
|
|
93
|
+
export type ConfiguredRunModelOverrides = {
|
|
94
|
+
judgeModelId?: string;
|
|
95
|
+
primaryModelId?: string;
|
|
96
|
+
};
|
|
97
|
+
export declare function parseConfiguredRunInput(parsedJson: unknown): ParsedConfiguredRunInput;
|
|
98
|
+
export declare function validateConfiguredRunInput(input: ParsedConfiguredRunInput, enabledModels: readonly SupportedModel[]): ConfiguredRunValidationResult;
|
|
99
|
+
/** Applies explicit CLI model UUIDs without mutating the parsed file input. */
|
|
100
|
+
export declare function applyModelOverrides(input: ParsedConfiguredRunInput, overrides: ConfiguredRunModelOverrides): ParsedConfiguredRunInput;
|
|
101
|
+
/** Fails fast when a sanitized public fixture was not given account model UUIDs. */
|
|
102
|
+
export declare function requireModelOverridesForPublicPlaceholders(input: ParsedConfiguredRunInput): void;
|
|
103
|
+
//# sourceMappingURL=validation.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"validation.d.ts","sourceRoot":"","sources":["../../src/configured-run/validation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,2BAA2B,CAAC;AACvE,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AA4EzD,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAiBzB,CAAC;AAEZ,MAAM,MAAM,sBAAsB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AAuC7E,MAAM,MAAM,wBAAwB,GAAG;IACrC,WAAW,EAAE;QACX,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,UAAU,EAAE,MAAM,CAAC;QACnB,YAAY,EAAE,MAAM,CAAC;QACrB,eAAe,CAAC,EAAE,MAAM,CAAC;QACzB,qBAAqB,EAAE,MAAM,CAAC;QAC9B,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB,CAAC;IACF,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,aAAa,EAAE,qBAAqB,CAAC;CACtC,CAAC;AAEF,MAAM,MAAM,6BAA6B,GAAG;IAC1C,KAAK,EAAE,IAAI,CAAC;IACZ,WAAW,EAAE,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAClD,aAAa,EAAE,MAAM,CAAC;IACtB,MAAM,EAAE;QACN,KAAK,EAAE,cAAc,CAAC;QACtB,OAAO,CAAC,EAAE,cAAc,CAAC;QACzB,WAAW,EAAE,cAAc,EAAE,CAAC;KAC/B,CAAC;CACH,CAAC;AAEF,MAAM,MAAM,2BAA2B,GAAG;IACxC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB,CAAC;AAwBF,wBAAgB,uBAAuB,CAAC,UAAU,EAAE,OAAO,GAAG,wBAAwB,CA4ErF;AAuCD,wBAAgB,0BAA0B,CACxC,KAAK,EAAE,wBAAwB,EAC/B,aAAa,EAAE,SAAS,cAAc,EAAE,GACvC,6BAA6B,CAgE/B;AAED,+EAA+E;AAC/E,wBAAgB,mBAAmB,CACjC,KAAK,EAAE,wBAAwB,EAC/B,SAAS,EAAE,2BAA2B,GACrC,wBAAwB,CAqB1B;AAED,oFAAoF;AACpF,wBAAgB,0CAA0C,CAAC,KAAK,EAAE,wBAAwB,GAAG,IAAI,CAiBhG"}
|