@leostera/evalkit 0.7.0 → 0.8.0
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/CHANGELOG.md +12 -1
- package/dist/cli.mjs +4 -4
- package/dist/dashboard/assets/index-C12IeG2y.css +2 -0
- package/dist/dashboard/assets/{index-BEBo_81Y.js → index-xlVsW6ZR.js} +14 -11
- package/dist/dashboard/index.html +2 -2
- package/dist/index.d.mts +19 -0
- package/dist/index.d.mts.map +1 -1
- package/dist/index.mjs +1 -1
- package/dist/runner.d.mts +20 -1
- package/dist/runner.d.mts.map +1 -1
- package/dist/runner.mjs +1 -1
- package/dist/{src-b2Wnmzbb.mjs → src-LeGSyDD2.mjs} +3 -0
- package/dist/{src-DhromKNu.mjs → src-mm8L_IxM.mjs} +68 -23
- package/package.json +1 -1
- package/dist/dashboard/assets/index-C24IquYZ.css +0 -2
|
@@ -4,8 +4,8 @@
|
|
|
4
4
|
<meta charset="UTF-8" />
|
|
5
5
|
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
|
|
6
6
|
<title>EvalKit</title>
|
|
7
|
-
<script type="module" crossorigin src="/assets/index-
|
|
8
|
-
<link rel="stylesheet" crossorigin href="/assets/index-
|
|
7
|
+
<script type="module" crossorigin src="/assets/index-xlVsW6ZR.js"></script>
|
|
8
|
+
<link rel="stylesheet" crossorigin href="/assets/index-C12IeG2y.css">
|
|
9
9
|
</head>
|
|
10
10
|
<body>
|
|
11
11
|
<div id="root"></div>
|
package/dist/index.d.mts
CHANGED
|
@@ -349,6 +349,7 @@ type EvalkitConfig = {
|
|
|
349
349
|
* Paths to leave out, as globs relative to the candidate workspace; a
|
|
350
350
|
* pattern without `/` matches that name at any depth. Replaces the default,
|
|
351
351
|
* `DEFAULT_ARTIFACT_EXCLUDE` (dependencies, build output, VCS metadata).
|
|
352
|
+
* A scenario's `workspace.ignore` patterns come after these.
|
|
352
353
|
*/
|
|
353
354
|
exclude?: readonly string[];
|
|
354
355
|
};
|
|
@@ -493,6 +494,8 @@ export declare const TrialMetadataSchema: Schema.Struct<{
|
|
|
493
494
|
version: Schema.optional<typeof Schema.String>;
|
|
494
495
|
}>>;
|
|
495
496
|
startedAt: typeof Schema.String;
|
|
497
|
+
/** The effective snapshot ignore patterns: the project's, then the scenario's. */
|
|
498
|
+
workspaceIgnore: Schema.optional<Schema.mutable<Schema.Array$<typeof Schema.String>>>;
|
|
496
499
|
datasets: Schema.optional<Schema.mutable<Schema.Array$<Schema.Struct<{
|
|
497
500
|
id: typeof Schema.String;
|
|
498
501
|
visibility: Schema.Literal<["candidate", "evaluator"]>;
|
|
@@ -556,6 +559,7 @@ export declare const TrialManifestSchema: Schema.Struct<{
|
|
|
556
559
|
version: Schema.optional<typeof Schema.String>;
|
|
557
560
|
}>>;
|
|
558
561
|
startedAt: typeof Schema.String;
|
|
562
|
+
workspaceIgnore: Schema.optional<Schema.mutable<Schema.Array$<typeof Schema.String>>>;
|
|
559
563
|
datasets: Schema.optional<Schema.mutable<Schema.Array$<Schema.Struct<{
|
|
560
564
|
id: typeof Schema.String;
|
|
561
565
|
visibility: Schema.Literal<["candidate", "evaluator"]>;
|
|
@@ -1232,6 +1236,19 @@ export type EvalDefinition<TAgent extends AutAdapter = AutAdapter> = {
|
|
|
1232
1236
|
policy?: EvalPolicy;
|
|
1233
1237
|
/** Any JSON-convertible object; converted when the eval is defined. */
|
|
1234
1238
|
metadata?: Record<string, unknown>;
|
|
1239
|
+
workspace?: EvalWorkspace;
|
|
1240
|
+
};
|
|
1241
|
+
/** How EvalKit treats this scenario's candidate workspace. */
|
|
1242
|
+
export type EvalWorkspace = {
|
|
1243
|
+
/**
|
|
1244
|
+
* Paths to leave out of the trial's snapshot, read like `.gitignore`: globs
|
|
1245
|
+
* relative to the workspace, where a pattern without `/` matches that name
|
|
1246
|
+
* at any depth and later patterns win. They come after the project's
|
|
1247
|
+
* `artifacts.exclude` (default `DEFAULT_ARTIFACT_EXCLUDE`), so `!pattern`
|
|
1248
|
+
* can keep a path the project leaves out, even inside an ignored directory:
|
|
1249
|
+
* `['!_build/doc/**']` keeps generated docs while dropping the rest of `_build`.
|
|
1250
|
+
*/
|
|
1251
|
+
ignore?: readonly string[];
|
|
1235
1252
|
};
|
|
1236
1253
|
export declare function defineEval<const T extends EvalDefinition>(definition: T): T;
|
|
1237
1254
|
export type EvalSuite<TEvals extends readonly EvalDefinition[] = readonly EvalDefinition[]> = {
|
|
@@ -1338,6 +1355,8 @@ export type TrialMetadata = {
|
|
|
1338
1355
|
};
|
|
1339
1356
|
aut?: AutIdentity;
|
|
1340
1357
|
startedAt: string;
|
|
1358
|
+
/** The effective snapshot ignore patterns: the project's `artifacts.exclude`, then the scenario's `workspace.ignore`. */
