@assemblyline-agents/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 +21 -0
- package/README.md +19 -0
- package/dist/add.d.ts +23 -0
- package/dist/add.d.ts.map +1 -0
- package/dist/add.js +352 -0
- package/dist/add.js.map +1 -0
- package/dist/args.d.ts +14 -0
- package/dist/args.d.ts.map +1 -0
- package/dist/args.js +95 -0
- package/dist/args.js.map +1 -0
- package/dist/auth.d.ts +8 -0
- package/dist/auth.d.ts.map +1 -0
- package/dist/auth.js +74 -0
- package/dist/auth.js.map +1 -0
- package/dist/channels.d.ts +25 -0
- package/dist/channels.d.ts.map +1 -0
- package/dist/channels.js +76 -0
- package/dist/channels.js.map +1 -0
- package/dist/checkpoints.d.ts +3 -0
- package/dist/checkpoints.d.ts.map +1 -0
- package/dist/checkpoints.js +60 -0
- package/dist/checkpoints.js.map +1 -0
- package/dist/deploy-definition.d.ts +14 -0
- package/dist/deploy-definition.d.ts.map +1 -0
- package/dist/deploy-definition.js +92 -0
- package/dist/deploy-definition.js.map +1 -0
- package/dist/deploy-lifecycle.d.ts +17 -0
- package/dist/deploy-lifecycle.d.ts.map +1 -0
- package/dist/deploy-lifecycle.js +25 -0
- package/dist/deploy-lifecycle.js.map +1 -0
- package/dist/deploy.d.ts +20 -0
- package/dist/deploy.d.ts.map +1 -0
- package/dist/deploy.js +462 -0
- package/dist/deploy.js.map +1 -0
- package/dist/dev.d.ts +13 -0
- package/dist/dev.d.ts.map +1 -0
- package/dist/dev.js +157 -0
- package/dist/dev.js.map +1 -0
- package/dist/eval-assertions.d.ts +5 -0
- package/dist/eval-assertions.d.ts.map +1 -0
- package/dist/eval-assertions.js +99 -0
- package/dist/eval-assertions.js.map +1 -0
- package/dist/eval-case-runner.d.ts +38 -0
- package/dist/eval-case-runner.d.ts.map +1 -0
- package/dist/eval-case-runner.js +247 -0
- package/dist/eval-case-runner.js.map +1 -0
- package/dist/eval-cases.d.ts +6 -0
- package/dist/eval-cases.d.ts.map +1 -0
- package/dist/eval-cases.js +151 -0
- package/dist/eval-cases.js.map +1 -0
- package/dist/eval-comparison.d.ts +8 -0
- package/dist/eval-comparison.d.ts.map +1 -0
- package/dist/eval-comparison.js +90 -0
- package/dist/eval-comparison.js.map +1 -0
- package/dist/eval-custom.d.ts +15 -0
- package/dist/eval-custom.d.ts.map +1 -0
- package/dist/eval-custom.js +118 -0
- package/dist/eval-custom.js.map +1 -0
- package/dist/eval-experiment.d.ts +19 -0
- package/dist/eval-experiment.d.ts.map +1 -0
- package/dist/eval-experiment.js +122 -0
- package/dist/eval-experiment.js.map +1 -0
- package/dist/eval-fixtures.d.ts +17 -0
- package/dist/eval-fixtures.d.ts.map +1 -0
- package/dist/eval-fixtures.js +59 -0
- package/dist/eval-fixtures.js.map +1 -0
- package/dist/eval-gateway-host.d.ts +34 -0
- package/dist/eval-gateway-host.d.ts.map +1 -0
- package/dist/eval-gateway-host.js +127 -0
- package/dist/eval-gateway-host.js.map +1 -0
- package/dist/eval-judge.d.ts +35 -0
- package/dist/eval-judge.d.ts.map +1 -0
- package/dist/eval-judge.js +105 -0
- package/dist/eval-judge.js.map +1 -0
- package/dist/eval-mocks.d.ts +10 -0
- package/dist/eval-mocks.d.ts.map +1 -0
- package/dist/eval-mocks.js +57 -0
- package/dist/eval-mocks.js.map +1 -0
- package/dist/eval-report.d.ts +7 -0
- package/dist/eval-report.d.ts.map +1 -0
- package/dist/eval-report.js +89 -0
- package/dist/eval-report.js.map +1 -0
- package/dist/eval-retry.d.ts +10 -0
- package/dist/eval-retry.d.ts.map +1 -0
- package/dist/eval-retry.js +37 -0
- package/dist/eval-retry.js.map +1 -0
- package/dist/eval-trajectory.d.ts +7 -0
- package/dist/eval-trajectory.d.ts.map +1 -0
- package/dist/eval-trajectory.js +92 -0
- package/dist/eval-trajectory.js.map +1 -0
- package/dist/eval-turns.d.ts +66 -0
- package/dist/eval-turns.d.ts.map +1 -0
- package/dist/eval-turns.js +187 -0
- package/dist/eval-turns.js.map +1 -0
- package/dist/eval-types.d.ts +271 -0
- package/dist/eval-types.d.ts.map +1 -0
- package/dist/eval-types.js +4 -0
- package/dist/eval-types.js.map +1 -0
- package/dist/eval.d.ts +26 -0
- package/dist/eval.d.ts.map +1 -0
- package/dist/eval.js +218 -0
- package/dist/eval.js.map +1 -0
- package/dist/help.d.ts +12 -0
- package/dist/help.d.ts.map +1 -0
- package/dist/help.js +323 -0
- package/dist/help.js.map +1 -0
- package/dist/hosts.d.ts +3 -0
- package/dist/hosts.d.ts.map +1 -0
- package/dist/hosts.js +51 -0
- package/dist/hosts.js.map +1 -0
- package/dist/index.d.ts +3 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +318 -0
- package/dist/index.js.map +1 -0
- package/dist/init.d.ts +13 -0
- package/dist/init.d.ts.map +1 -0
- package/dist/init.js +79 -0
- package/dist/init.js.map +1 -0
- package/dist/legacy-bin.d.ts +3 -0
- package/dist/legacy-bin.d.ts.map +1 -0
- package/dist/legacy-bin.js +8 -0
- package/dist/legacy-bin.js.map +1 -0
- package/dist/models.d.ts +5 -0
- package/dist/models.d.ts.map +1 -0
- package/dist/models.js +20 -0
- package/dist/models.js.map +1 -0
- package/dist/operator-client.d.ts +7 -0
- package/dist/operator-client.d.ts.map +1 -0
- package/dist/operator-client.js +62 -0
- package/dist/operator-client.js.map +1 -0
- package/dist/pm.d.ts +20 -0
- package/dist/pm.d.ts.map +1 -0
- package/dist/pm.js +70 -0
- package/dist/pm.js.map +1 -0
- package/dist/project-env.d.ts +8 -0
- package/dist/project-env.d.ts.map +1 -0
- package/dist/project-env.js +51 -0
- package/dist/project-env.js.map +1 -0
- package/dist/secrets.d.ts +3 -0
- package/dist/secrets.d.ts.map +1 -0
- package/dist/secrets.js +48 -0
- package/dist/secrets.js.map +1 -0
- package/dist/state.d.ts +3 -0
- package/dist/state.d.ts.map +1 -0
- package/dist/state.js +85 -0
- package/dist/state.js.map +1 -0
- package/package.json +67 -0
|
@@ -0,0 +1,271 @@
|
|
|
1
|
+
import type { JsonObject, JsonValue } from "@assemblyline-agents/core";
|
|
2
|
+
import type { RunStatus, RunTimeline } from "@assemblyline-agents/runtime";
