@lesliechan721/agent-skill-eval 0.1.0-beta.7 → 0.1.0-beta.9
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/dist/bin.js +149 -39
- package/dist/bin.js.map +1 -1
- package/dist/codex-eval-app-server.d.ts +199 -0
- package/dist/codex-eval-app-server.js +1366 -0
- package/dist/codex-eval-app-server.js.map +1 -0
- package/dist/codex-eval-cases.d.ts +43 -0
- package/dist/codex-eval-cases.js +327 -0
- package/dist/codex-eval-cases.js.map +1 -0
- package/dist/codex-eval-isolation.d.ts +61 -0
- package/dist/codex-eval-isolation.js +318 -0
- package/dist/codex-eval-isolation.js.map +1 -0
- package/dist/codex-eval-process-tree.d.ts +38 -0
- package/dist/codex-eval-process-tree.js +803 -0
- package/dist/codex-eval-process-tree.js.map +1 -0
- package/dist/codex-eval-runtime-config.d.ts +43 -0
- package/dist/codex-eval-runtime-config.js +607 -0
- package/dist/codex-eval-runtime-config.js.map +1 -0
- package/dist/codex-trigger-eval.d.ts +82 -0
- package/dist/codex-trigger-eval.js +789 -0
- package/dist/codex-trigger-eval.js.map +1 -0
- package/dist/codex-workflow.d.ts +188 -0
- package/dist/codex-workflow.js +1790 -0
- package/dist/codex-workflow.js.map +1 -0
- package/dist/credentials.d.ts +5 -0
- package/dist/credentials.js +30 -0
- package/dist/credentials.js.map +1 -0
- package/dist/custom-process.d.ts +58 -0
- package/dist/custom-process.js +409 -0
- package/dist/custom-process.js.map +1 -0
- package/dist/eval-audit.d.ts +38 -0
- package/dist/eval-audit.js +71 -0
- package/dist/eval-audit.js.map +1 -0
- package/dist/eval-file.d.ts +9 -0
- package/dist/eval-file.js +15 -0
- package/dist/eval-file.js.map +1 -0
- package/dist/eval-result.d.ts +64 -0
- package/dist/eval-result.js +144 -0
- package/dist/eval-result.js.map +1 -0
- package/dist/eval-run.d.ts +23 -0
- package/dist/eval-run.js +539 -0
- package/dist/eval-run.js.map +1 -0
- package/dist/eval-spec.d.ts +57 -0
- package/dist/eval-spec.js +120 -0
- package/dist/eval-spec.js.map +1 -0
- package/dist/eval-suite-binding.d.ts +4 -0
- package/dist/eval-suite-binding.js +19 -0
- package/dist/eval-suite-binding.js.map +1 -0
- package/dist/index.d.ts +34 -9
- package/dist/index.js +18 -6
- package/dist/index.js.map +1 -1
- package/dist/json-schema.js +75 -5
- package/dist/json-schema.js.map +1 -1
- package/dist/output-path.js +2 -2
- package/dist/process-tree.d.ts +19 -1
- package/dist/process-tree.js +321 -66
- package/dist/process-tree.js.map +1 -1
- package/dist/process.d.ts +0 -2
- package/dist/process.js +76 -25
- package/dist/process.js.map +1 -1
- package/dist/provider-doctor.d.ts +33 -0
- package/dist/provider-doctor.js +86 -0
- package/dist/provider-doctor.js.map +1 -0
- package/dist/providers.d.ts +82 -0
- package/dist/providers.js +630 -0
- package/dist/providers.js.map +1 -0
- package/dist/run-paths.d.ts +8 -0
- package/dist/run-paths.js +71 -0
- package/dist/run-paths.js.map +1 -0
- package/dist/runtime-skill.d.ts +6 -0
- package/dist/runtime-skill.js +87 -3
- package/dist/runtime-skill.js.map +1 -1
- package/dist/schemas/digests.json +3 -3
- package/dist/schemas/eval-audit.schema.json +64 -0
- package/dist/schemas/eval-result.schema.json +202 -0
- package/dist/schemas/eval-spec.schema.json +109 -0
- package/dist/schemas.d.ts +1 -3
- package/dist/schemas.js +0 -17
- package/dist/schemas.js.map +1 -1
- package/dist/selected-cases.d.ts +30 -0
- package/dist/selected-cases.js +388 -0
- package/dist/selected-cases.js.map +1 -0
- package/dist/types.d.ts +2 -34
- package/package.json +6 -1
- package/dist/bundle.d.ts +0 -93
- package/dist/bundle.js +0 -216
- package/dist/bundle.js.map +0 -1
- package/dist/capabilities.d.ts +0 -2
- package/dist/capabilities.js +0 -38
- package/dist/capabilities.js.map +0 -1
- package/dist/plan.d.ts +0 -79
- package/dist/plan.js +0 -285
- package/dist/plan.js.map +0 -1
- package/dist/run.d.ts +0 -2
- package/dist/run.js +0 -592
- package/dist/run.js.map +0 -1
- package/dist/schemas/aw-skill-eval-execution-plan.schema.json +0 -220
- package/dist/schemas/eval-run-bundle.schema.json +0 -408
- package/dist/schemas/eval-runtime-capabilities.schema.json +0 -85
package/dist/eval-run.js
ADDED
|
@@ -0,0 +1,539 @@
|
|
|
1
|
+
import { randomUUID } from 'node:crypto';
|
|
2
|
+
import { mkdir, readFile, rename, writeFile } from 'node:fs/promises';
|
|
3
|
+
import path from 'node:path';
|
|
4
|
+
import process from 'node:process';
|
|
5
|
+
import fs from 'fs-extra';
|
|
6
|
+
import { sha256, toPosixRelative } from './common.js';
|
|
7
|
+
import { credentialValues, redactSecrets } from './credentials.js';
|
|
8
|
+
import { CODEX_PROVIDER_ENVIRONMENT_ALLOWLIST, executeCustomProcessProvider, } from './custom-process.js';
|
|
9
|
+
import { validateEvalAuditFile } from './eval-audit.js';
|
|
10
|
+
import { computeEvalResultSummary, createEvalResultBinding, validateEvalResult, } from './eval-result.js';
|
|
11
|
+
import { validateEvalSpecFile } from './eval-spec.js';
|
|
12
|
+
import { assertEvalProviderSuiteKind } from './eval-suite-binding.js';
|
|
13
|
+
import { runBuiltinEvalProvider, } from './providers.js';
|
|
14
|
+
import { resolveCodexEvalRuntimeConfig } from './codex-eval-runtime-config.js';
|
|
15
|
+
import { ProviderProcessCleanupError } from './process.js';
|
|
16
|
+
import { executeCodexWorkflowProvider, } from './codex-workflow.js';
|
|
17
|
+
import { computeCanonicalDirectoryDigest, computeSkillDigest } from './runtime-skill.js';
|
|
18
|
+
import { parseSelectedCases, } from './selected-cases.js';
|
|
19
|
+
class BoundInputChangedError extends Error {
|
|
20
|
+
}
|
|
21
|
+
function secretValues(sourceEnv) {
|
|
22
|
+
return credentialValues(sourceEnv, CODEX_PROVIDER_ENVIRONMENT_ALLOWLIST);
|
|
23
|
+
}
|
|
24
|
+
function redactText(value, secrets) {
|
|
25
|
+
return redactSecrets(value, secrets);
|
|
26
|
+
}
|
|
27
|
+
function redactValue(value, secrets) {
|
|
28
|
+
if (typeof value === 'string')
|
|
29
|
+
return redactText(value, secrets);
|
|
30
|
+
if (Array.isArray(value))
|
|
31
|
+
return value.map((entry) => redactValue(entry, secrets));
|
|
32
|
+
if (value && typeof value === 'object') {
|
|
33
|
+
return Object.fromEntries(Object.entries(value).map(([key, entry]) => [
|
|
34
|
+
redactText(key, secrets),
|
|
35
|
+
redactValue(entry, secrets),
|
|
36
|
+
]));
|
|
37
|
+
}
|
|
38
|
+
return value;
|
|
39
|
+
}
|
|
40
|
+
async function inputDigests(calculated) {
|
|
41
|
+
return {
|
|
42
|
+
workspace: await computeCanonicalDirectoryDigest(calculated.paths.workspace),
|
|
43
|
+
target: await computeSkillDigest(calculated.paths.target),
|
|
44
|
+
cases: sha256(await readFile(calculated.paths.cases)),
|
|
45
|
+
previousTarget: calculated.paths.previousTarget