|
|
1359
|
+
workspaceIgnore?: string[];
|
|
1341
1360
|
/** From schema version 5: the datasets this trial used, with content digests. */
|
|
1342
1361
|
datasets?: TrialDataset[];
|
|
1343
1362
|
};
|
package/dist/index.d.mts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.mts","names":[],"sources":["../../core/src/identity.ts","../../core/src/dataset.ts","../../core/src/aut-events.ts","../../core/src/catalog.ts","../../core/src/canonical-json.ts","../../core/src/matrix.ts","../../core/src/config.ts","../../core/src/json.ts","../../core/src/definition-site.ts","../../core/src/schema.ts","../../core/src/trajectory.ts","../../core/src/index.ts","../../agents/src/index.ts"],"mappings":";;qBAEa,oBAAkB,OAAA;YACnB,sBAAsB,mBAAmB;;wBAGrC,SAAS;;qBAOZ,YAAU,OAAA,cAAA,OAAA;YAKX,cAAc,WAAW;YAEzB,YAAY,cAAc,eAAe,2BACxC;qBAEA,mBAAiB,OAAA,cAAA,OAAA;wBAMd,kBAAkB,cAAc,cAAc,MAAM,QAAK,OAAA,cAAA,OAAA;wBAWzD,YAAY,cAAc,cACxC,MAAM,OACN,MAAM,OACL,YAAY;wBAIC,iBAAiB,cAAc,cAC7C,eACA,eAAe;EACZ,MAAM;EAAO,MAAM;EAAM,KAAK,YAAY;;wBAY/B,aAAa,eAAe,eAAe,eAAe;;;;YCvD9D;;YAGA,6BAA6B;EAAe;;;;;;YAM5C,wBAAwB,SAAS,eAAe;;YAGhD;EACV,QAAQ;;EAER;;YAGU;;EAEV;EACA,QAAQ;;YAGE;EACV;EACA;EACA;EACA;;EAEA;;EAEA,YAAY;EACZ,QAAQ;;KAGL;EACH;EACA;;EAEA,aAAa;;EAEb;;YAGU,0BAA0B;EACpC,QAAQ;;EAER;;EAEA;EACA;EACA;;YAGU,0BAA0B,gBAAgB,kBACpD;;EAEE,SAAS,MAAM,uBAAuB,UAAU,QAAQ;;EAExD,SACE,QAAQ,SACR,OAAO,wBACJ,iBAAiB,QAAQ;EAC9B;EACA;EACA;;YAGQ;EAEN;EACA,QAAQ;EACR;EACA;;EAGA;EACA,SACE,MAAM,uBACH,iBAAiB,QAAQ;EAC9B,SACE,QAAQ,gBACR,OAAO,wBACJ,iBAAiB,QAAQ;;;YAIxB;WACD;WACA;WACA;WACA,YAAY;;WAEZ;WACA,QAAQ;;;wBAIH,oBAAoB,eAAe;;;;;;;wBAuBnC,cAAc,gBAAgB,gBAC5C,YAAY,0BAA0B,0BAA0B,WAC/D;wBAmFa,UAAU,iBAAiB,SAAS;;;;;;;YC/MxC;EACN;EAAiB;;EAEjB;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB;EAAkB;;EACrC;EAAe,OAAO;EAAe;;YAE/B;EACV;EACA;;EAEA;EACA;;;YAIU;EAGN;EACA,qBAAqB;EACrB,QAAQ;;;wBAIE,kBACd,OAAO,eACP,WAAW,OACV;;;;;wBA0Da,iBACd,QAAQ,YACR,cACA,WAAW,OACV;;;;;;;;;YCtGS;;EAEV;;YAGU;EACV;EACA;EACA;EACA;EACA,UAAU;IAAQ,MAAM;IAAkB;;;EAE1C;;;YAIU;EACN;EAAe;;EACf;EAAgB;;EAChB;;YAEM;EACV;EACA;EACA,YAAY;EACZ;EACA;;EAEA;IAAU;;IAAwB;;;EAElC;IAAU,OAAO;;IAAiB;;;EAElC;IAAU;IAAc;;EACxB;;YAGU;EACV;EACA,MAAM;EACN;EACA;;EAEA;EACA;;YAGU;;;EAEN;EAAc;;EACd;EAAe,aAAa;;EAC5B;EAA0B;EAAc,YAAY;;;;EAEpD;EAAoB,QAAQ;;YAEtB;EACV;EACA;EACA;EACA;EACA,OAAO;;EAEP,aAAa;EACb,UAAU;EACV;EACA;IAAU;IAAgB;IAAoB;;EAC9C,YAAY;EACZ,SAAS;EACT,WAAW;;EAEX;;wBAGc,aACd,YAAY,gBACZ,OAAO,uBACP,UAAS,iBACR;;;;wBC5Fa,oBAAoB,OAAO;;;KCK/B;;EAEV;;EAEA,aAAa,SAAS,wBAAwB;;EAE9C,YAAY;;KAEF;EACV;;EAEA;EACA,MAAM;EACN,YAAY;;KAEF;EACV;EACA;EACA,gBAAgB;EAChB,YAAY,SAAS,wBAAwB;EAC7C,WAAW;;EAEX,iBAAiB;;EAEjB,mBAAmB;;KAET,aAAa;EACvB;EACA,MAAM,YAAY;EAClB,MAAM,YAAY,kBAAkB,iBAAiB;;wBAWvC,uBAAuB,UAAU,sBAC/C,YAAY,IACX,aAAa;;;KCjDJ;;EAEV;EACA;EACA;EACA;;IAEE;;;EAGF,iBAAiB;EACjB,WAAW;;EAEX,SAAS,KAAK;IAA0C;;EACxD;IAAc;IAAsB;IAAiB;;EACrD;;EAEA
|
|
1
|
+
{"version":3,"file":"index.d.mts","names":[],"sources":["../../core/src/identity.ts","../../core/src/dataset.ts","../../core/src/aut-events.ts","../../core/src/catalog.ts","../../core/src/canonical-json.ts","../../core/src/matrix.ts","../../core/src/config.ts","../../core/src/json.ts","../../core/src/definition-site.ts","../../core/src/schema.ts","../../core/src/trajectory.ts","../../core/src/index.ts","../../agents/src/index.ts"],"mappings":";;qBAEa,oBAAkB,OAAA;YACnB,sBAAsB,mBAAmB;;wBAGrC,SAAS;;qBAOZ,YAAU,OAAA,cAAA,OAAA;YAKX,cAAc,WAAW;YAEzB,YAAY,cAAc,eAAe,2BACxC;qBAEA,mBAAiB,OAAA,cAAA,OAAA;wBAMd,kBAAkB,cAAc,cAAc,MAAM,QAAK,OAAA,cAAA,OAAA;wBAWzD,YAAY,cAAc,cACxC,MAAM,OACN,MAAM,OACL,YAAY;wBAIC,iBAAiB,cAAc,cAC7C,eACA,eAAe;EACZ,MAAM;EAAO,MAAM;EAAM,KAAK,YAAY;;wBAY/B,aAAa,eAAe,eAAe,eAAe;;;;YCvD9D;;YAGA,6BAA6B;EAAe;;;;;;YAM5C,wBAAwB,SAAS,eAAe;;YAGhD;EACV,QAAQ;;EAER;;YAGU;;EAEV;EACA,QAAQ;;YAGE;EACV;EACA;EACA;EACA;;EAEA;;EAEA,YAAY;EACZ,QAAQ;;KAGL;EACH;EACA;;EAEA,aAAa;;EAEb;;YAGU,0BAA0B;EACpC,QAAQ;;EAER;;EAEA;EACA;EACA;;YAGU,0BAA0B,gBAAgB,kBACpD;;EAEE,SAAS,MAAM,uBAAuB,UAAU,QAAQ;;EAExD,SACE,QAAQ,SACR,OAAO,wBACJ,iBAAiB,QAAQ;EAC9B;EACA;EACA;;YAGQ;EAEN;EACA,QAAQ;EACR;EACA;;EAGA;EACA,SACE,MAAM,uBACH,iBAAiB,QAAQ;EAC9B,SACE,QAAQ,gBACR,OAAO,wBACJ,iBAAiB,QAAQ;;;YAIxB;WACD;WACA;WACA;WACA,YAAY;;WAEZ;WACA,QAAQ;;;wBAIH,oBAAoB,eAAe;;;;;;;wBAuBnC,cAAc,gBAAgB,gBAC5C,YAAY,0BAA0B,0BAA0B,WAC/D;wBAmFa,UAAU,iBAAiB,SAAS;;;;;;;YC/MxC;EACN;EAAiB;;EAEjB;