|
|
3
|
+
export type EvalOutcome = "passed" | "failed" | "errored";
|
|
4
|
+
export type EvalTrajectoryMode = "contains" | "ordered" | "exact" | "forbid";
|
|
5
|
+
export type EvalValueMatchMode = "partial" | "exact";
|
|
6
|
+
export interface EvalTrajectoryStep {
|
|
7
|
+
tool: string;
|
|
8
|
+
arguments?: unknown;
|
|
9
|
+
argumentsMatch?: EvalValueMatchMode;
|
|
10
|
+
status?: string;
|
|
11
|
+
result?: unknown;
|
|
12
|
+
resultMatch?: EvalValueMatchMode;
|
|
13
|
+
}
|
|
14
|
+
export interface EvalCustomEvaluatorExpectation {
|
|
15
|
+
name: string;
|
|
16
|
+
config?: JsonObject;
|
|
17
|
+
/** Metric key used by minScore when the evaluator returns several metrics. */
|
|
18
|
+
metric?: string;
|
|
19
|
+
/** Require at least one numeric metric at or above this value. */
|
|
20
|
+
minScore?: number;
|
|
21
|
+
/** Defaults to true; fail when any returned metric explicitly has pass=false. */
|
|
22
|
+
requirePass?: boolean;
|
|
23
|
+
}
|
|
24
|
+
export interface EvalOutputExpectation {
|
|
25
|
+
contains?: string[];
|
|
26
|
+
matches?: string;
|
|
27
|
+
equals?: string;
|
|
28
|
+
json?: {
|
|
29
|
+
valid?: boolean;
|
|
30
|
+
jsonSchema?: unknown;
|
|
31
|
+
fields?: Record<string, unknown>;
|
|
32
|
+
};
|
|
33
|
+
}
|
|
34
|
+
/** Assertions that need no model call; usable per case and per conversation turn. */
|
|
35
|
+
export interface EvalDeterministicExpectations {
|
|
36
|
+
status?: RunStatus;
|
|
37
|
+
output?: EvalOutputExpectation;
|
|
38
|
+
toolsCalled?: string[];
|
|
39
|
+
toolsNotCalled?: string[];
|
|
40
|
+
trajectory?: {
|
|
41
|
+
mode: EvalTrajectoryMode;
|
|
42
|
+
steps: EvalTrajectoryStep[];
|
|
43
|
+
};
|
|
44
|
+
maxCostUsd?: number;
|
|
45
|
+
}
|
|
46
|
+
export interface EvalTurn {
|
|
47
|
+
message: string;
|
|
48
|
+
approve?: boolean;
|
|
49
|
+
tool?: string;
|
|
50
|
+
toolInput?: unknown;
|
|
51
|
+
/** Deterministic assertions against this turn alone; failures are prefixed `turn N:`. */
|
|
52
|
+
expect?: EvalDeterministicExpectations;
|
|
53
|
+
}
|
|
54
|
+
/** Exactly one of `result`, `results` (consumed per call, last repeats), or `error`. */
|
|
55
|
+
export interface EvalToolMock {
|
|
56
|
+
result?: JsonValue;
|
|
57
|
+
results?: JsonValue[];
|
|
58
|
+
error?: string;
|
|
59
|
+
}
|
|
60
|
+
export interface EvalMocks {
|
|
61
|
+
tools: Record<string, EvalToolMock>;
|
|
62
|
+
}
|
|
63
|
+
export interface EvalFixtures {
|
|
64
|
+
/** Sandbox files seeded before the first turn; relative paths land under /workspace. */
|
|
65
|
+
sandbox?: Record<string, string>;
|
|
66
|
+
/** Memory documents (path starting with /memory/) seeded at agent scope. */
|
|
67
|
+
memory?: Record<string, string>;
|
|
68
|
+
}
|
|
69
|
+
export interface EvalCase {
|
|
70
|
+
name: string;
|
|
71
|
+
description?: string;
|
|
72
|
+
tags?: string[];
|
|
73
|
+
/** Attempts for this case (1..100); wins over the --repetitions suite flag. */
|
|
74
|
+
repetitions?: number;
|
|
75
|
+
input: {
|
|
76
|
+
/** Single-turn message; mutually exclusive with `turns`. */
|
|
77
|
+
message?: string;
|
|
78
|
+
/** Scripted conversation; mutually exclusive with `message`. */
|
|
79
|
+
turns?: EvalTurn[];
|
|
80
|
+
channel?: string;
|
|
81
|
+
tool?: string;
|
|
82
|
+
toolInput?: unknown;
|
|
83
|
+
approve?: boolean;
|
|
84
|
+
recentHistory?: string[];
|
|
85
|
+
};
|
|
86
|
+
mocks?: EvalMocks;
|
|
87
|
+
fixtures?: EvalFixtures;
|
|
88
|
+
expect: EvalDeterministicExpectations & {
|
|
89
|
+
evaluators?: EvalCustomEvaluatorExpectation[];
|
|
90
|
+
judge?: {
|
|
91
|
+
criteria: string;
|
|
92
|
+
model?: string;
|
|
93
|
+
threshold?: number;
|
|
94
|
+
};
|
|
95
|
+
};
|
|
96
|
+
}
|
|
97
|
+
export interface DiscoveredEvalCase {
|
|
98
|
+
path: string;
|
|
99
|
+
name: string;
|
|
100
|
+
evalCase?: EvalCase;
|
|
101
|
+
error?: string;
|
|
102
|
+
}
|
|
103
|
+
export interface EvalTurnRecord {
|
|
104
|
+
message: string;
|
|
105
|
+
status: RunStatus;
|
|
106
|
+
response: string;
|
|
107
|
+
output?: JsonValue;
|
|
108
|
+
toolsCalled: string[];
|
|
109
|
+
costUsd: number;
|
|
110
|
+
runId: string;
|
|
111
|
+
}
|
|
112
|
+
export interface EvalActual {
|
|
113
|
+
status: RunStatus;
|
|
114
|
+
response: string;
|
|
115
|
+
output?: JsonValue;
|
|
116
|
+
toolsCalled: string[];
|
|
117
|
+
toolCalls?: RunTimeline["toolCalls"];
|
|
118
|
+
/** False when any model transaction lacks a provider-reported/reconciled cash receipt. */
|
|
119
|
+
costComplete: boolean;
|
|
120
|
+
costUsd: number;
|
|
121
|
+
/** Per-turn records for conversation cases; status/response/output reflect the final turn. */
|
|
122
|
+
turns?: EvalTurnRecord[];
|
|
123
|
+
}
|
|
124
|
+
export interface EvalMetric {
|
|
125
|
+
key: string;
|
|
126
|
+
score?: number;
|
|
127
|
+
value?: JsonValue;
|
|
128
|
+
pass?: boolean;
|
|
129
|
+
comment?: string;
|
|
130
|
+
evaluator?: string;
|
|
131
|
+
}
|
|
132
|
+
export type EvalModelComplete = (input: {
|
|
133
|
+
prompt: string;
|
|
134
|
+
systemPrompt?: string;
|
|
135
|
+
model?: string;
|
|
136
|
+
}) => Promise<{
|
|
137
|
+
text: string;
|
|
138
|
+
costUsd: number;
|
|
139
|
+
}>;
|
|
140
|
+
export interface EvalEvaluatorContext {
|
|
141
|
+
evalCase: EvalCase;
|
|
142
|
+
actual: EvalActual;
|
|
143
|
+
/** Final-turn timeline; `timelines` has one entry per conversation turn. */
|
|
144
|
+
timeline: RunTimeline;
|
|
145
|
+
timelines: RunTimeline[];
|
|
146
|
+
config: JsonObject;
|
|
147
|
+
repetition: number;
|
|
148
|
+
attempt: number;
|
|
149
|
+
/**
|
|
150
|
+
* Model completer bound to the suite judge configuration (defaults to the
|
|
151
|
+
* judge model). Cost is accounted into the case's judgeCostUsd.