|
|
46
|
+
? await computeSkillDigest(calculated.paths.previousTarget)
|
|
47
|
+
: null,
|
|
48
|
+
};
|
|
49
|
+
}
|
|
50
|
+
function sameDigests(left, right) {
|
|
51
|
+
return left.workspace === right.workspace &&
|
|
52
|
+
left.target === right.target &&
|
|
53
|
+
left.cases === right.cases &&
|
|
54
|
+
left.previousTarget === right.previousTarget;
|
|
55
|
+
}
|
|
56
|
+
async function assertInputsUnchanged(calculated, captured) {
|
|
57
|
+
let current;
|
|
58
|
+
try {
|
|
59
|
+
current = await inputDigests(calculated);
|
|
60
|
+
}
|
|
61
|
+
catch (error) {
|
|
62
|
+
throw new BoundInputChangedError(`Prepared Eval input changed: ${error.message}`);
|
|
63
|
+
}
|
|
64
|
+
if (!sameDigests(captured, current) || !digestsMatchSpec(calculated, current)) {
|
|
65
|
+
throw new BoundInputChangedError('Prepared Eval input digest changed or no longer matches EvalSpec digests.');
|
|
66
|
+
}
|
|
67
|
+
}
|
|
68
|
+
function digestsMatchSpec(calculated, digests) {
|
|
69
|
+
return digests.workspace === calculated.workspaceDigest
|
|
70
|
+
&& digests.workspace === calculated.spec.workspace.digest
|
|
71
|
+
&& digests.target === calculated.targetDigest
|
|
72
|
+
&& digests.cases === calculated.caseDigest
|
|
73
|
+
&& digests.previousTarget === calculated.previousTargetDigest;
|
|
74
|
+
}
|
|
75
|
+
function assertDigestsMatchSpec(calculated, digests) {
|
|
76
|
+
if (!digestsMatchSpec(calculated, digests)) {
|
|
77
|
+
throw new BoundInputChangedError('Prepared Eval input no longer matches EvalSpec digests.');
|
|
78
|
+
}
|
|
79
|
+
}
|
|
80
|
+
function triggerProviderCase(document, evalCase) {
|
|
81
|
+
if (!document.suite.providerSuite || !evalCase.providerCase) {
|
|
82
|
+
throw new Error('Selected trigger case has no canonical provider form.');
|
|
83
|
+
}
|
|
84
|
+
return {
|
|
85
|
+
suite: document.suite.providerSuite,
|
|
86
|
+
value: evalCase.providerCase,
|
|
87
|
+
};
|
|
88
|
+
}
|
|
89
|
+
export function adaptConversationProviderCase(document, evalCase) {
|
|
90
|
+
const participants = [...new Set([
|
|
91
|
+
document.suite.unitUnderTest,
|
|
92
|
+
...(document.suite.participants ?? []),
|
|
93
|
+
])];
|
|
94
|
+
const assertions = evalCase.assertions;
|
|
95
|
+
const input = evalCase.input && typeof evalCase.input === 'object' && !Array.isArray(evalCase.input)
|
|
96
|
+
? evalCase.input
|
|
97
|
+
: {};
|
|
98
|
+
const expected = evalCase.expected && typeof evalCase.expected === 'object' && !Array.isArray(evalCase.expected)
|
|
99
|
+
? evalCase.expected
|
|
100
|
+
: {};
|
|
101
|
+
const audit = evalCase.audit && typeof evalCase.audit === 'object' && !Array.isArray(evalCase.audit)
|
|
102
|
+
? evalCase.audit
|
|
103
|
+
: null;
|
|
104
|
+
const workspaceUnchanged = expected.workspaceUnchanged === undefined
|
|
105
|
+
? true
|
|
106
|
+
: expected.workspaceUnchanged;
|
|
107
|
+
const scriptedReplies = Array.isArray(input.scriptedReplies)
|
|
108
|
+
? input.scriptedReplies.map((reply) => {
|
|
109
|
+
if (typeof reply === 'string')
|
|
110
|
+
return reply;
|
|
111
|
+
const row = reply;
|
|
112
|
+
return {
|
|
113
|
+
answers_by_index: row.answersByIndex,
|
|
114
|
+
...(row.expectedQuestionCount === undefined
|
|
115
|
+
? {}
|
|
116
|
+
: { expected_question_count: row.expectedQuestionCount }),
|
|
117
|
+
unexpected_question_answer: row.unexpectedQuestionAnswer,
|
|
118
|
+
};
|
|
119
|
+
})
|
|
120
|
+
: [];
|
|
121
|
+
const workspaceFiles = input.workspaceFiles && typeof input.workspaceFiles === 'object'
|
|
122
|
+
&& !Array.isArray(input.workspaceFiles)
|
|
123
|
+
? input.workspaceFiles
|
|
124
|
+
: {};
|
|
125
|
+
const allowedWorkspaceChanges = Array.isArray(expected.allowedWorkspaceChanges)
|
|
126
|
+
? expected.allowedWorkspaceChanges
|
|
127
|
+
: [];
|
|
128
|
+
const artifacts = Array.isArray(expected.artifacts)
|
|
129
|
+
? expected.artifacts.map((artifact) => {
|
|
130
|
+
const row = artifact;
|
|
131
|
+
const { notContains, ...providerArtifact } = row;
|
|
132
|
+
return {
|
|
133
|
+
...providerArtifact,
|
|
134
|
+
...(Object.hasOwn(row, 'notContains') ? { not_contains: notContains } : {}),
|
|
135
|
+
};
|
|
136
|
+
})
|
|
137
|
+
: [];
|
|
138
|
+
return {
|
|
139
|
+
value: {
|
|
140
|
+
id: evalCase.id,
|
|
141
|
+
prompt: evalCase.prompt,
|
|
142
|
+
scripted_replies: scriptedReplies,
|
|
143
|
+
expected_output: expected.output ?? 'Evaluate the response against the approved assertions.',
|
|
144
|
+
...(Object.keys(workspaceFiles).length > 0 ? { workspace_files: workspaceFiles } : {}),
|
|
145
|
+
...(Array.isArray(input.pathPrepend) && input.pathPrepend.length > 0
|
|
146
|
+
? { environment: { pathPrepend: input.pathPrepend } }
|
|
147
|
+
: {}),
|
|
148
|
+
sandbox: workspaceUnchanged === false ? 'workspace-write' : 'read-only',
|
|
149
|
+
review: {
|
|
150
|
+
checks: assertions,
|
|
151
|
+
workspace_unchanged: workspaceUnchanged,
|
|
152
|
+
...(allowedWorkspaceChanges.length > 0
|
|
153
|
+
? { allowed_workspace_changes: allowedWorkspaceChanges }
|
|
154
|
+
: {}),
|
|
155
|
+
...(artifacts.length > 0 ? { artifacts } : {}),
|
|
156
|
+
},
|
|
157
|
+
},
|
|
158
|
+
skillName: document.suite.unitUnderTest,
|
|
159
|
+
skillNames: participants,
|
|
160
|
+
isolation: document.suite.runtime?.isolation === 'strict' ? 'strict' : 'standard',
|
|
161
|
+
suitePathPrepend: document.suite.runtime?.pathPrepend ?? [],
|
|
162
|
+
assertions,
|
|
163
|
+
...(audit && Object.hasOwn(audit, 'allowedVariance')
|
|
164
|
+
? { allowedVariance: audit.allowedVariance }
|
|
165
|
+
: {}),
|
|
166
|
+
};
|
|
167
|
+
}
|
|
168
|
+
function providerCase(document, evalCase) {
|
|
169
|
+
if (document.suite.kind === 'trigger')
|
|
170
|
+
return triggerProviderCase(document, evalCase);
|
|
171
|
+
if (document.suite.kind === 'conversation')
|
|
172
|
+
return adaptConversationProviderCase(document, evalCase);
|
|
173
|
+
return evalCase;
|
|
174
|
+
}
|
|
175
|
+
async function prepareAttempt(options) {
|
|
176
|
+
const attemptRoot = path.join(options.calculated.paths.outputDir, 'runs', options.caseId, options.variant, `attempt-${options.attempt}`);
|
|
177
|
+
const workspacePath = path.join(attemptRoot, 'workspace', 'subject');
|
|
178
|
+
const fixturesPath = path.join(attemptRoot, 'workspace', 'eval');
|
|
179
|
+
const runOutput = path.join(attemptRoot, 'output');
|
|
180
|
+
await fs.ensureDir(path.dirname(attemptRoot));
|
|
181
|
+
await mkdir(attemptRoot);
|
|
182
|
+
try {
|
|
183
|
+
await mkdir(path.join(attemptRoot, 'workspace'));
|
|
184
|
+
await fs.copy(path.join(options.snapshot.workspace, 'subject'), workspacePath, {
|
|
185
|
+
dereference: false,
|
|
186
|
+
overwrite: false,
|
|
187
|
+