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB;EAAkB;;EACrC;EAAe,OAAO;EAAe;;YAE/B;EACV;EACA;;EAEA;EACA;;;YAIU;EAGN;EACA,qBAAqB;EACrB,QAAQ;;;wBAIE,kBACd,OAAO,eACP,WAAW,OACV;;;;;wBA0Da,iBACd,QAAQ,YACR,cACA,WAAW,OACV;;;;;;;;;YCtGS;;EAEV;;YAGU;EACV;EACA;EACA;EACA;EACA,UAAU;IAAQ,MAAM;IAAkB;;;EAE1C;;;YAIU;EACN;EAAe;;EACf;EAAgB;;EAChB;;YAEM;EACV;EACA;EACA,YAAY;EACZ;EACA;;EAEA;IAAU;;IAAwB;;;EAElC;IAAU,OAAO;;IAAiB;;;EAElC;IAAU;IAAc;;EACxB;;YAGU;EACV;EACA,MAAM;EACN;EACA;;EAEA;EACA;;YAGU;;;EAEN;EAAc;;EACd;EAAe,aAAa;;EAC5B;EAA0B;EAAc,YAAY;;;;EAEpD;EAAoB,QAAQ;;YAEtB;EACV;EACA;EACA;EACA;EACA,OAAO;;EAEP,aAAa;EACb,UAAU;EACV;EACA;IAAU;IAAgB;IAAoB;;EAC9C,YAAY;EACZ,SAAS;EACT,WAAW;;EAEX;;wBAGc,aACd,YAAY,gBACZ,OAAO,uBACP,UAAS,iBACR;;;;wBC5Fa,oBAAoB,OAAO;;;KCK/B;;EAEV;;EAEA,aAAa,SAAS,wBAAwB;;EAE9C,YAAY;;KAEF;EACV;;EAEA;EACA,MAAM;EACN,YAAY;;KAEF;EACV;EACA;EACA,gBAAgB;EAChB,YAAY,SAAS,wBAAwB;EAC7C,WAAW;;EAEX,iBAAiB;;EAEjB,mBAAmB;;KAET,aAAa;EACvB;EACA,MAAM,YAAY;EAClB,MAAM,YAAY,kBAAkB,iBAAiB;;wBAWvC,uBAAuB,UAAU,sBAC/C,YAAY,IACX,aAAa;;;KCjDJ;;EAEV;EACA;EACA;EACA;;IAEE;;;EAGF,iBAAiB;EACjB,WAAW;;EAEX,SAAS,KAAK;IAA0C;;EACxD;IAAc;IAAsB;IAAiB;;EACrD;;EAEA;;;;;;;IAOE;;;EAGF;;EAEA,gBAAgB;EAChB,QAAQ;;KAGE;;qBAGC;;KAWD,eAAe;KAEf;;EAEV;;;;;;EAMA,UAAU,sBAAsB,eAAe,QAAQ;;;wBAIzC,mBAAmB,UAAU,eAC3C,eAAe,IACd;;;;qBC/DU,qBAAqB;WAGrB;WACA;WAHO;EAClB,YACW,cACA;;;;;;;;wBAYG,OAAO,gBAAgB,gBAAiB;;wBA4CxC,aACd,gBACA;GACI,cAAc;;;;;;;;;;;KC3DR;EAAmB;EAAc;;wBAqB7B,eAAe,iBAAiB;;;;qBCvBnC,iBAAe,OAAA;qBAOf,qBAAmB,OAAA;;;;;qBAMnB,mBAAiB,OAAA;;;;;;qBAQjB,iBAAiB,OAAO,OAAO;qBAoB/B,iBAAe,OAAA;;;;;;;;;;;;;;qBAiBf,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAejB,qBAAmB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBA2BnB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAIjB,qBAAmB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAKnB,aAAW,OAAA;;;;;qBAMX,oBAAkB,OAAA;;;;;;;;;;;;;;qBAalB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAYjB,wBAAsB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAgBtB,oBAAkB,OAAA;;;;;;;;;;;;qBAQlB,oBAAkB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAUlB,qBAAmB,OAAA;;;;;qBAKnB,oBAAkB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBASlB,kBAAgB,OAAA;;;;;;;;;;;;;;qBAmJhB,uBAAuB,OAAO,OAAO;qBAGrC,gBAAc,OAAA;;;;;;;wBClVX,qBAAqB,mBAAmB;;;YCgB5C;YACA,YACV,gBAAgB;GAAiB,cAAc;;YACrC;GAAgB,cAAc;;YAE9B;;EAEV;EACA;EACA;EACA;;YAGU;;;;;YAMA;EACV;EACA,gBAAgB;;YAGN,gBAAgB,QAAQ,OAAO,kBAAkB;YAEjD;EACV;EACA;EACA;;YAGU;EACV;EACA;EACA;;YAGU;EACN;EAAiB;;EAEjB;EACA;EACA,SAAS;EACT;;EAGA;EACA;EACA;EACA,WAAW;EACX;;EAGA;EACA;EACA;EACA,QAAQ;EACR;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB,SAAS;EAAW;;EACvC;EAAe,OAAO;EAAe;;YAE/B;EACN;EAAuB;;EAEvB;EACA;;EAEA;EACA;;EAEA;EAAmC;EAAc;;EACjD;EAAiC;EAAc;;EAE/C;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAwB;EAAgB;;EAExC;EACA;;EAEA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAyB;;EACzB;EAAe,OAAO;EAAe;;YAE/B,mBACT;EAAa;MAAoB;EAAgB;;YAExC;;EAEV;;YAGU,qBAAqB;YACrB,qBAAqB;YAErB;EACV;EACA;EACA;EACA;EACA,UAAU;;EAEV,aAAa;EACb,UAAU;;EAEV,MAAM;;EAEN,WAAW;;EAEX,oBAAoB;;YAGV,WAAW;WACZ,WAAW;;WAEX,WAAW;EACpB,MAAM;IACJ,SAAS;IACT,UAAU;;IAEV,UAAU,OAAO,yBAAyB;MACxC,QAAQ,WAAW;;YAGb,WAAW;;;;;EAKrB,KAAK,SAAS,WAAW,eAAe;;EAExC,UAAU;;;qBAIC,wBAAwB;;wBAKrB,cAAc,OAAO,kBAAkB;;;;;wBAUvC,kBAAkB,UAAU,YAAY,OAAO,IAAI;;YAMvD;;EAEV,YAAY;;EAEZ,MAAM;;EAEN;EACA;;YAGU;EACV;;EAEA,oBAAoB,OAAO;;YAGjB;EACV;EACA,aAAa;;YAGH;EACV;EACA,kBAAkB;IAAa;;EAC/B,6BAA6B,QAAQ;IAAY;;IAC/C;IACA;;EAEF;IACE;IACA;IACA;IACA,WAAW;IACX;IACA,SAAS;;EAEX;;YAGU;EACV;EACA;EACA;IAAY;IAAc,YAAY;;;YAE5B,iBACV,WAAW,YAAY,cAAc;;wBAGvB,eACd,UAAU,iCACT;;;;;wBAUa,KAAK,SAAS,sBAAsB;wBAIpC,MAAM,aAAa,aAAa;YAIpC;EACV,QAAQ;EACR,QAAQ;;EAER,SAAS;;;YAIC;EAAiB;EAAe;;YAEhC,sBAER;EAEE;EACA;EACA;;EAEA;;EAEA,QAAQ;;;;;;wBAOE,KAAK,iBAAiB;YAK1B;EACV;EACA;EACA;EACA;EACA;EACA,WAAW;EACX,QAAQ;EACR;EACA,QAAQ;;EAER;;YAGU;EACV,WAAW;EACX,WAAW;;YAGD;EACV,SAAS;EACT;IAAc,iBAAiB;;EAC/B,WAAW;;EAEX,OAAO;;EAEP,SAAS;;EAET,MAAM;;;;;;EAMN,sBAAsB;;YAGZ;EACV;EACA;EACA;;EAEA;EACA,IACE,SAAS,2BACE,aAAa,kBAAkB;;YAGlC;EACV;EACA;EACA;;EAEA;;YAGU,cAAc,kBAAkB;wBAE5B,UACd,cACA,KAAK,wBACL,UAAS,KAAK,iDACb;;;;;;YAqBS;EACV;EACA;;EAEA;;EAEA;EACA,IACE,SAAS,sCACa;;wBAGV,OACd,cACA,KAAK,eACL,UAAS,KAAK,6BACb;;YAMS;EACV;;EAEA;EACA;EACA;;EAEA;EACA,QAAQ;;wBAGM,MACd,cACA,SAAS,KAAK,2CACb;YAMS;EACV;;EAEA;;EAEA;;YAGU,eAAe,eAAe,aAAa;EACrD;EACA;EACA,OAAO;;EAEP,QAAQ;;EAER,oBAAoB;EACpB,YAAY;EACZ,SAAS;;EAET,mBAAmB;EACnB,SAAS;;EAET,WAAW;EACX,YAAY;;;YAIF;;;;;;;;;EASV;;wBAGc,iBAAiB,UAAU,gBAAgB,YAAY,IAAI;YA0D/D,UACV,wBAAwB,4BAA4B;EAEpD;EACA;EACA,OAAO;;wBAGO,kBAAkB,eAAe,WAC/C,OAAO,SACN;YAKS,mBAAmB,iBAAiB,YAAY;wBAE5C,YAAY;EAAS;;YAkBzB;;EAEV,gBAAgB;EAChB,iBAAiB;EACjB,mBAAmB;EACnB,IAAI,cAAc;EAClB,SAAS,cAAc;EACvB,YAAY;EACZ,iBAAiB;;EAEjB,QAAQ,UAAU,iBAAiB;;YAGzB;EACV;EACA;EACA;;YAGU;EACV;EACA;EACA;;;;;;wBAOc,oBACR,gCAAgC,oBACtC,eAAe,iBAAiB;YAuEtB;;;;;YAMA;;EAEV;;EAEA;;EAEA;EACA;EACA;;EAEA;EACA;IAAO;IAAkB;IAAmB;IAAc;;;;YAIhD;EACV,SAAS;EACT,OAAO;EACP,QAAQ;;;YAIE;EAAe;EAAa;;YAE5B;EACV;EACA;EACA,QAAQ;EACR;EACA;EACA,aAAa;EACb;IAAW;IAAY;;EACvB,MAAM;EACN;;EAEA,SAAS;;EAET,UAAU;;YAGA;EACV;EACA;EACA,QAAQ;EACR;EACA,UAAU;EACV;EACA;EACA,aAAa;EACb;IAAW;IAAY;;EACvB,MAAM;EACN;;EAEA;;EAEA,WAAW;;YAGD;EACV;EACA,YAAY;EACZ;;EAEA;;YAGU;EACV;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX,QAAQ;EACR;EACA,QAAQ;EACR;IAAoB;IAAoB;;;;qBAI7B;YAMD;EACV,SAAS;EACT,UAAU;EACV,cAAc;EACd;EACA;EACA;;YAGU;;EAEV;EACA;EACA;;YAGU;EACV,QAAQ;EACR;EACA;EACA,UAAU;EACV,YAAY;EACZ,QAAQ;;YAGE;EACV,QAAQ;EACR;EACA;EACA;EACA;EACA;;EAEA;EACA,QAAQ;;YAGE;EACV,YAAY,OAAO,kBAAkB;EACrC,YAAY,SAAS,eAAe;;EAEpC,cAAc,cAAc,MAAM,aAAa;EAC/C,SAAS,SAAS,eAAe;;YAGvB;WACD;EACT,WAAW,UAAU,gBAAgB,QAAQ;EAC7C,SAAS,SAAS,aAAa;;YAGrB;EACV,SACE,UAAU,aACV;IAAY,UAAU;MACrB,QAAQ;;YAGD;EACV;EACA,QAAQ;EACR;EACA,UAAU;EACV;EACA;EACA,QAAQ;EACR;EACA;EACA,UAAU;EACV,QAAQ;;YAGE;EACV;EACA;EACA;EACA;EACA;;YAGU,YAAY;EACtB;EACA;EACA;EACA;EACA,mBAAmB;;EAEnB,SAAS;;wBAYK,YAAY,iBAAiB;;;KCj1BjC;EACV;;EAEA;;;wBAIc,QAAQ,UAAS,eAAoB"}
|
package/dist/index.mjs
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { $ as ResourceUriSchema, A as RunSummarySchema, B as defineEvalMatrix, C as JudgeRunInfoSchema, D as RunMetadataSchema, E as RunManifestSchema, F as TrialScoringSchema, G as NotJsonError, H as DEFAULT_ARTIFACT_EXCLUDE, I as TrialSummarySchema, J as defineDataset, K as toJson, L as UsageSchema, M as TrajectoryEventSchema, N as TrialManifestSchema, O as RunSourceSchema, P as TrialMetadataSchema, Q as ResourceKindSchema, R as normalizeAutEvent, S as JsonValueSchema, T as RecordedErrorSchema, U as defineConfig, V as canonicalParameters, W as catalogEntry, X as relativeDatasetPath, Y as isDataset, Z as definitionSite, _ as parseTrajectoryJsonl, a as authoringId, at as resourceUuid, b as AutIdentitySchema, c as defineSuite, d as metric, et as UuidSchema, f as predicate, g as user, h as skip, i as agentRuntimes, it as resourceUriSchema, j as ScoreResultSchema, k as RunStatusSchema, l as expectToolCall, m as registerEvals, n as DEFAULT_AGENT_RUNTIMES, nt as parseResourceUri, o as defineAgent, p as recordError, q as toJsonObject, r as agent, rt as resourceUri, s as defineEval, t as BUILT_IN_METRICS, tt as authorId, u as judge, v as ApiErrorSchema, w as MetricResultSchema, x as CheckpointResultSchema, y as ArtifactEntrySchema, z as turnResultEvents } from "./src-
|
|
1
|
+
import { $ as ResourceUriSchema, A as RunSummarySchema, B as defineEvalMatrix, C as JudgeRunInfoSchema, D as RunMetadataSchema, E as RunManifestSchema, F as TrialScoringSchema, G as NotJsonError, H as DEFAULT_ARTIFACT_EXCLUDE, I as TrialSummarySchema, J as defineDataset, K as toJson, L as UsageSchema, M as TrajectoryEventSchema, N as TrialManifestSchema, O as RunSourceSchema, P as TrialMetadataSchema, Q as ResourceKindSchema, R as normalizeAutEvent, S as JsonValueSchema, T as RecordedErrorSchema, U as defineConfig, V as canonicalParameters, W as catalogEntry, X as relativeDatasetPath, Y as isDataset, Z as definitionSite, _ as parseTrajectoryJsonl, a as authoringId, at as resourceUuid, b as AutIdentitySchema, c as defineSuite, d as metric, et as UuidSchema, f as predicate, g as user, h as skip, i as agentRuntimes, it as resourceUriSchema, j as ScoreResultSchema, k as RunStatusSchema, l as expectToolCall, m as registerEvals, n as DEFAULT_AGENT_RUNTIMES, nt as parseResourceUri, o as defineAgent, p as recordError, q as toJsonObject, r as agent, rt as resourceUri, s as defineEval, t as BUILT_IN_METRICS, tt as authorId, u as judge, v as ApiErrorSchema, w as MetricResultSchema, x as CheckpointResultSchema, y as ArtifactEntrySchema, z as turnResultEvents } from "./src-LeGSyDD2.mjs";
|
|
2
2
|
//#region ../agents/src/index.ts
|
|
3
3
|
/** A local, one-prompt-per-turn AUT backed by Pi's non-interactive print mode. */