|
|
152
|
+
*/
|
|
153
|
+
complete: EvalModelComplete;
|
|
154
|
+
}
|
|
155
|
+
export interface EvalEvaluator {
|
|
156
|
+
name: string;
|
|
157
|
+
description?: string;
|
|
158
|
+
evaluate(context: EvalEvaluatorContext): EvalMetric | EvalMetric[] | Promise<EvalMetric | EvalMetric[]>;
|
|
159
|
+
}
|
|
160
|
+
export declare function defineEvaluator<T extends EvalEvaluator>(evaluator: T): T;
|
|
161
|
+
export interface EvalJudgeResult {
|
|
162
|
+
score: number;
|
|
163
|
+
pass: boolean;
|
|
164
|
+
reasoning: string;
|
|
165
|
+
costUsd: number;
|
|
166
|
+
model: string;
|
|
167
|
+
}
|
|
168
|
+
export interface EvalCaseResult {
|
|
169
|
+
caseId: string;
|
|
170
|
+
name: string;
|
|
171
|
+
path: string;
|
|
172
|
+
tags: string[];
|
|
173
|
+
outcome: EvalOutcome;
|
|
174
|
+
repetition: number;
|
|
175
|
+
/** Planned repetitions for this case (case field or suite flag) — the pass-distribution denominator. */
|
|
176
|
+
repetitionsPlanned?: number;
|
|
177
|
+
attempts: number;
|
|
178
|
+
durationMs: number;
|
|
179
|
+
runCostUsd: number;
|
|
180
|
+
judgeCostUsd: number;
|
|
181
|
+
failures: string[];
|
|
182
|
+
error?: string;
|
|
183
|
+
runId?: string;
|
|
184
|
+
status?: RunStatus;
|
|
185
|
+
response?: string;
|
|
186
|
+
output?: JsonValue;
|
|
187
|
+
turns?: EvalTurnRecord[];
|
|
188
|
+
judge?: EvalJudgeResult;
|
|
189
|
+
metrics?: EvalMetric[];
|
|
190
|
+
}
|
|
191
|
+
export interface EvalExecutionConfig {
|
|
192
|
+
filter?: string;
|
|
193
|
+
tag?: string;
|
|
194
|
+
judgeModel: string;
|
|
195
|
+
concurrency: number;
|
|
196
|
+
timeoutMs: number;
|
|
197
|
+
approve: boolean;
|
|
198
|
+
failFast: boolean;
|
|
199
|
+
repetitions: number;
|
|
200
|
+
retryAttempts: number;
|
|
201
|
+
retryBackoffMs: number;
|
|
202
|
+
}
|
|
203
|
+
export interface EvalExperimentManifest {
|
|
204
|
+
schemaVersion: 1;
|
|
205
|
+
id: string;
|
|
206
|
+
startedAt: string;
|
|
207
|
+
frameworkVersion: string;
|
|
208
|
+
agentRevision: string;
|
|
209
|
+
suiteFingerprint: string;
|
|
210
|
+
configFingerprint: string;
|
|
211
|
+
caseCount: number;
|
|
212
|
+
executionCount: number;
|
|
213
|
+
config: EvalExecutionConfig;
|
|
214
|
+
}
|
|
215
|
+
export interface EvalRegression {
|
|
216
|
+
caseId: string;
|
|
217
|
+
name: string;
|
|
218
|
+
baselinePassRate: number;
|
|
219
|
+
currentPassRate: number;
|
|
220
|
+
}
|
|
221
|
+
export interface EvalComparison {
|
|
222
|
+
baselineExperimentId: string;
|
|
223
|
+
regressions: EvalRegression[];
|
|
224
|
+
improvements: EvalRegression[];
|
|
225
|
+
addedCaseIds: string[];
|
|
226
|
+
removedCaseIds: string[];
|
|
227
|
+
passRateDelta: number;
|
|
228
|
+
costDeltaUsd: number;
|
|
229
|
+
costDeltaPercent?: number;
|
|
230
|
+
/** actual=null represents an unbounded percentage (zero-cost baseline to positive cost). */
|
|
231
|
+
gates: Array<{
|
|
232
|
+
name: string;
|
|
233
|
+
passed: boolean;
|
|
234
|
+
actual: number | null;
|
|
235
|
+
limit: number;
|
|
236
|
+
}>;
|
|
237
|
+
passed: boolean;
|
|
238
|
+
}
|
|
239
|
+
export interface EvalReport {
|
|
240
|
+
schemaVersion: 1;
|
|
241
|
+
experiment: EvalExperimentManifest;
|
|
242
|
+
startedAt: string;
|
|
243
|
+
durationMs: number;
|
|
244
|
+
agentRevision: string;
|
|
245
|
+
results: EvalCaseResult[];
|
|
246
|
+
comparison?: EvalComparison;
|
|
247
|
+
artifacts?: {
|
|
248
|
+
root: string;
|
|
249
|
+
experimentPath: string;
|
|
250
|
+
resultsPath: string;
|
|
251
|
+
};
|
|
252
|
+
summary: {
|
|
253
|
+
total: number;
|
|
254
|
+
uniqueCases: number;
|
|
255
|
+
repetitions: number;
|
|
256
|
+
passed: number;
|
|
257
|
+
failed: number;
|
|
258
|
+
errored: number;
|
|
259
|
+
passRate: number;
|
|
260
|
+
runCostUsd: number;
|
|
261
|
+
judgeCostUsd: number;
|
|
262
|
+
totalCostUsd: number;
|
|
263
|
+
byTag: Record<string, {
|
|
264
|
+
total: number;
|
|
265
|
+
passed: number;
|
|
266
|
+
failed: number;
|
|
267
|
+
errored: number;
|
|
268
|
+
}>;
|
|
269
|
+
};
|
|
270
|
+
}
|
|
271
|
+
//# sourceMappingURL=eval-types.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-types.d.ts","sourceRoot":"","sources":["../src/eval-types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,2BAA2B,CAAC;AACvE,OAAO,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,8BAA8B,CAAC;AAE3E,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,QAAQ,GAAG,SAAS,CAAC;AAC1D,MAAM,MAAM,kBAAkB,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,GAAG,QAAQ,CAAC;AAC7E,MAAM,MAAM,kBAAkB,GAAG,SAAS,GAAG,OAAO,CAAC;AAErD,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,cAAc,CAAC,EAAE,kBAAkB,CAAC;IACpC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,WAAW,CAAC,EAAE,kBAAkB,CAAC;CAClC;AAED,MAAM,WAAW,8BAA8B;IAC7C,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,UAAU,CAAC;IACpB,8EAA8E;IAC9E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,kEAAkE;IAClE,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iFAAiF;IACjF,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE;QACL,KAAK,CAAC,EAAE,OAAO,CAAC;QAChB,UAAU,CAAC,EAAE,OAAO,CAAC;QACrB,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KAClC,CAAC;CACH;AAED,qFAAqF;AACrF,MAAM,WAAW,6BAA6B;IAC5C,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,MAAM,CAAC,EAAE,qBAAqB,CAAC;IAC/B,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,UAAU,CAAC,EAAE;QAAE,IAAI,EAAE,kBAAkB,CAAC;QAAC,KAAK,EAAE,kBAAkB,EAAE,CAAA;KAAE,CAAC;IACvE,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,yFAAyF;IACzF,MAAM,CAAC,EAAE,6BAA6B,CAAC;CACxC;AAED,wFAAwF;AACxF,MAAM,WAAW,YAAY;IAC3B,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,OAAO,CAAC,EAAE,SAAS,EAAE,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;CACrC;AAED,MAAM,WAAW,YAAY;IAC3B,wFAAwF;IACxF,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,4EAA4E;IAC5E,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjC;AAED,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,+EAA+E;IAC/E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE;QACL,4DAA4D;QAC5D,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,gEAAgE;QAChE