errorOnExist: true,
|
|
188
|
+
});
|
|
189
|
+
await fs.copy(path.join(options.snapshot.workspace, 'eval'), fixturesPath, {
|
|
190
|
+
dereference: false,
|
|
191
|
+
overwrite: false,
|
|
192
|
+
errorOnExist: true,
|
|
193
|
+
});
|
|
194
|
+
await mkdir(runOutput);
|
|
195
|
+
}
|
|
196
|
+
catch (error) {
|
|
197
|
+
await fs.remove(attemptRoot);
|
|
198
|
+
throw error;
|
|
199
|
+
}
|
|
200
|
+
const target = options.variant === 'current'
|
|
201
|
+
? options.snapshot.currentTarget
|
|
202
|
+
: options.variant === 'previous-version'
|
|
203
|
+
? options.snapshot.previousTarget
|
|
204
|
+
: null;
|
|
205
|
+
if (target) {
|
|
206
|
+
try {
|
|
207
|
+
const destination = path.join(workspacePath, '.agents', 'skills', options.document.suite.unitUnderTest);
|
|
208
|
+
if (await fs.pathExists(destination)) {
|
|
209
|
+
throw new Error('Prepared subject unexpectedly contains the unit-under-test Skill.');
|
|
210
|
+
}
|
|
211
|
+
await fs.ensureDir(path.dirname(destination));
|
|
212
|
+
await fs.copy(target, destination, {
|
|
213
|
+
dereference: false,
|
|
214
|
+
overwrite: false,
|
|
215
|
+
errorOnExist: true,
|
|
216
|
+
});
|
|
217
|
+
}
|
|
218
|
+
catch (error) {
|
|
219
|
+
await fs.remove(attemptRoot);
|
|
220
|
+
throw error;
|
|
221
|
+
}
|
|
222
|
+
}
|
|
223
|
+
return { attemptRoot, workspacePath, fixturesPath, runOutput };
|
|
224
|
+
}
|
|
225
|
+
async function createRunInputSnapshot(calculated, captured, casesBytes) {
|
|
226
|
+
const root = path.join(calculated.paths.outputDir, '.run-input');
|
|
227
|
+
const workspace = path.join(root, 'workspace');
|
|
228
|
+
const currentTarget = path.join(root, 'targets', 'current');
|
|
229
|
+
const previousTarget = calculated.paths.previousTarget
|
|
230
|
+
? path.join(root, 'targets', 'previous')
|
|
231
|
+
: null;
|
|
232
|
+
const cases = path.join(root, 'cases.json');
|
|
233
|
+
await mkdir(root);
|
|
234
|
+
try {
|
|
235
|
+
await mkdir(workspace);
|
|
236
|
+
await mkdir(path.join(root, 'targets'));
|
|
237
|
+
await fs.copy(path.join(calculated.paths.workspace, 'subject'), path.join(workspace, 'subject'), {
|
|
238
|
+
dereference: false,
|
|
239
|
+
overwrite: false,
|
|
240
|
+
errorOnExist: true,
|
|
241
|
+
});
|
|
242
|
+
await fs.copy(path.join(calculated.paths.workspace, 'eval'), path.join(workspace, 'eval'), {
|
|
243
|
+
dereference: false,
|
|
244
|
+
overwrite: false,
|
|
245
|
+
errorOnExist: true,
|
|
246
|
+
});
|
|
247
|
+
await fs.copy(calculated.paths.target, currentTarget, {
|
|
248
|
+
dereference: false,
|
|
249
|
+
overwrite: false,
|
|
250
|
+
errorOnExist: true,
|
|
251
|
+
});
|
|
252
|
+
await writeFile(cases, casesBytes, { flag: 'wx', mode: 0o600 });
|
|
253
|
+
if (calculated.paths.previousTarget && previousTarget) {
|
|
254
|
+
await fs.copy(calculated.paths.previousTarget, previousTarget, {
|
|
255
|
+
dereference: false,
|
|
256
|
+
overwrite: false,
|
|
257
|
+
errorOnExist: true,
|
|
258
|
+
});
|
|
259
|
+
}
|
|
260
|
+
const snapshotDigests = {
|
|
261
|
+
workspace: await computeCanonicalDirectoryDigest(workspace),
|
|
262
|
+
target: await computeSkillDigest(currentTarget),
|
|
263
|
+
cases: sha256(await readFile(cases)),
|
|
264
|
+
previousTarget: previousTarget ? await computeSkillDigest(previousTarget) : null,
|
|
265
|
+
};
|
|
266
|
+
if (!sameDigests(captured, snapshotDigests)) {
|
|
267
|
+
throw new BoundInputChangedError('Run input snapshot does not match the prepared Eval input.');
|
|
268
|
+
}
|
|
269
|
+
assertDigestsMatchSpec(calculated, snapshotDigests);
|
|
270
|
+
return { root, workspace, currentTarget, previousTarget, cases };
|
|
271
|
+
}
|
|
272
|
+
catch (error) {
|
|
273
|
+
await fs.remove(root);
|
|
274
|
+
throw error;
|
|
275
|
+
}
|
|
276
|
+
}
|
|
277
|
+
async function assertSnapshotUnchanged(calculated, snapshot, captured) {
|
|
278
|
+
const current = {
|
|
279
|
+
workspace: await computeCanonicalDirectoryDigest(snapshot.workspace),
|
|
280
|
+
target: await computeSkillDigest(snapshot.currentTarget),
|
|
281
|
+
cases: sha256(await readFile(snapshot.cases)),
|
|
282
|
+
previousTarget: snapshot.previousTarget
|
|
283
|
+
? await computeSkillDigest(snapshot.previousTarget)
|
|
284
|
+
: null,
|
|
285
|
+
};
|
|
286
|
+
if (!sameDigests(captured, current) || !digestsMatchSpec(calculated, current)) {
|
|
287
|
+
throw new BoundInputChangedError('Run input snapshot digest changed or no longer matches EvalSpec digests.');
|
|
288
|
+
}
|
|
289
|
+
}
|
|
290
|
+
function isTimeout(error) {
|
|
291
|
+
return /timed out|timeout/u.test(error instanceof Error ? error.message : String(error));
|
|
292
|
+
}
|
|
293
|
+
function isCleanupFailure(error) {
|
|
294
|
+
return error instanceof ProviderProcessCleanupError ||
|
|
295
|
+
/cleanup|termination|process-tree/u.test(error instanceof Error ? error.message : String(error));
|
|
296
|
+
}
|
|
297
|
+
function abortError() {
|
|
298
|
+
return new Error('Eval execution was aborted.');
|
|
299
|
+
}
|
|
300
|
+
async function writeResultSafely(resultPath, result, binding) {
|
|
301
|
+
const validation = await validateEvalResult(result, binding);
|
|
302
|
+
if (!validation.valid)
|
|
303
|
+
throw new Error(`Generated EvalResult is invalid: ${validation.errors.join('; ')}`);
|
|
304
|
+
if (await fs.pathExists(resultPath))
|
|
305
|
+
throw new Error('Eval result.json already exists.');
|
|
306
|
+
const bytes = Buffer.from(`${JSON.stringify(result, null, 2)}\n`);
|
|
307
|
+
const temporary = path.join(path.dirname(resultPath), `.result.${randomUUID()}.tmp`);
|
|
308
|
+
await writeFile(temporary, bytes, { flag: 'wx', mode: 0o600 });
|
|
309
|
+
try {
|
|
310
|
+
if (await fs.pathExists(resultPath))
|
|
311
|
+
throw new Error('Eval result.json appeared before commit.');
|
|
312
|
+
await rename(temporary, resultPath);
|
|
313
|
+
}
|
|
314
|
+
catch (error) {
|
|
315
|
+
await fs.remove(temporary).catch(() => undefined);
|
|
316
|
+
throw error;
|
|
317
|
+
}
|
|
318
|
+
return sha256(bytes);
|
|
319
|
+
}
|
|
320
|
+
export async function runEvalSpec(specPath, options) {
|
|
321
|
+
if (!options?.auditPath)
|
|
322
|
+
throw new Error('Eval execution requires auditPath.');
|
|
323
|
+
const specValidation = await validateEvalSpecFile(specPath);
|
|
324
|
+
if (!specValidation.valid || !specValidation.calculated) {
|
|
325
|
+
throw new Error(`EvalSpec is invalid: ${specValidation.errors.join('; ')}`);
|
|
326
|
+
}
|
|
327
|
+
const calculated = specValidation.calculated;
|
|
328
|
+
const specRoot = await fs.realpath(path.dirname(path.resolve(specPath)));
|
|
329
|
+
const casesBytes = await readFile(calculated.paths.cases);
|
|
330
|
+
if (sha256(casesBytes) !== calculated.caseDigest) {