|
|
4
4
|
function piAgent(options = {}) {
|
package/dist/runner.d.mts
CHANGED
|
@@ -642,6 +642,19 @@ type EvalDefinition<TAgent extends AutAdapter = AutAdapter> = {
|
|
|
642
642
|
policy?: EvalPolicy;
|
|
643
643
|
/** Any JSON-convertible object; converted when the eval is defined. */
|
|
644
644
|
metadata?: Record<string, unknown>;
|
|
645
|
+
workspace?: EvalWorkspace;
|
|
646
|
+
};
|
|
647
|
+
/** How EvalKit treats this scenario's candidate workspace. */
|
|
648
|
+
type EvalWorkspace = {
|
|
649
|
+
/**
|
|
650
|
+
* Paths to leave out of the trial's snapshot, read like `.gitignore`: globs
|
|
651
|
+
* relative to the workspace, where a pattern without `/` matches that name
|
|
652
|
+
* at any depth and later patterns win. They come after the project's
|
|
653
|
+
* `artifacts.exclude` (default `DEFAULT_ARTIFACT_EXCLUDE`), so `!pattern`
|
|
654
|
+
* can keep a path the project leaves out, even inside an ignored directory:
|
|
655
|
+
* `['!_build/doc/**']` keeps generated docs while dropping the rest of `_build`.
|
|
656
|
+
*/
|
|
657
|
+
ignore?: readonly string[];
|
|
645
658
|
};
|
|
646
659
|
type EvalRegistryMetadata = {
|
|
647
660
|
id: string;
|
|
@@ -720,6 +733,8 @@ type TrialMetadata = {
|
|
|
720
733
|
};
|
|
721
734
|
aut?: AutIdentity;
|
|
722
735
|
startedAt: string;
|
|
736
|
+
/** The effective snapshot ignore patterns: the project's `artifacts.exclude`, then the scenario's `workspace.ignore`. */
|
|
737
|
+
workspaceIgnore?: string[];
|
|
723
738
|
/** From schema version 5: the datasets this trial used, with content digests. */
|
|
724
739
|
datasets?: TrialDataset[];
|
|
725
740
|
};
|
|
@@ -920,6 +935,7 @@ export declare const readTrialManifest: (root: string, runId: string, trialId: s
|
|
|
920
935
|
readonly version?: string | undefined;
|
|
921
936
|
} | undefined;
|
|
922
937
|
readonly startedAt: string;
|
|
938
|
+
readonly workspaceIgnore?: string[] | undefined;
|
|
923
939
|
readonly datasets?: {
|
|
924
940
|
readonly id: string;
|
|
925
941
|
readonly visibility: "candidate" | "evaluator";
|
|
@@ -1152,7 +1168,10 @@ export type RunEvalOptions = {
|
|
|
1152
1168
|
workspaceRoot?: string;
|
|
1153
1169
|
/** With `workspaceRoot`: which trial workspaces to keep afterwards. Default `'failed'`. */
|
|
1154
1170
|
keepWorkspaces?: KeepSandboxes;
|
|
1155
|
-
/**
|
|
1171
|
+
/**
|
|
1172
|
+
* Candidate paths left out of the artifact snapshot, before the scenario's
|
|
1173
|
+
* own `workspace.ignore`. Default `DEFAULT_ARTIFACT_EXCLUDE`.
|
|
1174
|
+
*/
|
|
1156
1175
|
artifactExclude?: readonly string[];
|
|
1157
1176
|
/** Repository and commit the eval code came from, recorded in the manifest. */
|
|
1158
1177
|
source?: RunSource;
|
package/dist/runner.d.mts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runner.d.mts","names":["_$_evalkit_core0.JsonValue","_$_evalkit_core0.TrajectoryEvent"],"sources":["../../runner/src/errors.ts","../../core/src/identity.ts","../../core/src/dataset.ts","../../core/src/aut-events.ts","../../core/src/catalog.ts","../../core/src/matrix.ts","../../core/src/config.ts","../../core/src/index.ts","../../runner/src/local-report-store.ts","../../runner/src/datasets.ts","../../runner/src/local-report-reader.ts","../../runner/src/matrix.ts","../../runner/src/workspace.ts","../../runner/src/index.ts"],"mappings":";;;;;;qBAEa,0BAA0B;EACrC;EACA;;;;;qBAGW,qBAAqB;EAChC;EACA;;;;;qBAGW,oBAAoB;EAC/B;EACA;;;;;qBAGW,qBAAqB;EAChC;EACA;;;;;qBASW,iCAAiC;EAG5C;EACA;EACA;EACA;;;;;;qBAIW,uBAAuB;EAClC;;;;cCrCW,oBAAkB,OAAA;KACnB,sBAAsB,mBAAmB;KAiBzC,YAAY,cAAc,eAAe,2BACxC;;;;KCdD;;KAGA,6BAA6B;EAAe;;;;;;KAM5C,wBAAwB,SAAS,eAAe;;KAGhD;EACV,QAAQ;;EAER;;KAGU;;EAEV;EACA,QAAQ;;KAGE;EACV;EACA;EACA;EACA;;EAEA;;EAEA,YAAY;EACZ,QAAQ;;KAoCE;EAEN;EACA,QAAQ;EACR;EACA;;EAGA;EACA,SACE,MAAM,uBACH,iBAAiB,QAAQ;EAC9B,SACE,QAAQ,gBACR,OAAO,wBACJ,iBAAiB,QAAQ;;;KAIxB;WACD;WACA;WACA;WACA,YAAY;;WAEZ;WACA,QAAQ;;;;;;;;KC/FP;EACN;EAAiB;;EAEjB;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB;EAAkB;;EACrC;EAAe,OAAO;EAAe;;KAE/B;EACV;EACA;;EAEA;EACA;;;KAIU;EAGN;EACA,qBAAqB;EACrB,QAAQ;;;;KC5BF;EACV;EACA;EACA;EACA;EACA,UAAU;IAAQ,MAAM;IAAkB;;;EAE1C;;;KAIU;EACN;EAAe;;EACf;EAAgB;;EAChB;;KAEM;EACV;EACA;EACA,YAAY;EACZ;EACA;;EAEA;IAAU;;IAAwB;;;EAElC;IAAU,OAAO;;IAAiB;;;EAElC;IAAU;IAAc;;EACxB;;KAGU;EACV;EACA,MAAM;EACN;EACA;;EAEA;EACA;;KAGU;;;EAEN;EAAc;;EACd;EAAe,aAAa;;EAC5B;EAA0B;EAAc,YAAY;;;;EAEpD;EAAoB,QAAQ;;KAEtB;EACV;EACA;EACA;EACA;EACA,OAAO;;EAEP,aAAa;EACb,UAAU;EACV;EACA;IAAU;IAAgB;IAAoB;;EAC9C,YAAY;EACZ,SAAS;EACT,WAAW;;EAEX;;;;KChFU;;EAEV;;EAEA,aAAa,SAAS,wBAAwB;;EAE9C,YAAY;;KAEF;EACV;;EAEA;EACA,MAAM;EACN,YAAY;;KAEF;EACV;EACA;EACA,gBAAgB;EAChB,YAAY,SAAS,wBAAwB;EAC7C,WAAW;;EAEX,iBAAiB;;EAEjB,mBAAmB;;KAET,aAAa;EACvB;EACA,MAAM,YAAY;EAClB,MAAM,YAAY,kBAAkB,iBAAiB;;;;