,KAAK,CAAC,EAAE,QAAQ,EAAE,CAAC;QACnB,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,SAAS,CAAC,EAAE,OAAO,CAAC;QACpB,OAAO,CAAC,EAAE,OAAO,CAAC;QAClB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;KAC1B,CAAC;IACF,KAAK,CAAC,EAAE,SAAS,CAAC;IAClB,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,MAAM,EAAE,6BAA6B,GAAG;QACtC,UAAU,CAAC,EAAE,8BAA8B,EAAE,CAAC;QAC9C,KAAK,CAAC,EAAE;YAAE,QAAQ,EAAE,MAAM,CAAC;YAAC,KAAK,CAAC,EAAE,MAAM,CAAC;YAAC,SAAS,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC;KAClE,CAAC;CACH;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,SAAS,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,SAAS,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,SAAS,CAAC,EAAE,WAAW,CAAC,WAAW,CAAC,CAAC;IACrC,0FAA0F;IAC1F,YAAY,EAAE,OAAO,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;IAChB,8FAA8F;IAC9F,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,UAAU;IACzB,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,SAAS,CAAC;IAClB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,iBAAiB,GAAG,CAAC,KAAK,EAAE;IACtC,MAAM,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB,KAAK,OAAO,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CAAC;AAEjD,MAAM,WAAW,oBAAoB;IACnC,QAAQ,EAAE,QAAQ,CAAC;IACnB,MAAM,EAAE,UAAU,CAAC;IACnB,4EAA4E;IAC5E,QAAQ,EAAE,WAAW,CAAC;IACtB,SAAS,EAAE,WAAW,EAAE,CAAC;IACzB,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB;;;OAGG;IACH,QAAQ,EAAE,iBAAiB,CAAC;CAC7B;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,OAAO,EAAE,oBAAoB,GAAG,UAAU,GAAG,UAAU,EAAE,GAAG,OAAO,CAAC,UAAU,GAAG,UAAU,EAAE,CAAC,CAAC;CACzG;AAED,wBAAgB,eAAe,CAAC,CAAC,SAAS,aAAa,EAAE,SAAS,EAAE,CAAC,GAAG,CAAC,CAExE;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,OAAO,EAAE,WAAW,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,wGAAwG;IACxG,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,OAAO,CAAC,EAAE,UAAU,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,mBAAmB;IAClC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,OAAO,CAAC;IACjB,QAAQ,EAAE,OAAO,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,sBAAsB;IACrC,aAAa,EAAE,CAAC,CAAC;IACjB,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;IAClB,gBAAgB,EAAE,MAAM,CAAC;IACzB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;IACzB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,EAAE,mBAAmB,CAAC;CAC7B;AAED,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,cAAc;IAC7B,oBAAoB,EAAE,MAAM,CAAC;IAC7B,WAAW,EAAE,cAAc,EAAE,CAAC;IAC9B,YAAY,EAAE,cAAc,EAAE,CAAC;IAC/B,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,4FAA4F;IAC5F,KAAK,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACtF,MAAM,EAAE,OAAO,CAAC;CACjB;AAED,MAAM,WAAW,UAAU;IACzB,aAAa,EAAE,CAAC,CAAC;IACjB,UAAU,EAAE,sBAAsB,CAAC;IACnC,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE,cAAc,EAAE,CAAC;IAC1B,UAAU,CAAC,EAAE,cAAc,CAAC;IAC5B,SAAS,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC;IAC1E,OAAO,EAAE;QACP,KAAK,EAAE,MAAM,CAAC;QACd,WAAW,EAAE,MAAM,CAAC;QACpB,WAAW,EAAE,MAAM,CAAC;QACpB,MAAM,EAAE,MAAM,CAAC;QACf,MAAM,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,UAAU,EAAE,MAAM,CAAC;QACnB,YAAY,EAAE,MAAM,CAAC;QACrB,YAAY,EAAE,MAAM,CAAC;QACrB,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,MAAM,EAAE,MAAM,CAAC;YAAC,MAAM,EAAE,MAAM,CAAC;YAAC,OAAO,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;KAC3F,CAAC;CACH"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-types.js","sourceRoot":"","sources":["../src/eval-types.ts"],"names":[],"mappings":"AAuKA,MAAM,UAAU,eAAe,CAA0B,SAAY;IACnE,OAAO,SAAS,CAAC;AACnB,CAAC"}
|
package/dist/eval.d.ts
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
1
|
+
import { type ParsedArgs } from "./args.js";
|
|
2
|
+
import { type EvalCaseRuntimeDependencies } from "./eval-case-runner.js";
|
|
3
|
+
import type { EvalEvaluator, EvalReport } from "./eval-types.js";
|
|
4
|
+
export { evaluateCase, evaluateExpectations } from "./eval-assertions.js";
|
|
5
|
+
export { discoverCases } from "./eval-cases.js";
|
|
6
|
+
export { compareEvalReports, loadBaselineReport } from "./eval-comparison.js";
|
|
7
|
+
export { EvalEvaluatorLoader, evaluateCustomExpectations } from "./eval-custom.js";
|
|
8
|
+
export { seedEvalFixtures, validateFixtures } from "./eval-fixtures.js";
|
|
9
|
+
export { createToolStubs, validateMocks } from "./eval-mocks.js";
|
|
10
|
+
export { conversationIdForCase, EVAL_CONVERSATION_ID, normalizeTurns, runConversation, type EvalRuntimeHost } from "./eval-turns.js";
|
|
11
|
+
export { createGatewayEvalHost, remoteEvalIncompatibilities } from "./eval-gateway-host.js";
|
|
12
|
+
export { createExperimentArtifacts, createExperimentManifest, fingerprintEvalSuite, fingerprintValue, writeExperimentManifest, writeExperimentReport } from "./eval-experiment.js";
|
|
13
|
+
export { createDefaultJudgeCompleter, EvalJudgeResponseError, judgeCase, robustParseJudge } from "./eval-judge.js";
|
|
14
|
+
export { createEvalReport, formatEvalReport } from "./eval-report.js";
|
|
15
|
+
export { isTransientProviderError, retryTransient } from "./eval-retry.js";
|
|
16
|
+
export { evaluateTrajectory } from "./eval-trajectory.js";
|
|
17
|
+
export { defineEvaluator } from "./eval-types.js";
|
|
18
|
+
export type * from "./eval-types.js";
|
|
19
|
+
export interface EvalCommandDependencies extends EvalCaseRuntimeDependencies {
|
|
20
|
+
evaluators?: Record<string, EvalEvaluator>;
|
|
21
|
+
env?: Record<string, string | undefined>;
|
|
22
|
+
now?: () => Date;
|
|
23
|
+
log?: (line: string) => void;
|
|
24
|
+
}
|
|
25
|
+
export declare function runEvalCommand(agentRoot: string, parsed: ParsedArgs, dependencies?: EvalCommandDependencies): Promise<EvalReport>;
|
|
26
|
+