|
|
331
|
+
throw new BoundInputChangedError('Selected cases changed after EvalSpec validation.');
|
|
332
|
+
}
|
|
333
|
+
const document = parseSelectedCases(JSON.parse(casesBytes.toString('utf8')));
|
|
334
|
+
if (document.suite.unitUnderTest !== calculated.spec.source.skill) {
|
|
335
|
+
throw new Error('Selected cases unitUnderTest does not match EvalSpec source.skill.');
|
|
336
|
+
}
|
|
337
|
+
assertEvalProviderSuiteKind(calculated.spec.provider, document.suite.kind);
|
|
338
|
+
const selectedCaseIds = document.cases.map((entry) => entry.id);
|
|
339
|
+
const auditValidation = await validateEvalAuditFile(options.auditPath, {
|
|
340
|
+
spec: calculated.spec,
|
|
341
|
+
selectedCaseIds,
|
|
342
|
+
});
|
|
343
|
+
if (!auditValidation.valid || !auditValidation.calculated) {
|
|
344
|
+
throw new Error(`EvalAudit does not admit execution: ${auditValidation.errors.join('; ')}`);
|
|
345
|
+
}
|
|
346
|
+
const resultBinding = createEvalResultBinding(calculated, auditValidation.calculated.auditDigest, document);
|
|
347
|
+
const captured = await inputDigests(calculated);
|
|
348
|
+
assertDigestsMatchSpec(calculated, captured);
|
|
349
|
+
await assertInputsUnchanged(calculated, captured);
|
|
350
|
+
const sourceEnv = options.sourceEnv
|
|
351
|
+
?? options.builtinProviderDependencies?.sourceEnv
|
|
352
|
+
?? process.env;
|
|
353
|
+
const runtime = await resolveCodexEvalRuntimeConfig({
|
|
354
|
+
sourceEnv,
|
|
355
|
+
inheritUserConfig: true,
|
|
356
|
+
modelOverride: options.builtinProviderDependencies?.model
|
|
357
|
+
?? sourceEnv.AW_SKILL_EVAL_CODEX_MODEL,
|
|
358
|
+
});
|
|
359
|
+
const secrets = () => [
|
|
360
|
+
...secretValues(sourceEnv),
|
|
361
|
+
...runtime.sensitiveValues,
|
|
362
|
+
];
|
|
363
|
+
const builtinProviderDependencies = {
|
|
364
|
+
...options.builtinProviderDependencies,
|
|
365
|
+
sourceEnv,
|
|
366
|
+
model: runtime.model ?? undefined,
|
|
367
|
+
codexRuntime: runtime,
|
|
368
|
+
};
|
|
369
|
+
const startedAt = new Date().toISOString();
|
|
370
|
+
const runs = [];
|
|
371
|
+
let stop = false;
|
|
372
|
+
await assertInputsUnchanged(calculated, captured);
|
|
373
|
+
await fs.ensureDir(path.dirname(calculated.paths.outputDir));
|
|
374
|
+
await mkdir(calculated.paths.outputDir);
|
|
375
|
+
let snapshot;
|
|
376
|
+
try {
|
|
377
|
+
snapshot = await createRunInputSnapshot(calculated, captured, casesBytes);
|
|
378
|
+
}
|
|
379
|
+
catch (error) {
|
|
380
|
+
await fs.remove(calculated.paths.outputDir);
|
|
381
|
+
throw error;
|
|
382
|
+
}
|
|
383
|
+
for (const evalCase of document.cases) {
|
|
384
|
+
if (stop)
|
|
385
|
+
break;
|
|
386
|
+
for (const variant of resultBinding.allowedVariants) {
|
|
387
|
+
if (stop)
|
|
388
|
+
break;
|
|
389
|
+
for (let attempt = 1; attempt <= calculated.spec.runs; attempt += 1) {
|
|
390
|
+
const attemptStarted = Date.now();
|
|
391
|
+
let status = 'failed';
|
|
392
|
+
let usage = null;
|
|
393
|
+
let grade = null;
|
|
394
|
+
let error = null;
|
|
395
|
+
let artifacts;
|
|
396
|
+
let runOutput = path.join(calculated.paths.outputDir, 'runs', evalCase.id, variant, `attempt-${attempt}`, 'output');
|
|
397
|
+
try {
|
|
398
|
+
if (options.signal?.aborted)
|
|
399
|
+
throw abortError();
|
|
400
|
+
await assertInputsUnchanged(calculated, captured);
|
|
401
|
+
await assertSnapshotUnchanged(calculated, snapshot, captured);
|
|
402
|
+
const prepared = await prepareAttempt({
|
|
403
|
+
calculated,
|
|
404
|
+
snapshot,
|
|
405
|
+
document,
|
|
406
|
+
caseId: evalCase.id,
|
|
407
|
+
variant,
|
|
408
|
+
attempt,
|
|
409
|
+
});
|
|
410
|
+
runOutput = prepared.runOutput;
|
|
411
|
+
const input = {
|
|
412
|
+
case: providerCase(document, evalCase),
|
|
413
|
+
variant,
|
|
414
|
+
attempt,
|
|
415
|
+
workspacePath: prepared.workspacePath,
|
|
416
|
+
fixturesPath: prepared.fixturesPath,
|
|
417
|
+
outputDir: prepared.runOutput,
|
|
418
|
+
timeoutMs: calculated.spec.timeoutMs,
|
|
419
|
+
...(options.signal ? { signal: options.signal } : {}),
|
|
420
|
+
};
|
|
421
|
+
const providerResult = typeof calculated.spec.provider === 'string'
|
|
422
|
+
? await (options.runBuiltinProvider ?? runBuiltinEvalProvider)(calculated.spec.provider, { ...input, evalRoot: prepared.attemptRoot }, builtinProviderDependencies)
|
|
423
|
+
: calculated.spec.provider.type === 'custom-process'
|
|
424
|
+
? await (options.executeCustomProvider ?? executeCustomProcessProvider)({
|
|
425
|
+
command: calculated.spec.provider.command,
|
|
426
|
+
evalRoot: prepared.attemptRoot,
|
|
427
|
+
input,
|
|
428
|
+
sourceEnv,
|
|
429
|
+
codexRuntime: runtime,
|
|
430
|
+
})
|
|
431
|
+
: await (options.executeCodexWorkflowProvider ?? executeCodexWorkflowProvider)({
|
|
432
|
+
provider: {
|
|
433
|
+
...calculated.spec.provider,
|
|
434
|
+
command: calculated.spec.provider.command,
|
|
435
|
+
},
|
|
436
|
+
evalRoot: prepared.attemptRoot,
|
|
437
|
+
input,
|
|
438
|
+
sourceEnv,
|
|
439
|
+
codexRuntime: runtime,
|
|
440
|
+
codexBin: builtinProviderDependencies.codexBin,
|
|
441
|
+
});
|
|
442
|
+
if (options.signal?.aborted)
|
|
443
|
+
throw abortError();
|
|
444
|
+
const sanitized = redactValue(providerResult, secrets());
|
|
445
|
+
usage = sanitized.usage === null ? null : { ...sanitized.usage };
|
|
446
|
+
grade = sanitized.grade;
|
|
447
|
+
artifacts = typeof calculated.spec.provider === 'object'
|
|
448
|
+
&& calculated.spec.provider.type === 'codex-workflow'
|
|
449
|
+
&& sanitized.artifacts
|
|
450
|
+
? [
|
|
451
|
+
...sanitized.artifacts.map((artifact) => ({ ...artifact })),
|
|
452
|
+
{ path: 'provider-result.json', role: 'result' },
|
|
453
|
+
]
|
|
454
|
+
: undefined;
|
|
455
|
+
await writeFile(path.join(prepared.runOutput, 'provider-result.json'), `${JSON.stringify(sanitized, null, 2)}\n`, { flag: 'wx' });
|
|
456
|
+
status = 'completed';
|
|
457
|
+
}
|
|
458
|
+
catch (caught) {
|
|
459
|
+
status = isTimeout(caught) ? 'timeout' : 'failed';
|
|
460
|
+
usage = null;
|
|
461
|
+
grade = null;
|
|
462
|
+
artifacts = undefined;
|
|
463
|
+
error = { message: redactText(caught instanceof Error ? caught.message : String(caught), secrets()) || 'Provider failed.' };
|
|
464
|
+
if (options.signal?.aborted
|
|
465
|
+
|| caught instanceof BoundInputChangedError
|
|
466
|
+
|| isCleanupFailure(caught)
|
|
467
|
+
|| status === 'timeout'
|
|
468
|
+
|| (typeof calculated.spec.provider === 'object'
|
|
469
|
+
&& calculated.spec.provider.type === 'codex-workflow'))
|
|
470
|
+
stop = true;
|
|
471
|
+
}
|
|
472
|
+
try {
|
|
473
|
+
await assertInputsUnchanged(calculated, captured);
|
|
474
|
+
await assertSnapshotUnchanged(calculated, snapshot, captured);