|
|
1
|
+
{"version":3,"file":"runner.d.mts","names":["_$_evalkit_core0.JsonValue","_$_evalkit_core0.TrajectoryEvent"],"sources":["../../runner/src/errors.ts","../../core/src/identity.ts","../../core/src/dataset.ts","../../core/src/aut-events.ts","../../core/src/catalog.ts","../../core/src/matrix.ts","../../core/src/config.ts","../../core/src/index.ts","../../runner/src/local-report-store.ts","../../runner/src/datasets.ts","../../runner/src/local-report-reader.ts","../../runner/src/matrix.ts","../../runner/src/workspace.ts","../../runner/src/index.ts"],"mappings":";;;;;;qBAEa,0BAA0B;EACrC;EACA;;;;;qBAGW,qBAAqB;EAChC;EACA;;;;;qBAGW,oBAAoB;EAC/B;EACA;;;;;qBAGW,qBAAqB;EAChC;EACA;;;;;qBASW,iCAAiC;EAG5C;EACA;EACA;EACA;;;;;;qBAIW,uBAAuB;EAClC;;;;cCrCW,oBAAkB,OAAA;KACnB,sBAAsB,mBAAmB;KAiBzC,YAAY,cAAc,eAAe,2BACxC;;;;KCdD;;KAGA,6BAA6B;EAAe;;;;;;KAM5C,wBAAwB,SAAS,eAAe;;KAGhD;EACV,QAAQ;;EAER;;KAGU;;EAEV;EACA,QAAQ;;KAGE;EACV;EACA;EACA;EACA;;EAEA;;EAEA,YAAY;EACZ,QAAQ;;KAoCE;EAEN;EACA,QAAQ;EACR;EACA;;EAGA;EACA,SACE,MAAM,uBACH,iBAAiB,QAAQ;EAC9B,SACE,QAAQ,gBACR,OAAO,wBACJ,iBAAiB,QAAQ;;;KAIxB;WACD;WACA;WACA;WACA,YAAY;;WAEZ;WACA,QAAQ;;;;;;;;KC/FP;EACN;EAAiB;;EAEjB;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB;EAAkB;;EACrC;EAAe,OAAO;EAAe;;KAE/B;EACV;EACA;;EAEA;EACA;;;KAIU;EAGN;EACA,qBAAqB;EACrB,QAAQ;;;;KC5BF;EACV;EACA;EACA;EACA;EACA,UAAU;IAAQ,MAAM;IAAkB;;;EAE1C;;;KAIU;EACN;EAAe;;EACf;EAAgB;;EAChB;;KAEM;EACV;EACA;EACA,YAAY;EACZ;EACA;;EAEA;IAAU;;IAAwB;;;EAElC;IAAU,OAAO;;IAAiB;;;EAElC;IAAU;IAAc;;EACxB;;KAGU;EACV;EACA,MAAM;EACN;EACA;;EAEA;EACA;;KAGU;;;EAEN;EAAc;;EACd;EAAe,aAAa;;EAC5B;EAA0B;EAAc,YAAY;;;;EAEpD;EAAoB,QAAQ;;KAEtB;EACV;EACA;EACA;EACA;EACA,OAAO;;EAEP,aAAa;EACb,UAAU;EACV;EACA;IAAU;IAAgB;IAAoB;;EAC9C,YAAY;EACZ,SAAS;EACT,WAAW;;EAEX;;;;KChFU;;EAEV;;EAEA,aAAa,SAAS,wBAAwB;;EAE9C,YAAY;;KAEF;EACV;;EAEA;EACA,MAAM;EACN,YAAY;;KAEF;EACV;EACA;EACA,gBAAgB;EAChB,YAAY,SAAS,wBAAwB;EAC7C,WAAW;;EAEX,iBAAiB;;EAEjB,mBAAmB;;KAET,aAAa;EACvB;EACA,MAAM,YAAY;EAClB,MAAM,YAAY,kBAAkB,iBAAiB;;;;KCH3C;;;KCfA;KACA,YACV,gBAAgB;GAAiB,cAAc;;KACrC;GAAgB,cAAc;;KAE9B;;EAEV;EACA;EACA;EACA;;KAGU;;;;;KAMA;EACV;EACA,gBAAgB;;KAGN,gBAAgB,QAAQ,OAAO,kBAAkB;KAEjD;EACV;EACA;EACA;;KAGU;EACV;EACA;EACA;;KAGU;EACN;EAAiB;;EAEjB;EACA;EACA,SAAS;EACT;;EAGA;EACA;EACA;EACA,WAAW;EACX;;EAGA;EACA;EACA;EACA,QAAQ;EACR;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB,SAAS;EAAW;;EACvC;EAAe,OAAO;EAAe;;KAE/B;EACN;EAAuB;;EAEvB;EACA;;EAEA;EACA;;EAEA;EAAmC;EAAc;;EACjD;EAAiC;EAAc;;EAE/C;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAwB;EAAgB;;EAExC;EACA;;EAEA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAyB;;EACzB;EAAe,OAAO;EAAe;;KAE/B,mBACT;EAAa;MAAoB;EAAgB;;KAExC;;EAEV;;KAGU,qBAAqB;KACrB,qBAAqB;KAErB;EACV;EACA;EACA;EACA;EACA,UAAU;;EAEV,aAAa;EACb,UAAU;;EAEV,MAAM;;EAEN,WAAW;;EAEX,oBAAoB;;KAGV,WAAW;WACZ,WAAW;;WAEX,WAAW;EACpB,MAAM;IACJ,SAAS;IACT,UAAU;;IAEV,UAAU,OAAO,yBAAyB;MACxC,QAAQ,WAAW;;KAGb,WAAW;;;;;EAKrB,KAAK,SAAS,WAAW,eAAe;;EAExC,UAAU;;;KAyBA;;EAEV,YAAY;;EAEZ,MAAM;;EAEN;EACA;;KAGU;EACV;;EAEA,oBAAoB,OAAO;;KAGjB;EACV;EACA,aAAa;;KAGH;EACV;EACA,kBAAkB;IAAa;;EAC/B,6BAA6B,QAAQ;IAAY;;IAC/C;IACA;;EAEF;IACE;IACA;IACA;IACA,WAAW;IACX;IACA,SAAS;;EAEX;;KAGU;EACV;EACA;EACA;IAAY;IAAc,YAAY;;;KAE5B,iBACV,WAAW,YAAY,cAAc;KAuB3B;EACV,QAAQ;EACR,QAAQ;;EAER,SAAS;;;KAIC;EAAiB;EAAe;;KAEhC,sBAER;EAEE;EACA;EACA;;EAEA;;EAEA,QAAQ;;KAYF;EACV;EACA;EACA;EACA;EACA;EACA,WAAW;EACX,QAAQ;EACR;EACA,QAAQ;;EAER;;KAGU;EACV,WAAW;EACX,WAAW;;KAGD;EACV,SAAS;EACT;IAAc,iBAAiB;;EAC/B,WAAW;;EAEX,OAAO;;EAEP,SAAS;;EAET,MAAM;;;;;;EAMN,sBAAsB;;KAGZ;EACV;EACA;EACA;;EAEA;EACA,IACE,SAAS,2BACE,aAAa,kBAAkB;;KAGlC;EACV;EACA;EACA;;EAEA;;KAGU,cAAc,kBAAkB;;;;;;KA2BhC;EACV;EACA;;EAEA;;EAEA;EACA,IACE,SAAS,sCACa;;;KAad;EACV;;EAEA;EACA;EACA;;EAEA;EACA,QAAQ;;KAYE;EACV;;EAEA;;EAEA;;KAGU,eAAe,eAAe,aAAa;EACrD;EACA;EACA,OAAO;;EAEP,QAAQ;;EAER,oBAAoB;EACpB,YAAY;EACZ,SAAS;;EAET,mBAAmB;EACnB,SAAS;;EAET,WAAW;EACX,YAAY;;;KAIF;;;;;;;;;EASV;;KA6GU;EACV;EACA;EACA;;KAGU;EACV;EACA;EACA;;KAgFU;;;;;KAMA;;