//# sourceMappingURL=eval.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../src/eval.ts"],"names":[],"mappings":"AAEA,OAAO,EAA2B,KAAK,UAAU,EAAE,MAAM,WAAW,CAAC;AACrE,OAAO,EAA0B,KAAK,2BAA2B,EAAE,MAAM,uBAAuB,CAAC;AAejG,OAAO,KAAK,EAGV,aAAa,EAEb,UAAU,EACX,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EAAE,YAAY,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAC1E,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC9E,OAAO,EAAE,mBAAmB,EAAE,0BAA0B,EAAE,MAAM,kBAAkB,CAAC;AACnF,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AACxE,OAAO,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,qBAAqB,EAAE,oBAAoB,EAAE,cAAc,EAAE,eAAe,EAAE,KAAK,eAAe,EAAE,MAAM,iBAAiB,CAAC;AACrI,OAAO,EAAE,qBAAqB,EAAE,2BAA2B,EAAE,MAAM,wBAAwB,CAAC;AAC5F,OAAO,EACL,yBAAyB,EACzB,wBAAwB,EACxB,oBAAoB,EACpB,gBAAgB,EAChB,uBAAuB,EACvB,qBAAqB,EACtB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,2BAA2B,EAAE,sBAAsB,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnH,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,wBAAwB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAC3E,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAClD,mBAAmB,iBAAiB,CAAC;AAErC,MAAM,WAAW,uBAAwB,SAAQ,2BAA2B;IAC1E,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC;IAC3C,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,SAAS,CAAC,CAAC;IACzC,GAAG,CAAC,EAAE,MAAM,IAAI,CAAC;IACjB,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;CAC9B;AAQD,wBAAsB,cAAc,CAClC,SAAS,EAAE,MAAM,EACjB,MAAM,EAAE,UAAU,EAClB,YAAY,GAAE,uBAA4B,GACzC,OAAO,CAAC,UAAU,CAAC,CAuHrB"}
|
package/dist/eval.js
ADDED
|
@@ -0,0 +1,218 @@
|
|
|
1
|
+
import { relative, resolve } from "node:path";
|
|
2
|
+
import { buildAgent } from "@assemblyline-agents/compiler";
|
|
3
|
+
import { booleanFlag, stringFlag } from "./args.js";
|
|
4
|
+
import { runEvalCaseWithRetries } from "./eval-case-runner.js";
|
|
5
|
+
import { discoverCases } from "./eval-cases.js";
|
|
6
|
+
import { createGatewayEvalHost, remoteEvalIncompatibilities } from "./eval-gateway-host.js";
|
|
7
|
+
import { compareEvalReports, loadBaselineReport } from "./eval-comparison.js";
|
|
8
|
+
import { EvalEvaluatorLoader } from "./eval-custom.js";
|
|
9
|
+
import { createExperimentArtifacts, createExperimentManifest, writeExperimentManifest, writeExperimentReport } from "./eval-experiment.js";
|
|
10
|
+
import { createDefaultJudgeCompleter } from "./eval-judge.js";
|
|
11
|
+
import { createEvalReport, formatEvalReport } from "./eval-report.js";
|
|
12
|
+
export { evaluateCase, evaluateExpectations } from "./eval-assertions.js";
|
|
13
|
+
export { discoverCases } from "./eval-cases.js";
|
|
14
|
+
export { compareEvalReports, loadBaselineReport } from "./eval-comparison.js";
|
|
15
|
+
export { EvalEvaluatorLoader, evaluateCustomExpectations } from "./eval-custom.js";
|
|
16
|
+
export { seedEvalFixtures, validateFixtures } from "./eval-fixtures.js";
|
|
17
|
+
export { createToolStubs, validateMocks } from "./eval-mocks.js";
|
|
18
|
+
export { conversationIdForCase, EVAL_CONVERSATION_ID, normalizeTurns, runConversation } from "./eval-turns.js";
|
|
19
|
+
export { createGatewayEvalHost, remoteEvalIncompatibilities } from "./eval-gateway-host.js";
|
|
20
|
+
export { createExperimentArtifacts, createExperimentManifest, fingerprintEvalSuite, fingerprintValue, writeExperimentManifest, writeExperimentReport } from "./eval-experiment.js";
|
|
21
|
+
export { createDefaultJudgeCompleter, EvalJudgeResponseError, judgeCase, robustParseJudge } from "./eval-judge.js";
|
|
22
|
+
export { createEvalReport, formatEvalReport } from "./eval-report.js";
|
|
23
|
+
export { isTransientProviderError, retryTransient } from "./eval-retry.js";
|
|
24
|
+
export { evaluateTrajectory } from "./eval-trajectory.js";
|
|
25
|
+
export { defineEvaluator } from "./eval-types.js";
|
|
26
|
+
export async function runEvalCommand(agentRoot, parsed, dependencies = {}) {
|
|
27
|
+
const env = dependencies.env ?? process.env;
|
|
28
|
+
const now = dependencies.now ?? (() => new Date());
|
|
29
|
+
const started = now();
|
|
30
|
+
const filter = stringFlag(parsed, "filter");
|
|
31
|
+
const tag = stringFlag(parsed, "tag");
|
|
32
|
+
const discovered = discoverCases(agentRoot, {
|
|
33
|
+
...(filter ? { filter } : {}),
|
|
34
|
+
...(tag ? { tag } : {})
|
|
35
|
+
});
|
|
36
|
+
const concurrency = positiveIntegerFlag(parsed, "concurrency", 1);
|
|
37
|
+
const timeoutMs = positiveIntegerFlag(parsed, "timeout", 120) * 1_000;
|
|
38
|
+
const repetitions = positiveIntegerFlag(parsed, "repetitions", 1);
|
|
39
|
+
const retryAttempts = nonnegativeIntegerFlag(parsed, "retry-attempts", 2);
|
|
40
|
+
const retryBackoffMs = nonnegativeIntegerFlag(parsed, "retry-backoff-ms", 250);
|
|
41
|
+
const approve = booleanFlag(parsed, "approve");
|
|
42
|
+
const failFast = booleanFlag(parsed, "fail-fast");
|
|
43
|
+
const outDir = stringFlag(parsed, "out");
|
|
44
|
+
const artifact = buildAgent({ root: agentRoot, ...(outDir ? { outDir } : {}) });
|
|
45
|
+
const judgeModel = stringFlag(parsed, "judge-model") ?? env.OPENEVE_EVAL_JUDGE_MODEL ?? artifact.manifest.agent.model;
|
|
46
|
+
const config = {
|
|
47
|
+
...(filter ? { filter } : {}),
|
|
48
|
+
...(tag ? { tag } : {}),
|
|
49
|
+
judgeModel,
|
|
50
|
+
concurrency,
|
|
51
|
+
timeoutMs,
|
|
52
|
+
approve,
|
|
53
|
+
failFast,
|
|
54
|
+
repetitions,
|
|
55
|
+
retryAttempts,
|
|
56
|
+
retryBackoffMs
|
|
57
|
+
};
|
|
58
|
+
const experiment = createExperimentManifest({
|
|
59
|
+
startedAt: started.toISOString(),
|
|
60
|
+
agentRevision: artifact.manifest.agentRevision,
|
|
61
|
+
agentRoot,
|
|
62
|
+
cases: discovered,
|
|
63
|
+
config
|
|
64
|
+
});
|
|
65
|
+
const artifacts = createExperimentArtifacts(artifact.root, experiment);
|
|
66
|
+
writeExperimentManifest(experiment, artifacts);
|
|
67
|
+
const evaluatorLoader = new EvalEvaluatorLoader(agentRoot, resolve(artifacts.root, "evaluator-cache"), dependencies.evaluators);
|
|
68
|
+
const executions = expandExecutions(discovered, repetitions);
|
|
69
|
+
const judgeComplete = dependencies.judgeComplete ?? createDefaultJudgeCompleter(env);
|
|
70
|
+
// --url targets a deployed gateway instead of the in-process runtime. Cases
|
|
71
|
+
// that depend on local adapters (fixtures, mocks) or run() inputs the HTTP
|
|
72
|
+
// surface lacks (multi-turn, seeded history) are refused upfront rather than
|
|
73
|
+
// silently graded against a runtime that never saw them.