|
|
475
|
+
}
|
|
476
|
+
catch (changed) {
|
|
477
|
+
status = 'failed';
|
|
478
|
+
usage = null;
|
|
479
|
+
grade = null;
|
|
480
|
+
artifacts = undefined;
|
|
481
|
+
error = { message: redactText(changed.message, secrets()) };
|
|
482
|
+
stop = true;
|
|
483
|
+
}
|
|
484
|
+
runs.push({
|
|
485
|
+
caseId: evalCase.id,
|
|
486
|
+
variant,
|
|
487
|
+
attempt,
|
|
488
|
+
status,
|
|
489
|
+
durationMs: Math.max(0, Date.now() - attemptStarted),
|
|
490
|
+
outputDir: toPosixRelative(specRoot, runOutput),
|
|
491
|
+
usage,
|
|
492
|
+
grade,
|
|
493
|
+
error,
|
|
494
|
+
...(artifacts ? { artifacts } : {}),
|
|
495
|
+
});
|
|
496
|
+
if (stop)
|
|
497
|
+
break;
|
|
498
|
+
}
|
|
499
|
+
}
|
|
500
|
+
}
|
|
501
|
+
try {
|
|
502
|
+
await assertInputsUnchanged(calculated, captured);
|
|
503
|
+
await assertSnapshotUnchanged(calculated, snapshot, captured);
|
|
504
|
+
}
|
|
505
|
+
catch (changed) {
|
|
506
|
+
if (runs.length === 0)
|
|
507
|
+
throw changed;
|
|
508
|
+
const last = runs[runs.length - 1];
|
|
509
|
+
last.status = 'failed';
|
|
510
|
+
last.usage = null;
|
|
511
|
+
last.grade = null;
|
|
512
|
+
delete last.artifacts;
|
|
513
|
+
last.error = { message: redactText(changed.message, secrets()) };
|
|
514
|
+
}
|
|
515
|
+
if (options.signal?.aborted && runs.length > 0) {
|
|
516
|
+
const last = runs[runs.length - 1];
|
|
517
|
+
last.status = 'failed';
|
|
518
|
+
last.usage = null;
|
|
519
|
+
last.grade = null;
|
|
520
|
+
delete last.artifacts;
|
|
521
|
+
last.error = { message: abortError().message };
|
|
522
|
+
}
|
|
523
|
+
const result = {
|
|
524
|
+
version: 1,
|
|
525
|
+
specDigest: calculated.specDigest,
|
|
526
|
+
auditDigest: auditValidation.calculated.auditDigest,
|
|
527
|
+
targetDigest: calculated.targetDigest,
|
|
528
|
+
caseDigest: calculated.caseDigest,
|
|
529
|
+
provider: calculated.spec.provider,
|
|
530
|
+
startedAt,
|
|
531
|
+
finishedAt: new Date().toISOString(),
|
|
532
|
+
runs,
|
|
533
|
+
summary: computeEvalResultSummary(runs),
|
|
534
|
+
};
|
|
535
|
+
const resultPath = path.join(calculated.paths.outputDir, 'result.json');
|
|
536
|
+
const resultDigest = await writeResultSafely(resultPath, result, resultBinding);
|
|
537
|
+
return { valid: true, resultPath, resultDigest, result };
|
|
538
|
+
}
|
|
539
|
+
//# sourceMappingURL=eval-run.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"eval-run.js","sourceRoot":"","sources":["../src/eval-run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAA;AACxC,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAA;AACrE,OAAO,IAAI,MAAM,WAAW,CAAA;AAC5B,OAAO,OAAO,MAAM,cAAc,CAAA;AAClC,OAAO,EAAE,MAAM,UAAU,CAAA;AACzB,OAAO,EAAE,MAAM,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA;AACrD,OAAO,EAAE,gBAAgB,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAA;AAClE,OAAO,EACL,oCAAoC,EACpC,4BAA4B,GAK7B,MAAM,qBAAqB,CAAA;AAC5B,OAAO,EAAE,qBAAqB,EAAE,MAAM,iBAAiB,CAAA;AACvD,OAAO,EACL,wBAAwB,EACxB,uBAAuB,EACvB,kBAAkB,GAInB,MAAM,kBAAkB,CAAA;AACzB,OAAO,EAAE,oBAAoB,EAA2B,MAAM,gBAAgB,CAAA;AAC9E,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAA;AACrE,OAAO,EACL,sBAAsB,GAGvB,MAAM,gBAAgB,CAAA;AACvB,OAAO,EAAE,6BAA6B,EAAE,MAAM,gCAAgC,CAAA;AAC9E,OAAO,EAAE,2BAA2B,EAAE,MAAM,cAAc,CAAA;AAC1D,OAAO,EACL,4BAA4B,GAE7B,MAAM,qBAAqB,CAAA;AAC5B,OAAO,EAAE,+BAA+B,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAA;AACxF,OAAO,EACL,kBAAkB,GAGnB,MAAM,qBAAqB,CAAA;AAyC5B,MAAM,sBAAuB,SAAQ,KAAK;CAAG;AAE7C,SAAS,YAAY,CAAC,SAA4B;IAChD,OAAO,gBAAgB,CAAC,SAAS,EAAE,oCAAoC,CAAC,CAAA;AAC1E,CAAC;AAED,SAAS,UAAU,CAAC,KAAa,EAAE,OAA0B;IAC3D,OAAO,aAAa,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;AACtC,CAAC;AAED,SAAS,WAAW,CAAI,KAAQ,EAAE,OAA0B;IAC1D,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,UAAU,CAAC,KAAK,EAAE,OAAO,CAAM,CAAA;IACrE,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC,CAAM,CAAA;IACvF,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACvC,OAAO,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC;YACpE,UAAU,CAAC,GAAG,EAAE,OAAO,CAAC;YACxB,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC;SAC5B,CAAC,CAAM,CAAA;IACV,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED,KAAK,UAAU,YAAY,CAAC,UAA8B;IACxD,OAAO;QACL,SAAS,EAAE,MAAM,+BAA+B,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC;QAC5E,MAAM,EAAE,MAAM,kBAAkB,CAAC,UAAU,CAAC,KAAK,CAAC,MAAM,CAAC;QACzD,KAAK,EAAE,MAAM,CAAC,MAAM,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;QACrD,cAAc,EAAE,UAAU,CAAC,KAAK,CAAC,cAAc;YAC7C,CAAC,CAAC,MAAM,kBAAkB,CAAC,UAAU,CAAC,KAAK,CAAC,cAAc,CAAC;YAC3D,CAAC,CAAC,IAAI;KACT,CAAA;AACH,CAAC;AAED,SAAS,WAAW,CAAC,IAAuB,EAAE,KAAwB;IACpE,OAAO,IAAI,CAAC,SAAS,KAAK,KAAK,CAAC,SAAS;QACvC,IAAI,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM;QAC5B,IAAI,CAAC,KAAK,KAAK,KAAK,CAAC,KAAK;QAC1B,IAAI,CAAC,cAAc,KAAK,KAAK,CAAC,cAAc,CAAA;AAChD,CAAC;AAED,KAAK,UAAU,qBAAqB,CAClC,UAA8B,EAC9B,QAA2B;IAE3B,IAAI,OAA0B,CAAA;IAC9B,IAAI,CAAC;QACH,OAAO,GAAG,MAAM,YAAY,CAAC,UAAU,CAAC,CAAA;IAC1C,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,IAAI,sBAAsB,CAAC,gCAAiC,KAAe,CAAC,OAAO,EAAE,CAAC,CAAA;IAC9F,CAAC;IACD,IAAI,CAAC,WAAW,CAAC,QAAQ,EAAE,OAAO,CAAC,IAAI,CAAC,gBAAgB,CAAC,UAAU,EAAE,OAAO,CAAC,EAAE,CAAC;QAC9E,MAAM,IAAI,sBAAsB,CAC9B,2EAA2E,CAC5E,CAAA;IACH,CAAC;AACH,CAAC;AAED,SAAS,gBAAgB,CACvB,UAA8B,EAC9B,OAA0B;IAE1B,OAAO,OAAO,CAAC,SAAS,KAAK,UAAU,CAAC,eAAe;WAClD,OAAO,CAAC,SAAS,KAAK,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM;WACtD,OAAO,CAAC,MAAM,KAAK,UAAU,CAAC,YAAY;WAC1C,OAAO,CAAC,KAAK,KAAK,UAAU,CAAC,UAAU;WACvC,OAAO,CAAC,cAAc,KAAK,UAAU,CAAC,oBAAoB,CAAA;AACjE,CAAC;AAED,SAAS,sBAAsB,CAC7B,UAA8B,EAC9B,OAA0B;IAE1B,IAAI,CAAC,gBAAgB,CAAC,UAAU,EAAE,OAAO,CAAC,EAAE,CAAC;QAC3C,MAAM,IAAI,sBAAsB,CAAC,yDAAyD,CAAC,CAAA;IAC7F,CAAC;AACH,CAAC;AAED,SAAS,mBAAmB,CAC1B,QAA+B,EAC/B,QAA0B;IAE1B,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,aAAa,IAAI,CAAC,QAAQ,CAAC,YAAY,EAAE,CAAC;QAC5D,MAAM,IAAI,KAAK,CAAC,uDAAuD,CAAC,CAAA;IAC1E,CAAC;IACD,OAAO;QACL,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,aAAa;QACnC,KAAK,EAAE,QAAQ,CAAC,YAAY;KAC7B,CAAA;AACH,CAAC;AAED,MAAM,UAAU,6BAA6B,CAC3C,QAA+B,EAC/B,QAA0B;IAE1B,MAAM,YAAY,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC;YAC/B,QAAQ,CAAC,KAAK,CAAC,aAAa;YAC5B,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;SACvC,CAAC,CAAC,CAAA;IACH,MAAM,UAAU,GAAG,QAAQ,CAAC,UAAuB,CAAA;IACnD,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,IAAI,OAAO,QAAQ,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC;QAClG,CAAC,CAAC,QAAQ,CAAC,KAAgC;QAC3C,CAAC,CAAC,EAAE,CAAA;IACN,MAAM,QAAQ,GAAG,QAAQ,CAAC,QAAQ,IAAI,OAAO,QAAQ,CAAC,QAAQ,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC;QAC9G,CAAC,CAAC,QAAQ,CAAC,QAAmC;QAC9C,CAAC,CAAC,EAAE,CAAA;IACN,