EAEV;;EAEA;;EAEA;EACA;EACA;;EAEA;EACA;IAAO;IAAkB;IAAmB;IAAc;;;;KAIhD;EACV,SAAS;EACT,OAAO;EACP,QAAQ;;;KAIE;EAAe;EAAa;;KAE5B;EACV;EACA;EACA,QAAQ;EACR;EACA;EACA,aAAa;EACb;IAAW;IAAY;;EACvB,MAAM;EACN;;EAEA,SAAS;;EAET,UAAU;;KAGA;EACV;EACA;EACA,QAAQ;EACR;EACA,UAAU;EACV;EACA;EACA,aAAa;EACb;IAAW;IAAY;;EACvB,MAAM;EACN;;EAEA;;EAEA,WAAW;;KAGD;EACV;EACA,YAAY;EACZ;;EAEA;;KAGU;EACV;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX,QAAQ;EACR;EACA,QAAQ;EACR;IAAoB;IAAoB;;;KAU9B;EACV,SAAS;EACT,UAAU;EACV,cAAc;EACd;EACA;EACA;;KAGU;;EAEV;EACA;EACA;;KAGU;EACV,QAAQ;EACR;EACA;EACA,UAAU;EACV,YAAY;EACZ,QAAQ;;KAGE;EACV,QAAQ;EACR;EACA;EACA;EACA;EACA;;EAEA;EACA,QAAQ;;KAGE;EACV,YAAY,OAAO,kBAAkB;EACrC,YAAY,SAAS,eAAe;;EAEpC,cAAc,cAAc,MAAM,aAAa;EAC/C,SAAS,SAAS,eAAe;;KAGvB;WACD;EACT,WAAW,UAAU,gBAAgB,QAAQ;EAC7C,SAAS,SAAS,aAAa;;KAGrB;EACV,SACE,UAAU,aACV;IAAY,UAAU;MACrB,QAAQ;;KAGD;EACV;EACA,QAAQ;EACR;EACA,UAAU;EACV;EACA;EACA,QAAQ;EACR;EACA;EACA,UAAU;EACV,QAAQ;;KAGE;EACV;EACA;EACA;EACA;EACA;;KAGU,YAAY;EACtB;EACA;EACA;EACA;EACA,mBAAmB;;EAEnB,SAAS;;;;;wBC5sBK,iBAAiB,wBAAwB;;;;KCiI7C;EAAqB;EAAc;;;;;;;wBAOzB,mBACpB,SAAS,SACT,eACC,QAAQ;;wBA6BW,kBACpB,SAAS,SACT,cACA,eACC,QAAQ,WAAW;;;qBC3RT,kBAAe,cAAgB,kBAAe;;;;;;0BAA9CA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAEA,iBAAc,cAAgB,kBAAe;;;;;;;;;;;;;;qBAE7C,oBAAiB,cAChB,eACC,oBACE;;;;;;;0BAHJA;;;;;;;;;;;;;;;;;;;;;;qBASA,mBAAgB,cACf,eACC,oBACE;;;;;;;;;;;0BAHJA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBASA,mBAAgB,cACf,eACC,oBACE;;;;;;;wBAHJA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;wBASS,gBACpB,cACA,eACA,kBAAe,QAHKC;;;;;wBAiBN,YAAY,OAAO;;;;KC7DvB;EACV;EACA;EACA;EACA;;;wBAIc,UACd,QAAQ,YACR,SAAS,KACP;EAGA,YAAY;;EAEZ,YAAY,MAAM;EAClB,YACE,MAAM,gBACN,QAAQ,qBACE;IAEb,OAAO,OAAO;;;;wBChBD,mBAAmB,cAAc;;;YC0DrC;EACV,QAAQ;EACR;EACA;EACA;;EAEA;;EAEA,aAAa;;EAEb,SAAS;;EAET,UAAU;;EAEV;;EAEA,YAAY,OAAO;;EAEnB;;EAEA,iBAAiB;;;;;EAKjB;;EAEA,SAAS;;EAET,UAAU;;EAEV;;;;;EAKA,SAAS;;;;;EAKT,WAAW,OAAO,UAAU,OAAO;EACnC,YAAY;;;YAIF;EACV;EACA;EACA;EACA;EACA;EACA,aAAa;;;wBASC,QACd,YAAY,gBACZ,SAAS,iBACR,OAAO,OAAO"}
|
package/dist/runner.mjs
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import { _ as CheckpointExecutionError, a as readRunSummary, b as ScoringError, c as readTrialScoring, d as trialWorkspacePath, f as staticDatasetFile, g as CancelledError, h as AutExecutionError, i as readRunManifest, l as readTrialSummary, n as runMatrix, o as readTrialEvents, p as staticDatasetFiles, r as ownerIsGone, s as readTrialManifest, t as runEval, u as localReportStore, v as DatasetError, y as ReportError } from "./src-
|
|
1
|
+
import { _ as CheckpointExecutionError, a as readRunSummary, b as ScoringError, c as readTrialScoring, d as trialWorkspacePath, f as staticDatasetFile, g as CancelledError, h as AutExecutionError, i as readRunManifest, l as readTrialSummary, n as runMatrix, o as readTrialEvents, p as staticDatasetFiles, r as ownerIsGone, s as readTrialManifest, t as runEval, u as localReportStore, v as DatasetError, y as ReportError } from "./src-mm8L_IxM.mjs";
|
|
2
2
|
export { AutExecutionError, CancelledError, CheckpointExecutionError, DatasetError, ReportError, ScoringError, localReportStore, ownerIsGone, readRunManifest, readRunSummary, readTrialEvents, readTrialManifest, readTrialScoring, readTrialSummary, runEval, runMatrix, staticDatasetFile, staticDatasetFiles, trialWorkspacePath };
|
|
@@ -619,6 +619,8 @@ const TrialMetadataSchema = Schema.Struct({
|
|
|
619
619
|
matrix: Schema.optional(MatrixCellSchema),
|
|
620
620
|
aut: Schema.optional(AutIdentitySchema),
|
|
621
621
|
startedAt: Schema.String,
|
|
622
|
+
/** The effective snapshot ignore patterns: the project's, then the scenario's. */
|
|
623
|
+
workspaceIgnore: Schema.optional(Schema.mutable(Schema.Array(Schema.String))),
|
|
622
624
|
datasets: Schema.optional(Schema.mutable(Schema.Array(Schema.Struct({
|
|
623
625
|
id: Schema.String,
|
|
624
626
|
visibility: Schema.Literal("candidate", "evaluator"),
|
|
@@ -957,6 +959,7 @@ function defineEval(definition) {
|
|
|
957
959
|
authoringId(definition);
|
|
958
960
|
if ("fixtures" in definition) throw new Error(`Eval ${definition.id} uses fixtures, which were replaced by datasets: [defineDataset(...)]`);
|
|
959
961
|
if (definition.metadata !== void 0) toJsonObject(definition.metadata, `${definition.id}.metadata`);
|
|
962
|
+