|
|
74
|
+
const gatewayUrl = stringFlag(parsed, "url");
|
|
75
|
+
let createRuntime = dependencies.createRuntime;
|
|
76
|
+
if (gatewayUrl && !createRuntime) {
|
|
77
|
+
const token = stringFlag(parsed, "token") ?? env.OPENEVE_ADMIN_TOKEN;
|
|
78
|
+
if (!token)
|
|
79
|
+
throw new Error("Remote eval targets require --token or ASSEMBLY_LINE_ADMIN_TOKEN.");
|
|
80
|
+
const host = createGatewayEvalHost({ url: gatewayUrl, token });
|
|
81
|
+
const capabilities = await host.capabilities().catch(() => []);
|
|
82
|
+
const incompatible = discovered.flatMap((item) => {
|
|
83
|
+
const reasons = item.evalCase ? remoteEvalIncompatibilities(item.evalCase, capabilities) : [];
|
|
84
|
+
return reasons.length > 0 ? [`${item.name} (${reasons.join(", ")})`] : [];
|
|
85
|
+
});
|
|
86
|
+
if (incompatible.length > 0) {
|
|
87
|
+
throw new Error(`${incompatible.length} eval case(s) need the in-process runtime and cannot run against --url:\n- ${incompatible.join("\n- ")}\nRun them without --url, or exclude them with --tag/--filter.`);
|
|
88
|
+
}
|
|
89
|
+
createRuntime = () => host;
|
|
90
|
+
}
|
|
91
|
+
const results = await executePool(executions, concurrency, failFast, async (execution, index) => {
|
|
92
|
+
const { item, repetition, repetitionsPlanned } = execution;
|
|
93
|
+
const caseId = caseIdentifier(agentRoot, item.path);
|
|
94
|
+
if (!item.evalCase)
|
|
95
|
+
return malformedResult(item, caseId);
|
|
96
|
+
const caseRoot = resolve(artifacts.root, "cases", `${String(index + 1).padStart(4, "0")}-${safeName(item.name)}`, `repetition-${String(repetition).padStart(2, "0")}`);
|
|
97
|
+
return runEvalCaseWithRetries({
|
|
98
|
+
item: { ...item, evalCase: item.evalCase },
|
|
99
|
+
caseId,
|
|
100
|
+
repetition,
|
|
101
|
+
repetitionsPlanned,
|
|
102
|
+
caseRoot,
|
|
103
|
+
artifactRoot: artifact.root,
|
|
104
|
+
evaluatorLoader,
|
|
105
|
+
approve,
|
|
106
|
+
timeoutMs,
|
|
107
|
+
judgeModel,
|
|
108
|
+
retryAttempts,
|
|
109
|
+
retryBackoffMs,
|
|
110
|
+
env,
|
|
111
|
+
dependencies: {
|
|
112
|
+
...(createRuntime ? { createRuntime } : {}),
|
|
113
|
+
judgeComplete,
|
|
114
|
+
...(dependencies.sleep ? { sleep: dependencies.sleep } : {})
|
|
115
|
+
}
|
|
116
|
+
});
|
|
117
|
+
});
|
|
118
|
+
const durationMs = Math.max(0, now().getTime() - started.getTime());
|
|
119
|
+
const reportArtifacts = {
|
|
120
|
+
root: artifacts.root,
|
|
121
|
+
experimentPath: artifacts.experimentPath,
|
|
122
|
+
resultsPath: artifacts.resultsPath
|
|
123
|
+
};
|
|
124
|
+
let report = createEvalReport(experiment, durationMs, results, { artifacts: reportArtifacts });
|
|
125
|
+
const baselinePath = stringFlag(parsed, "baseline");
|
|
126
|
+
if (baselinePath) {
|
|
127
|
+
const baseline = loadBaselineReport(baselinePath);
|
|
128
|
+
const comparison = compareEvalReports(baseline, report, {
|
|
129
|
+
maxRegressions: nonnegativeIntegerFlag(parsed, "max-regressions", 0),
|
|
130
|
+
...(parsed.flags["max-cost-increase-percent"] === undefined
|
|
131
|
+
? {}
|
|
132
|
+
: { maxCostIncreasePercent: nonnegativeNumberFlag(parsed, "max-cost-increase-percent") })
|
|
133
|
+
});
|
|
134
|
+
report = createEvalReport(experiment, durationMs, results, { artifacts: reportArtifacts, comparison });
|
|
135
|
+
}
|
|
136
|
+
writeExperimentReport(report, artifacts);
|
|
137
|
+
(dependencies.log ?? console.log)(booleanFlag(parsed, "json") ? JSON.stringify(report, null, 2) : formatEvalReport(report));
|
|
138
|
+
if (report.summary.failed + report.summary.errored > 0 || report.comparison?.passed === false)
|
|
139
|
+
process.exitCode = 1;
|
|
140
|
+
return report;
|
|
141
|
+
}
|
|
142
|
+
async function executePool(cases, concurrency, failFast, execute) {
|
|
143
|
+
const results = new Array(cases.length);
|
|
144
|
+
let next = 0;
|
|
145
|
+
let stopped = false;
|
|
146
|
+
const workers = Array.from({ length: Math.min(concurrency, Math.max(1, cases.length)) }, async () => {
|
|
147
|
+
while (!stopped) {
|
|
148
|
+
const index = next;
|
|
149
|
+
next += 1;
|
|
150
|
+
const item = cases[index];
|
|
151
|
+
if (!item)
|
|
152
|
+
return;
|
|
153
|
+
const result = await execute(item, index);
|
|
154
|
+
results[index] = result;
|
|
155
|
+
if (failFast && result.outcome !== "passed")
|
|
156
|
+
stopped = true;
|
|
157
|
+
}
|
|
158
|
+
});
|
|
159
|
+
await Promise.all(workers);
|
|
160
|
+
return results.filter((result) => result !== undefined);
|
|
161
|
+
}
|
|
162
|
+
function expandExecutions(cases, repetitions) {
|
|
163
|
+
return cases.flatMap((item) => {
|
|
164
|
+
if (!item.evalCase)
|
|
165
|
+
return [{ item, repetition: 1, repetitionsPlanned: 1 }];
|
|
166
|
+
const planned = item.evalCase.repetitions ?? repetitions;
|
|
167
|
+
return Array.from({ length: planned }, (_unused, index) => ({ item, repetition: index + 1, repetitionsPlanned: planned }));
|
|
168
|
+
});
|
|
169
|
+
}
|
|
170
|
+
function malformedResult(item, caseId) {
|
|
171
|
+
return {
|
|
172
|
+
caseId,
|
|
173
|
+
name: item.name,
|
|
174
|
+
path: item.path,
|
|
175
|
+
tags: [],
|
|
176
|
+
outcome: "errored",
|
|
177
|
+
repetition: 1,
|
|
178
|
+
repetitionsPlanned: 1,
|
|
179
|
+
attempts: 1,
|
|
180
|
+
durationMs: 0,
|
|
181
|
+
runCostUsd: 0,
|
|
182
|
+
judgeCostUsd: 0,
|
|
183
|
+
failures: [],
|
|
184
|
+
error: `Invalid eval case: ${item.error ?? "unknown error"}`
|
|
185
|
+
};
|
|
186
|
+
}
|
|
187
|
+
function positiveIntegerFlag(parsed, name, fallback) {
|
|
188
|
+
const raw = parsed.flags[name];
|
|
189
|
+
if (raw === undefined)
|
|
190
|
+
return fallback;
|
|
191
|
+
const value = typeof raw === "string" ? Number(raw) : Number.NaN;
|
|
192
|
+
if (!Number.isInteger(value) || value <= 0)
|
|
193
|
+
throw new Error(`--${name} must be a positive integer.`);
|
|
194
|
+
return value;
|
|
195
|
+
}
|
|
196
|
+
function nonnegativeIntegerFlag(parsed, name, fallback) {
|
|
197
|
+
const raw = parsed.flags[name];
|
|
198
|
+
if (raw === undefined)
|
|
199
|
+
return fallback;
|
|
200
|
+
const value = typeof raw === "string" ? Number(raw) : Number.NaN;
|
|
201
|
+
if (!Number.isInteger(value) || value < 0)