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,IAAI,OAAO,QAAQ,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC;QAClG,CAAC,CAAC,QAAQ,CAAC,KAAgC;QAC3C,CAAC,CAAC,IAAI,CAAA;IACR,MAAM,kBAAkB,GAAG,QAAQ,CAAC,kBAAkB,KAAK,SAAS;QAClE,CAAC,CAAC,IAAI;QACN,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAA;IAC/B,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,eAAe,CAAC;QAC1D,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YAClC,IAAI,OAAO,KAAK,KAAK,QAAQ;gBAAE,OAAO,KAAK,CAAA;YAC3C,MAAM,GAAG,GAAG,KAAgC,CAAA;YAC5C,OAAO;gBACL,gBAAgB,EAAE,GAAG,CAAC,cAAc;gBACpC,GAAG,CAAC,GAAG,CAAC,qBAAqB,KAAK,SAAS;oBACzC,CAAC,CAAC,EAAE;oBACJ,CAAC,CAAC,EAAE,uBAAuB,EAAE,GAAG,CAAC,qBAAqB,EAAE,CAAC;gBAC3D,0BAA0B,EAAE,GAAG,CAAC,wBAAwB;aACzD,CAAA;QACH,CAAC,CAAC;QACJ,CAAC,CAAC,EAAE,CAAA;IACN,MAAM,cAAc,GAAG,KAAK,CAAC,cAAc,IAAI,OAAO,KAAK,CAAC,cAAc,KAAK,QAAQ;WAClF,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,cAAc,CAAC;QACvC,CAAC,CAAC,KAAK,CAAC,cAAyC;QACjD,CAAC,CAAC,EAAE,CAAA;IACN,MAAM,uBAAuB,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAC;QAC7E,CAAC,CAAC,QAAQ,CAAC,uBAAuB;QAClC,CAAC,CAAC,EAAE,CAAA;IACN,MAAM,SAAS,GAAG,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC;QACjD,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE;YAClC,MAAM,GAAG,GAAG,QAAmC,CAAA;YAC/C,MAAM,EAAE,WAAW,EAAE,GAAG,gBAAgB,EAAE,GAAG,GAAG,CAAA;YAChD,OAAO;gBACL,GAAG,gBAAgB;gBACnB,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC5E,CAAA;QACH,CAAC,CAAC;QACJ,CAAC,CAAC,EAAE,CAAA;IACN,OAAO;QACL,KAAK,EAAE;YACL,EAAE,EAAE,QAAQ,CAAC,EAAE;YACf,MAAM,EAAE,QAAQ,CAAC,MAAM;YACvB,gBAAgB,EAAE,eAAe;YACjC,eAAe,EAAE,QAAQ,CAAC,MAAM,IAAI,wDAAwD;YAC5F,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtF,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,KAAK,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC;gBAClE,CAAC,CAAC,EAAE,WAAW,EAAE,EAAE,WAAW,EAAE,KAAK,CAAC,WAAW,EAAE,EAAE;gBACrD,CAAC,CAAC,EAAE,CAAC;YACP,OAAO,EAAE,kBAAkB,KAAK,KAAK,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,WAAW;YACvE,MAAM,EAAE;gBACN,MAAM,EAAE,UAAU;gBAClB,mBAAmB,EAAE,kBAAkB;gBACvC,GAAG,CAAC,uBAAuB,CAAC,MAAM,GAAG,CAAC;oBACpC,CAAC,CAAC,EAAE,yBAAyB,EAAE,uBAAuB,EAAE;oBACxD,CAAC,CAAC,EAAE,CAAC;gBACP,GAAG,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC/C;SACF;QACD,SAAS,EAAE,QAAQ,CAAC,KAAK,CAAC,aAAa;QACvC,UAAU,EAAE,YAAY;QACxB,SAAS,EAAE,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,UAAU;QACjF,gBAAgB,EAAE,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,IAAI,EAAE;QAC3D,UAAU;QACV,GAAG,CAAC,KAAK,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,EAAE,iBAAiB,CAAC;YAClD,CAAC,CAAC,EAAE,eAAe,EAAE,KAAK,CAAC,eAAe,EAAE;YAC5C,CAAC,CAAC,EAAE,CAAC;KACR,CAAA;AACH,CAAC;AAED,SAAS,YAAY,CAAC,QAA+B,EAAE,QAA0B;IAC/E,IAAI,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,SAAS;QAAE,OAAO,mBAAmB,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAA;IACrF,IAAI,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,cAAc;QAAE,OAAO,6BAA6B,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAA;IACpG,OAAO,QAAQ,CAAA;AACjB,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,OAO7B;IAMC,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAC3B,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,EAClC,MAAM,EACN,OAAO,CAAC,MAAM,EACd,OAAO,CAAC,OAAO,EACf,WAAW,OAAO,CAAC,OAAO,EAAE,CAC7B,CAAA;IACD,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,WAAW,EAAE,SAAS,CAAC,CAAA;IACpE,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,WAAW,EAAE,MAAM,CAAC,CAAA;IAChE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAA;IAClD,MAAM,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAA;IAC7C,MAAM,KAAK,CAAC,WAAW,CAAC,CAAA;IACxB,IAAI,CAAC;QACH,MAAM,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC,CAAA;QAChD,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE,aAAa,EAAE;YAC7E,WAAW,EAAE,KAAK;YAClB,SAAS,EAAE,KAAK;YAChB,YAAY,EAAE,IAAI;SACnB,CAAC,CAAA;QACF,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC,EAAE,YAAY,EAAE;YACzE,WAAW,EAAE,KAAK;YAClB,SAAS,EAAE,KAAK;YAChB,YAAY,EAAE,IAAI;SACnB,CAAC,CAAA;QACF,MAAM,KAAK,CAAC,SAAS,CAAC,CAAA;IACxB,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,CAAA;QAC5B,MAAM,KAAK,CAAA;IACb,CAAC;IACD,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,KAAK,SAAS;QAC1C,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,aAAa;QAChC,CAAC,CAAC,OAAO,CAAC,OAAO,KAAK,kBAAkB;YACtC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,cAAc;YACjC,CAAC,CAAC,IAAI,CAAA;IACV,IAAI,MAAM,EAAE,CAAC;QACX,IAAI,CAAC;YACH,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAC3B,aAAa,EACb,SAAS,EACT,QAAQ,EACR,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,aAAa,CACrC,CAAA;YACD,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;gBACrC,MAAM,IAAI,KAAK,CAAC,mEAAmE,CAAC,CAAA;YACtF,CAAC;YACD,MAAM,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAA;YAC7C,MAAM,EAAE,CAAC,IAAI,CAAC,MAAM,EAAE,WAAW,EAAE;gBACjC,WAAW,EAAE,KAAK;gBAClB,SAAS,EAAE,KAAK;gBAChB,YAAY,EAAE,IAAI;aACnB,CAAC,CAAA;QACJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,CAAA;YAC5B,MAAM,KAAK,CAAA;QACb,CAAC;IACH,CAAC;IACD,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE,YAAY,EAAE,SAAS,EAAE,CAAA;AAChE,CAAC;AAED,KAAK,UAAU,sBAAsB,CACnC,UAA8B,EAC9B,QAA2B,EAC3B,UAAkB;IAElB,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,EAAE,YAAY,CAAC,CAAA;IAChE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,WAAW,CAAC,CAAA;IAC9C,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,CAAA;IAC3D,MAAM,cAAc,GAAG,UAAU,CAAC,KAAK,CAAC,cAAc;QACpD,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,SAAS,EAAE,UAAU,CAAC;QACxC,CAAC,CAAC,IAAI,CAAA;IACR,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,YAAY,CAAC,CAAA;IAC3C,MAAM,KAAK,CAAC,IAAI,CAAC,CAAA;IACjB,IAAI,CAAC;QACH,MAAM,KAAK,CAAC,SAAS,CAAC,CAAA;QACtB,MAAM,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC,CAAA;QACvC,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,SAAS,CAAC,EAAE;YAC/F,WAAW,EAAE,KAAK;YAClB,SAAS,EAAE,KAAK;YAChB,YAAY,EAAE,IAAI;SACnB,CAAC,CAAA;QACF,MAAM,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,EAAE,MAAM,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,CAAC,EAAE;YACzF,WAAW,EAAE,KAAK;YAClB,SAAS,EAAE,KAAK;YAChB,YAAY,EAAE,IAAI;SACnB,CAAC,CAAA;QACF,MAAM,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,MAAM,EAAE,aAAa,EAAE;YACpD,WAAW,EAAE,KAAK;YAClB,SAAS,EAAE,KAAK;YAChB,YAAY,EAAE,IAAI;SACnB,CAAC,CAAA;QACF,MAAM,SAAS,CAAC,KAAK,EAAE,UAAU,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAA;QAC/D,IAAI,UAAU,CAAC,KAAK,CAAC,cAAc,IAAI,cAAc,EAAE,CAAC;YACtD,MAAM,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,cAAc,EAAE,cAAc,EAAE;gBAC7D,WAAW,EAAE,KAAK;gBAClB,SAAS,EAAE,KAAK;gBAChB,YAAY,EAAE,IAAI;aACnB,CAAC,CAAA;QACJ,CAAC;QACD,MAAM,eAAe,GAAsB;YACzC,SAAS,EAAE,MAAM,+BAA+B,CAAC,SAAS,CAAC;YAC3D,MAAM,EAAE,MAAM