|
|
202
|
+
throw new Error(`--${name} must be a non-negative integer.`);
|
|
203
|
+
return value;
|
|
204
|
+
}
|
|
205
|
+
function nonnegativeNumberFlag(parsed, name) {
|
|
206
|
+
const raw = parsed.flags[name];
|
|
207
|
+
const value = typeof raw === "string" ? Number(raw) : Number.NaN;
|
|
208
|
+
if (!Number.isFinite(value) || value < 0)
|
|
209
|
+
throw new Error(`--${name} must be a non-negative number.`);
|
|
210
|
+
return value;
|
|
211
|
+
}
|
|
212
|
+
function caseIdentifier(agentRoot, path) {
|
|
213
|
+
return relative(agentRoot, path).replaceAll("\\", "/");
|
|
214
|
+
}
|
|
215
|
+
function safeName(value) {
|
|
216
|
+
return value.toLocaleLowerCase().replace(/[^a-z0-9._-]+/gu, "-").replace(/^-+|-+$/gu, "").slice(0, 80) || "case";
|
|
217
|
+
}
|
|
218
|
+
//# sourceMappingURL=eval.js.map
|
package/dist/eval.js.map
ADDED
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval.js","sourceRoot":"","sources":["../src/eval.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,+BAA+B,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAE,UAAU,EAAmB,MAAM,WAAW,CAAC;AACrE,OAAO,EAAE,sBAAsB,EAAoC,MAAM,uBAAuB,CAAC;AACjG,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,qBAAqB,EAAE,2BAA2B,EAAE,MAAM,wBAAwB,CAAC;AAC5F,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC9E,OAAO,EAAE,mBAAmB,EAAE,MAAM,kBAAkB,CAAC;AACvD,OAAO,EACL,yBAAyB,EACzB,wBAAwB,EAGxB,uBAAuB,EACvB,qBAAqB,EACtB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,2BAA2B,EAAE,MAAM,iBAAiB,CAAC;AAC9D,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAStE,OAAO,EAAE,YAAY,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAC1E,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC9E,OAAO,EAAE,mBAAmB,EAAE,0BAA0B,EAAE,MAAM,kBAAkB,CAAC;AACnF,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AACxE,OAAO,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,qBAAqB,EAAE,oBAAoB,EAAE,cAAc,EAAE,eAAe,EAAwB,MAAM,iBAAiB,CAAC;AACrI,OAAO,EAAE,qBAAqB,EAAE,2BAA2B,EAAE,MAAM,wBAAwB,CAAC;AAC5F,OAAO,EACL,yBAAyB,EACzB,wBAAwB,EACxB,oBAAoB,EACpB,gBAAgB,EAChB,uBAAuB,EACvB,qBAAqB,EACtB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,2BAA2B,EAAE,sBAAsB,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnH,OAAO,EAAE,gBAAgB,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,wBAAwB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAC3E,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAgBlD,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,SAAiB,EACjB,MAAkB,EAClB,eAAwC,EAAE;IAE1C,MAAM,GAAG,GAAG,YAAY,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC;IAC5C,MAAM,GAAG,GAAG,YAAY,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;IACtB,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAC;IAC5C,MAAM,GAAG,GAAG,UAAU,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IACtC,MAAM,UAAU,GAAG,aAAa,CAAC,SAAS,EAAE;QAC1C,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7B,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACxB,CAAC,CAAC;IACH,MAAM,WAAW,GAAG,mBAAmB,CAAC,MAAM,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC;IAClE,MAAM,SAAS,GAAG,mBAAmB,CAAC,MAAM,EAAE,SAAS,EAAE,GAAG,CAAC,GAAG,KAAK,CAAC;IACtE,MAAM,WAAW,GAAG,mBAAmB,CAAC,MAAM,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC;IAClE,MAAM,aAAa,GAAG,sBAAsB,CAAC,MAAM,EAAE,gBAAgB,EAAE,CAAC,CAAC,CAAC;IAC1E,MAAM,cAAc,GAAG,sBAAsB,CAAC,MAAM,EAAE,kBAAkB,EAAE,GAAG,CAAC,CAAC;IAC/E,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IAC/C,MAAM,QAAQ,GAAG,WAAW,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;IAClD,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IACzC,MAAM,QAAQ,GAAG,UAAU,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;IAChF,MAAM,UAAU,GAAG,UAAU,CAAC,MAAM,EAAE,aAAa,CAAC,IAAI,GAAG,CAAC,wBAAwB,IAAI,QAAQ,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,CAAC;IACtH,MAAM,MAAM,GAAwB;QAClC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7B,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvB,UAAU;QACV,WAAW;QACX,SAAS;QACT,OAAO;QACP,QAAQ;QACR,WAAW;QACX,aAAa;QACb,cAAc;KACf,CAAC;IACF,MAAM,UAAU,GAAG,wBAAwB,CAAC;QAC1C,SAAS,EAAE,OAAO,CAAC,WAAW,EAAE;QAChC,aAAa,EAAE,QAAQ,CAAC,QAAQ,CAAC,aAAa;QAC9C,SAAS;QACT,KAAK,EAAE,UAAU;QACjB,MAAM;KACP,CAAC,CAAC;IACH,MAAM,SAAS,GAAG,yBAAyB,CAAC,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;IACvE,uBAAuB,CAAC,UAAU,EAAE,SAAS,CAAC,CAAC;IAC/C,MAAM,eAAe,GAAG,IAAI,mBAAmB,CAAC,SAAS,EAAE,OAAO,CAAC,SAAS,CAAC,IAAI,EAAE,iBAAiB,CAAC,EAAE,YAAY,CAAC,UAAU,CAAC,CAAC;IAChI,MAAM,UAAU,GAAG,gBAAgB,CAAC,UAAU,EAAE,WAAW,CAAC,CAAC;IAC7D,MAAM,aAAa,GAAG,YAAY,CAAC,aAAa,IAAI,2BAA2B,CAAC,GAAG,CAAC,CAAC;IACrF,4EAA4E;IAC5E,2EAA2E;IAC3E,6EAA6E;IAC7E,yDAAyD;IACzD,MAAM,UAAU,GAAG,UAAU,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;IAC7C,IAAI,aAAa,GAAG,YAAY,CAAC,aAAa,CAAC;IAC/C,IAAI,UAAU,IAAI,CAAC,aAAa,EAAE,CAAC;QACjC,MAAM,KAAK,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,GAAG,CAAC,mBAAmB,CAAC;QACrE,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAC;QACjG,MAAM,IAAI,GAAG,qBAAqB,CAAC,EAAE,GAAG,EAAE,UAAU,EAAE,KAAK,EAAE,CAAC,CAAC;QAC/D,MAAM,YAAY,GAAG,MAAM,IAAI,CAAC,YAAY,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAc,CAAC,CAAC;QAC3E,MAAM,YAAY,GAAG,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE;YAC/C,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,2BAA2B,CAAC,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAC9F,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,KAAK,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5E,CAAC,CAAC,CAAC;QACH,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,MAAM,IAAI,KAAK,CACb,GAAG,YAAY,CAAC,MAAM,8EAA8E,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,gEAAgE,CAC9L,CAAC;QACJ,CAAC;QACD,aAAa,GAAG,GAAG,EAAE,CAAC,IAAI,CAAC;IAC7B,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,UAAU,EAAE,WAAW,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,EAAE;QAC9F,MAAM,EAAE,IAAI,EAAE,UAAU,EAAE,kBAAkB,EAAE,GAAG,SAAS,CAAC;QAC3D,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC,IAAI,CAAC,QAAQ;YAAE,OAAO,eAAe,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzD,MAAM,QAAQ,GAAG,OAAO,CACtB,SAAS,CAAC,IAAI,EACd,OAAO,EACP,GAAG,