,kBAAkB,CAAC,aAAa,CAAC;YAC/C,KAAK,EAAE,MAAM,CAAC,MAAM,QAAQ,CAAC,KAAK,CAAC,CAAC;YACpC,cAAc,EAAE,cAAc,CAAC,CAAC,CAAC,MAAM,kBAAkB,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,IAAI;SACjF,CAAA;QACD,IAAI,CAAC,WAAW,CAAC,QAAQ,EAAE,eAAe,CAAC,EAAE,CAAC;YAC5C,MAAM,IAAI,sBAAsB,CAAC,4DAA4D,CAAC,CAAA;QAChG,CAAC;QACD,sBAAsB,CAAC,UAAU,EAAE,eAAe,CAAC,CAAA;QACnD,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,aAAa,EAAE,cAAc,EAAE,KAAK,EAAE,CAAA;IAClE,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;QACrB,MAAM,KAAK,CAAA;IACb,CAAC;AACH,CAAC;AAED,KAAK,UAAU,uBAAuB,CACpC,UAA8B,EAC9B,QAA0B,EAC1B,QAA2B;IAE3B,MAAM,OAAO,GAAsB;QACjC,SAAS,EAAE,MAAM,+BAA+B,CAAC,QAAQ,CAAC,SAAS,CAAC;QACpE,MAAM,EAAE,MAAM,kBAAkB,CAAC,QAAQ,CAAC,aAAa,CAAC;QACxD,KAAK,EAAE,MAAM,CAAC,MAAM,QAAQ,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QAC7C,cAAc,EAAE,QAAQ,CAAC,cAAc;YACrC,CAAC,CAAC,MAAM,kBAAkB,CAAC,QAAQ,CAAC,cAAc,CAAC;YACnD,CAAC,CAAC,IAAI;KACT,CAAA;IACD,IAAI,CAAC,WAAW,CAAC,QAAQ,EAAE,OAAO,CAAC,IAAI,CAAC,gBAAgB,CAAC,UAAU,EAAE,OAAO,CAAC,EAAE,CAAC;QAC9E,MAAM,IAAI,sBAAsB,CAC9B,0EAA0E,CAC3E,CAAA;IACH,CAAC;AACH,CAAC;AAED,SAAS,SAAS,CAAC,KAAc;IAC/B,OAAO,oBAAoB,CAAC,IAAI,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAA;AAC1F,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAc;IACtC,OAAO,KAAK,YAAY,2BAA2B;QACjD,mCAAmC,CAAC,IAAI,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAA;AACpG,CAAC;AAED,SAAS,UAAU;IACjB,OAAO,IAAI,KAAK,CAAC,6BAA6B,CAAC,CAAA;AACjD,CAAC;AAED,KAAK,UAAU,iBAAiB,CAC9B,UAAkB,EAClB,MAAkB,EAClB,OAA0B;IAE1B,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAC5D,IAAI,CAAC,UAAU,CAAC,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IAC1G,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,UAAU,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAA;IACxF,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAA;IACjE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,WAAW,UAAU,EAAE,MAAM,CAAC,CAAA;IACpF,MAAM,SAAS,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAA;IAC9D,IAAI,CAAC;QACH,IAAI,MAAM,EAAE,CAAC,UAAU,CAAC,UAAU,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAA;QAChG,MAAM,MAAM,CAAC,SAAS,EAAE,UAAU,CAAC,CAAA;IACrC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAA;QACjD,MAAM,KAAK,CAAA;IACb,CAAC;IACD,OAAO,MAAM,CAAC,KAAK,CAAC,CAAA;AACtB,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,QAAgB,EAChB,OAA2B;IAE3B,IAAI,CAAC,OAAO,EAAE,SAAS;QAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAA;IAC9E,MAAM,cAAc,GAAG,MAAM,oBAAoB,CAAC,QAAQ,CAAC,CAAA;IAC3D,IAAI,CAAC,cAAc,CAAC,KAAK,IAAI,CAAC,cAAc,CAAC,UAAU,EAAE,CAAC;QACxD,MAAM,IAAI,KAAK,CAAC,wBAAwB,cAAc,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IAC7E,CAAC;IACD,MAAM,UAAU,GAAG,cAAc,CAAC,UAAU,CAAA;IAC5C,MAAM,QAAQ,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAA;IACxE,MAAM,UAAU,GAAG,MAAM,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,KAAK,CAAC,CAAA;IACzD,IAAI,MAAM,CAAC,UAAU,CAAC,KAAK,UAAU,CAAC,UAAU,EAAE,CAAC;QACjD,MAAM,IAAI,sBAAsB,CAAC,mDAAmD,CAAC,CAAA;IACvF,CAAC;IACD,MAAM,QAAQ,GAAG,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAA;IAC5E,IAAI,QAAQ,CAAC,KAAK,CAAC,aAAa,KAAK,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;QAClE,MAAM,IAAI,KAAK,CAAC,oEAAoE,CAAC,CAAA;IACvF,CAAC;IACD,2BAA2B,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;IAC1E,MAAM,eAAe,GAAG,QAAQ,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;IAC/D,MAAM,eAAe,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,SAAS,EAAE;QACrE,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,eAAe;KAChB,CAAC,CAAA;IACF,IAAI,CAAC,eAAe,CAAC,KAAK,IAAI,CAAC,eAAe,CAAC,UAAU,EAAE,CAAC;QAC1D,MAAM,IAAI,KAAK,CAAC,uCAAuC,eAAe,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IAC7F,CAAC;IACD,MAAM,aAAa,GAAG,uBAAuB,CAC3C,UAAU,EACV,eAAe,CAAC,UAAU,CAAC,WAAW,EACtC,QAAQ,CACT,CAAA;IACD,MAAM,QAAQ,GAAG,MAAM,YAAY,CAAC,UAAU,CAAC,CAAA;IAC/C,sBAAsB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;IAC5C,MAAM,qBAAqB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;IACjD,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS;WAC9B,OAAO,CAAC,2BAA2B,EAAE,SAAS;WAC9C,OAAO,CAAC,GAAG,CAAA;IAChB,MAAM,OAAO,GAAG,MAAM,6BAA6B,CAAC;QAClD,SAAS;QACT,iBAAiB,EAAE,IAAI;QACvB,aAAa,EAAE,OAAO,CAAC,2BAA2B,EAAE,KAAK;eACpD,SAAS,CAAC,yBAAyB;KACzC,CAAC,CAAA;IACF,MAAM,OAAO,GAAG,GAAa,EAAE,CAAC;QAC9B,GAAG,YAAY,CAAC,SAAS,CAAC;QAC1B,GAAG,OAAO,CAAC,eAAe;KAC3B,CAAA;IACD,MAAM,2BAA2B,GAAoC;QACnE,GAAG,OAAO,CAAC,2BAA2B;QACtC,SAAS;QACT,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,SAAS;QACjC,YAAY,EAAE,OAAO;KACtB,CAAA;IACD,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAA;IAC1C,MAAM,IAAI,GAAoB,EAAE,CAAA;IAChC,IAAI,IAAI,GAAG,KAAK,CAAA;IAChB,MAAM,qBAAqB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;IACjD,MAAM,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAA;IAC5D,MAAM,KAAK,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC,CAAA;IACvC,IAAI,QAA0B,CAAA;IAC9B,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,sBAAsB,CAAC,UAAU,EAAE,QAAQ,EAAE,UAAU,CAAC,CAAA;IAC3E,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,EAAE,CAAC,MAAM,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC,CAAA;QAC3C,MAAM,KAAK,CAAA;IACb,CAAC;IACD,KAAK,MAAM,QAAQ,IAAI,QAAQ,CAAC,KAAK,EAAE,CAAC;QACtC,IAAI,IAAI;YAAE,MAAK;QACf,KAAK,MAAM,OAAO,IAAI,aAAa,CAAC,eAAe,EAAE,CAAC;YACpD,IAAI,IAAI;gBAAE,MAAK;YACf,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC;gBACpE,MAAM,cAAc,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;gBACjC,IAAI,MAAM,GAA4B,QAAQ,CAAA;gBAC9C,IAAI,KAAK,GAA2B,IAAI,CAAA;gBACxC,IAAI,KAAK,GAA2B,IAAI,CAAA;gBACxC,IAAI,KAAK,GAA2B,IAAI,CAAA;gBACxC,IAAI,SAAqC,CAAA;gBACzC,IAAI,SAAS,GAAG,IAAI,CAAC,IAAI,CACvB,UAAU,CAAC,KAAK,CAAC,SAAS,EAC1B,MAAM,EACN,QAAQ,CAAC,EAAE,EACX,OAAO,EACP,WAAW,OAAO,EAAE,EACpB,QAAQ,CACT,CAAA;gBACD,IAAI,CAAC;oBACH,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO;wBAAE,MAAM,UAAU,EAAE,CAAA;oBAC/C,MAAM,qBAAqB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;oBACjD,MAAM,uBAAuB,CAAC,UAAU,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAA;oBAC7D,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC;wBACpC,UAAU;wBACV,QAAQ;wBACR,QAAQ;wBACR,MAAM,EAAE,QAAQ,CAAC,EAAE;wBACnB,OAAO;wBACP,OAAO;qBACR,CAAC,CAAA;oBACF,SAAS,GAAG,QAAQ,CAAC,SAAS,CAAA;oBAC9B,MAAM,KAAK,GAAyB;wBAClC,IAAI,EAAE,YAAY,CAAC,QAAQ,EAAE,QAAQ,CAAC;wBACtC,OAAO;wBACP,OAAO;wBACP,aAAa,EAAE,QAAQ,CAAC,aAAa;wBACrC,YAAY,EAAE,QAAQ,CAAC,YAAY;wBACnC,SAAS,EAAE,QAAQ,CAAC,SAAS;wBAC7B,SAAS,EAAE,UAAU,CAAC,IAAI,CAAC,SAAS;wBACpC,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;qBACtD,CAAA;oBACD,MAAM,cAAc,GAAG,OAAO,UAAU,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;wBACjE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,kBAAkB,IAAI,sBAAsB,CAAC,CAC1D,UAAU,CAAC,IAAI,CAAC,QAAQ,EACxB,EAAE,GAAG,KAAK,EAAE,QAAQ,EAAE,QAAQ,CAAC,WAAW,EAAE,EAC5C,2BAA2B,CAC5B;wBACH,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,gBAAgB;4BAClD,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,qBAAqB,IAAI,4BAA4B,CAAC,CAAC;gCACpE,OAAO,EAAE,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAgC;gCAClE,QAAQ,EAAE,QAAQ,CAAC,WAAW;gCAC9B,KAAK;gCACL,SAAS;gCACT,YAAY,EAAE,OAAO;6BACtB,CAAC;4BACJ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,4BAA4B,IAAI,4BAA4B,CAAC,CAAC;gCAC3E,QAAQ,EAAE;oCACR,GAAG,UAAU,CAAC,IAAI,CAAC,QAAQ;oCAC3B,OAAO,EAAE,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAgC;iCACnE;gCACD,QAAQ,EAAE,QAAQ,CAAC,WAAW;gCAC9B,KAAK;gCACL,SAAS;gCACT,YAAY,EAAE,OAAO;gCACrB,QAAQ,EAAE,2BAA2B,CAAC,QAAQ;6BAC/C,CAAC,CAAA;oBACR,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO;wBAAE,MAAM,UAAU,EAAE,CAAA;oBAC/C,MAAM,SAAS,GAAG,WAAW,CAAC,cAAc,EAAE,OAAO,EAAE,CAAC,CAAA;oBACxD,KAAK,GAAG,SAAS,CAAC,KAAK,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,GAAG,SAAS,CAAC,KAAK,EAAE,CAAA;oBAChE,KAAK,GAAG,SAAS,CAAC,KAAK,CAAA;oBACvB,SAAS,GAAG,OAAO,UAAU,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;2BACnD,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,gBAAgB;2BAClD,SAAS,CAAC,SAAS;wBACtB,CAAC,CAAC;4BACE,GAAG,SAAS,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,QAAQ,EAAE,CAAC,CAAC;4BAC3D,EAAE,IAAI,EAAE,sBAAsB,EAAE,IAAI,EAAE,QAAQ,EAAE;yBACjD;wBACH,CAAC,CAAC,SAAS,CAAA;oBACb,MAAM,SAAS,CACb,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,EAAE,sBAAsB,CAAC,EACrD,GAAG,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EACzC,EAAE,IAAI,EAAE,IAAI,EAAE,CACf,CAAA;oBACD,MAAM,GAAG,WAAW,CAAA;gBACtB,CAAC;gBAAC,OAAO,MAAM,EAAE,CAAC;oBAChB,MAAM,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAA;oBACjD,KAAK,GAAG,IAAI,CAAA;oBACZ,KAAK,GAAG,IAAI,CAAA;oBACZ,SAAS,GAAG,SAAS,CAAA;oBACrB,KAAK,GAAG,EAAE,OAAO,EAAE,UAAU,CAAC,MAAM,YAAY,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,CAAC,IAAI,kBAAkB,EAAE,CAAA;oBAC3H,IACE,OAAO,CAAC,MAAM,EAAE,OAAO;2BACpB,MAAM,YAAY,sBAAsB;2BACxC,gBAAgB,CAAC,MAAM,CAAC;2BACxB,MAAM,KAAK,SAAS;2BACpB,CACD,OAAO,UAAU,CAAC,IAAI,CAAC,QAAQ,KAAK,QAAQ;+BACzC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,gBAAgB,CACtD;wBACD,IAAI,GAAG,IAAI,CAAA;gBACf,CAAC;gBACD,IAAI,CAAC;oBACH,MAAM,qBAAqB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;oBACjD,MAAM,uBAAuB,CAAC,UAAU,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAA;gBAC/D,CAAC;gBAAC,OAAO,OAAO,EAAE,CAAC;oBACjB,MAAM,GAAG,QAAQ,CAAA;oBACjB,KAAK,GAAG,IAAI,CAAA;oBACZ,KAAK,GAAG,IAAI,CAAA;oBACZ,SAAS,GAAG,SAAS,CAAA;oBACrB,KAAK,GAAG,EAAE,OAAO,EAAE,UAAU,CAAE,OAAiB,CAAC,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,CAAA;oBACtE,IAAI,GAAG,IAAI,CAAA;gBACb,CAAC;gBACD,IAAI,CAAC,IAAI,CAAC;oBACR,MAAM,EAAE,QAAQ,CAAC,EAAE;oBACnB,OAAO;oBACP,OAAO;oBACP,MAAM;oBACN,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,cAAc,CAAC;oBACpD,SAAS,EAAE,eAAe,CAAC,QAAQ,EAAE,SAAS,CAAC;oBAC/C,KAAK;oBACL,KAAK;oBACL,KAAK;oBACL,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;iBACpC,CAAC,CAAA;gBACF,IAAI,IAAI;oBAAE,MAAK;YACjB,CAAC;QACH,CAAC;IACH,CAAC;IACD,IAAI,CAAC;QACH,MAAM,qBAAqB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;QACjD,MAAM,uBAAuB,CAAC,UAAU,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAA;IAC/D,CAAC;IAAC,OAAO,OAAO,EAAE,CAAC;QACjB,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;YAAE,MAAM,OAAO,CAAA;QACpC,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAE,CAAA;QACnC,IAAI,CAAC,MAAM,GAAG,QAAQ,CAAA;QACtB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAA;QACjB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAA;QACjB,OAAO,IAAI,CAAC,SAAS,CAAA;QACrB,IAAI,CAAC,KAAK,GAAG,EAAE,OAAO,EAAE,UAAU,CAAE,OAAiB,CAAC,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,CAAA;IAC7E,CAAC;IACD,IAAI,OAAO,CAAC,MAAM,EAAE,OAAO,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAE,CAAA;QACnC,IAAI,CAAC,MAAM,GAAG,QAAQ,CAAA;QACtB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAA;QACjB,IAAI,CAAC,KAAK,GAAG,IAAI,CAAA;QACjB,OAAO,IAAI,CAAC,SAAS,CAAA;QACrB,IAAI,CAAC,KAAK,GAAG,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC,OAAO,EAAE,CAAA;IAChD,CAAC;IACD,MAAM,MAAM,GAAe;QACzB,OAAO,EAAE,CAAC;QACV,UAAU,EAAE,UAAU,CAAC,UAAU;QACjC,WAAW,EAAE,eAAe,CAAC,UAAU,CAAC,WAAW;QACnD,YAAY,EAAE,UAAU,CAAC,YAAY;QACrC,UAAU,EAAE,UAAU,CAAC,UAAU;QACjC,QAAQ,EAAE,UAAU,CAAC,IAAI,CAAC,QAAQ;QAClC,SAAS;QACT,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACpC,IAAI;QACJ,OAAO,EAAE,wBAAwB,CAAC,IAAI,CAAC;KACxC,CAAA;IACD,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,SAAS,EAAE,aAAa,CAAC,CAAA;IACvE,MAAM,YAAY,GAAG,MAAM,iBAAiB,CAAC,UAAU,EAAE,MAAM,EAAE,aAAa,CAAC,CAAA;IAC/E,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,EAAE,CAAA;AAC1D,CAAC"}
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
import type { Digest, ValidationResult } from './types.js';
|
|
2
|
+
export type EvalProvider = 'codex-trigger' | 'codex-conversation' | {
|
|
3
|
+
type: 'custom-process';
|
|
4
|
+
command: string[];
|
|
5
|
+
} | {
|
|
6
|
+
type: 'codex-workflow';
|
|
7
|
+
command: string[];
|
|
8
|
+
allowedModels: string[];
|
|
9
|
+
};
|
|
10
|
+
export interface EvalSpecPathDigest {
|
|
11
|
+
path: string;
|
|
12
|
+
digest: Digest;
|
|
13
|
+
}
|
|
14
|
+
export type EvalBaseline = {
|
|
15
|
+
type: 'none';
|
|
16
|
+
} | {
|
|
17
|
+
type: 'without-skill';
|
|
18
|
+
} | {
|
|
19
|
+
type: 'previous-version';
|
|
20
|
+
target: EvalSpecPathDigest;
|
|
21
|
+
};
|
|
22
|
+
export interface EvalSpec {
|
|
23
|
+
version: 1;
|
|
24
|
+
source: {
|
|
25
|
+
plugin: string;
|
|
26
|
+
skill: string;
|
|
27
|
+
suite: string;
|
|
28
|
+
profile: string;
|
|
29
|
+
};
|
|
30
|
+
workspace: EvalSpecPathDigest;
|
|
31
|
+
target: EvalSpecPathDigest;
|
|
32
|
+
cases: EvalSpecPathDigest;
|
|
33
|
+
provider: EvalProvider;
|
|
34
|
+
runs: number;
|
|
35
|
+
timeoutMs: number;
|
|
36
|
+
baseline: EvalBaseline;
|
|
37
|
+
outputDir: string;
|
|
38
|
+
}
|
|
39
|
+
export interface EvalSpecCalculated {
|
|
40
|
+
spec: EvalSpec;
|
|
41
|
+
specDigest: Digest;
|
|
42
|
+
workspaceDigest: Digest;
|
|
43
|
+
providerDigest: Digest;
|
|
44
|
+
targetDigest: Digest;
|
|
45
|
+
caseDigest: Digest;
|
|
46
|
+
previousTargetDigest: Digest | null;
|
|
47
|
+
paths: {
|
|
48
|
+
workspace: string;
|
|
49
|
+
target: string;
|
|
50
|
+
cases: string;
|
|
51
|
+
previousTarget: string | null;
|
|
52
|
+
outputDir: string;
|
|
53
|
+
};
|
|
54
|
+
}
|
|
55
|
+
export declare function validateEvalSpec(value: unknown): Promise<ValidationResult<EvalSpec>>;
|
|
56
|
+
export declare function parseEvalSpec(value: unknown): Promise<EvalSpec>;
|
|
57
|
+
export declare function validateEvalSpecFile(specPath: string): Promise<ValidationResult<EvalSpecCalculated>>;
|