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAC9D,cAAc,MAAM,CAAC,UAAU,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CACpD,CAAC;QACF,OAAO,sBAAsB,CAAC;YAC5B,IAAI,EAAE,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE;YAC1C,MAAM;YACN,UAAU;YACV,kBAAkB;YAClB,QAAQ;YACR,YAAY,EAAE,QAAQ,CAAC,IAAI;YAC3B,eAAe;YACf,OAAO;YACP,SAAS;YACT,UAAU;YACV,aAAa;YACb,cAAc;YACd,GAAG;YACH,YAAY,EAAE;gBACZ,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC3C,aAAa;gBACb,GAAG,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,YAAY,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC7D;SACF,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,OAAO,EAAE,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC,CAAC;IACpE,MAAM,eAAe,GAAG;QACtB,IAAI,EAAE,SAAS,CAAC,IAAI;QACpB,cAAc,EAAE,SAAS,CAAC,cAAc;QACxC,WAAW,EAAE,SAAS,CAAC,WAAW;KACnC,CAAC;IACF,IAAI,MAAM,GAAG,gBAAgB,CAAC,UAAU,EAAE,UAAU,EAAE,OAAO,EAAE,EAAE,SAAS,EAAE,eAAe,EAAE,CAAC,CAAC;IAC/F,MAAM,YAAY,GAAG,UAAU,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;IACpD,IAAI,YAAY,EAAE,CAAC;QACjB,MAAM,QAAQ,GAAG,kBAAkB,CAAC,YAAY,CAAC,CAAC;QAClD,MAAM,UAAU,GAAG,kBAAkB,CAAC,QAAQ,EAAE,MAAM,EAAE;YACtD,cAAc,EAAE,sBAAsB,CAAC,MAAM,EAAE,iBAAiB,EAAE,CAAC,CAAC;YACpE,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,2BAA2B,CAAC,KAAK,SAAS;gBACzD,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,sBAAsB,EAAE,qBAAqB,CAAC,MAAM,EAAE,2BAA2B,CAAC,EAAE,CAAC;SAC5F,CAAC,CAAC;QACH,MAAM,GAAG,gBAAgB,CAAC,UAAU,EAAE,UAAU,EAAE,OAAO,EAAE,EAAE,SAAS,EAAE,eAAe,EAAE,UAAU,EAAE,CAAC,CAAC;IACzG,CAAC;IACD,qBAAqB,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IACzC,CAAC,YAAY,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC,CAAC,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,MAAM,CAAC,CAAC,CAAC;IAC5H,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,OAAO,GAAG,CAAC,IAAI,MAAM,CAAC,UAAU,EAAE,MAAM,KAAK,KAAK;QAAE,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;IACpH,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,KAAK,UAAU,WAAW,CACxB,KAAsB,EACtB,WAAmB,EACnB,QAAiB,EACjB,OAAwE;IAExE,MAAM,OAAO,GAAsC,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC3E,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,IAAI,OAAO,GAAG,KAAK,CAAC;IACpB,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,KAAK,IAAI,EAAE;QAClG,OAAO,CAAC,OAAO,EAAE,CAAC;YAChB,MAAM,KAAK,GAAG,IAAI,CAAC;YACnB,IAAI,IAAI,CAAC,CAAC;YACV,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;YAC1B,IAAI,CAAC,IAAI;gBAAE,OAAO;YAClB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;YAC1C,OAAO,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC;YACxB,IAAI,QAAQ,IAAI,MAAM,CAAC,OAAO,KAAK,QAAQ;gBAAE,OAAO,GAAG,IAAI,CAAC;QAC9D,CAAC;IACH,CAAC,CAAC,CAAC;IACH,MAAM,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAC3B,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM,EAA4B,EAAE,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;AACpF,CAAC;AAED,SAAS,gBAAgB,CAAC,KAA2B,EAAE,WAAmB;IACxE,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE;QAC5B,IAAI,CAAC,IAAI,CAAC,QAAQ;YAAE,OAAO,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,kBAAkB,EAAE,CAAC,EAAE,CAAC,CAAC;QAC5E,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,WAAW,IAAI,WAAW,CAAC;QACzD,OAAO,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,GAAG,CAAC,EAAE,kBAAkB,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;IAC7H,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,eAAe,CAAC,IAAwB,EAAE,MAAc;IAC/D,OAAO;QACL,MAAM;QACN,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,IAAI,EAAE,EAAE;QACR,OAAO,EAAE,SAAS;QAClB,UAAU,EAAE,CAAC;QACb,kBAAkB,EAAE,CAAC;QACrB,QAAQ,EAAE,CAAC;QACX,UAAU,EAAE,CAAC;QACb,UAAU,EAAE,CAAC;QACb,YAAY,EAAE,CAAC;QACf,QAAQ,EAAE,EAAE;QACZ,KAAK,EAAE,sBAAsB,IAAI,CAAC,KAAK,IAAI,eAAe,EAAE;KAC7D,CAAC;AACJ,CAAC;AAED,SAAS,mBAAmB,CAAC,MAAkB,EAAE,IAAY,EAAE,QAAgB;IAC7E,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,IAAI,GAAG,KAAK,SAAS;QAAE,OAAO,QAAQ,CAAC;IACvC,MAAM,KAAK,GAAG,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;IACjE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,KAAK,IAAI,8BAA8B,CAAC,CAAC;IACrG,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,sBAAsB,CAAC,MAAkB,EAAE,IAAY,EAAE,QAAgB;IAChF,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,IAAI,GAAG,KAAK,SAAS;QAAE,OAAO,QAAQ,CAAC;IACvC,MAAM,KAAK,GAAG,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;IACjE,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,KAAK,IAAI,kCAAkC,CAAC,CAAC;IACxG,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,qBAAqB,CAAC,MAAkB,EAAE,IAAY;IAC7D,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,MAAM,KAAK,GAAG,OAAO,GAAG,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;IACjE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,KAAK,IAAI,iCAAiC,CAAC,CAAC;IACtG,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,cAAc,CAAC,SAAiB,EAAE,IAAY;IACrD,OAAO,QAAQ,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;AACzD,CAAC;AAED,SAAS,QAAQ,CAAC,KAAa;IAC7B,OAAO,KAAK,CAAC,iBAAiB,EAAE,CAAC,OAAO,CAAC,iBAAiB,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,MAAM,CAAC;AACnH,CAAC"}
|
package/dist/help.d.ts
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
export declare function printHelp(command?: string): void;
|
|
2
|
+
export declare function printCommandHelp(command: string): void;
|
|
3
|
+
export declare function isKnownCommand(command: string): boolean;
|
|
4
|
+
export declare function printIssues(issues: {
|
|
5
|
+
level: string;
|
|
6
|
+
code: string;
|
|
7
|
+
message: string;
|
|
8
|
+
path?: string;
|
|
9
|
+
}[], options?: {
|
|
10
|
+
agentRoot?: string;
|
|
11
|
+
}): void;
|
|
12
|
+
//# sourceMappingURL=help.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"help.d.ts","sourceRoot":"","sources":["../src/help.ts"],"names":[],"mappings":"AAoQA,wBAAgB,SAAS,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,CAuBhD;AAED,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAsBtD;AAED,wBAAgB,cAAc,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAEvD;AAED,wBAAgB,WAAW,CACzB,MAAM,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAA;CAAE,EAAE,EACzE,OAAO,GAAE;IAAE,SAAS,CAAC,EAAE,MAAM,CAAA;CAAO,GACnC,IAAI,CAQN"}
|