for (const [index, pattern] of (definition.workspace?.ignore ?? []).entries()) if (typeof pattern !== "string" || pattern.replace(/^!/, "").trim() === "") throw new Error(`Eval ${definition.id} workspace.ignore[${index}] must be a non-empty glob, got ${JSON.stringify(pattern)}`);
|
|
960
963
|
const datasetIds = /* @__PURE__ */ new Set();
|
|
961
964
|
for (const [index, dataset] of (definition.datasets ?? []).entries()) {
|
|
962
965
|
if (!isDataset(dataset)) throw new Error(`Eval ${definition.id} datasets[${index}] is not a dataset; create it with defineDataset(...)`);
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { A as RunSummarySchema, D as RunMetadataSchema, E as RunManifestSchema, F as TrialScoringSchema, H as DEFAULT_ARTIFACT_EXCLUDE, I as TrialSummarySchema, K as toJson, M as TrajectoryEventSchema, N as TrialManifestSchema, P as TrialMetadataSchema, R as normalizeAutEvent, S as JsonValueSchema, V as canonicalParameters, X as relativeDatasetPath, _ as parseTrajectoryJsonl, a as authoringId, i as agentRuntimes, j as ScoreResultSchema, p as recordError, q as toJsonObject, rt as resourceUri, x as CheckpointResultSchema, z as turnResultEvents } from "./src-
|
|
1
|
+
import { A as RunSummarySchema, D as RunMetadataSchema, E as RunManifestSchema, F as TrialScoringSchema, H as DEFAULT_ARTIFACT_EXCLUDE, I as TrialSummarySchema, K as toJson, M as TrajectoryEventSchema, N as TrialManifestSchema, P as TrialMetadataSchema, R as normalizeAutEvent, S as JsonValueSchema, V as canonicalParameters, X as relativeDatasetPath, _ as parseTrajectoryJsonl, a as authoringId, i as agentRuntimes, j as ScoreResultSchema, p as recordError, q as toJsonObject, rt as resourceUri, x as CheckpointResultSchema, z as turnResultEvents } from "./src-LeGSyDD2.mjs";
|
|
2
2
|
import * as Schema from "effect/Schema";
|
|
3
3
|
import { Data, Effect, Either, Exit, Scope } from "effect";
|
|
4
4
|
import os from "node:os";
|
|
@@ -63,41 +63,82 @@ function globMatcher(pattern) {
|
|
|
63
63
|
}
|
|
64
64
|
//#endregion
|
|
65
65
|
//#region ../runner/src/snapshot.ts
|
|
66
|
-
|
|
67
|
-
function
|
|
68
|
-
const
|
|
69
|
-
|
|
70
|
-
|
|
71
|
-
|
|
72
|
-
|
|
73
|
-
|
|
66
|
+
const GLOB_CHARS = /[*?{]/;
|
|
67
|
+
function ignoreRule(pattern) {
|
|
68
|
+
const negated = pattern.startsWith("!");
|
|
69
|
+
const trimmed = (negated ? pattern.slice(1) : pattern).replace(/^\.\//, "").replace(/\/$/, "");
|
|
70
|
+
const match = globMatcher(trimmed);
|
|
71
|
+
if (!trimmed.includes("/")) return {
|
|
72
|
+
negated,
|
|
73
|
+
match: (relativePath) => match(path.posix.basename(relativePath)),
|
|
74
|
+
mayMatchBelow: () => true
|
|
75
|
+
};
|
|
76
|
+
const segments = trimmed.split("/");
|
|
77
|
+
const firstGlob = segments.findIndex((segment) => GLOB_CHARS.test(segment));
|
|
78
|
+
const prefix = segments.slice(0, firstGlob === -1 ? segments.length : firstGlob).join("/");
|
|
79
|
+
return {
|
|
80
|
+
negated,
|
|
81
|
+
match,
|
|
82
|
+
mayMatchBelow: (directory) => prefix === "" || prefix.startsWith(`${directory}/`) || prefix === directory || directory.startsWith(`${prefix}/`)
|
|
83
|
+
};
|
|
84
|
+
}
|
|
85
|
+
/**
|
|
86
|
+
* Ignore rules for candidate snapshots, read like `.gitignore`: globs relative
|
|
87
|
+
* to the workspace root, where a pattern without `/` matches that name at any
|
|
88
|
+
* depth, and later patterns win. `!pattern` re-includes a path; unlike Git, it
|
|
89
|
+
* also works below an ignored directory (`_build` then `!_build/doc/**`).
|
|
90
|
+
*/
|
|
91
|
+
function ignoreRules(patterns) {
|
|
92
|
+
const rules = patterns.map(ignoreRule);
|
|
93
|
+
return {
|
|
94
|
+
/** Whether a path is ignored, given whether its directory was. */
|
|
95
|
+
ignored(relativePath, parentIgnored = false) {
|
|
96
|
+
let ignored = parentIgnored;
|
|
97
|
+
for (const rule of rules) if (rule.match(relativePath)) ignored = !rule.negated;
|
|
98
|
+
return ignored;
|
|
99
|
+
},
|
|
100
|
+
/** Whether an ignored directory must still be walked for re-included paths. */
|
|
101
|
+
mayReinclude(directory) {
|
|
102
|
+
return rules.some((rule) => rule.negated && rule.mayMatchBelow(directory));
|
|
103
|
+
}
|
|
104
|
+
};
|
|
74
105
|
}
|
|
75
106
|
/**
|
|
76
107
|
* Copies the candidate workspace's regular files into the trial report,
|
|
77
|
-
* leaving out
|
|
78
|
-
* never followed, so a snapshot cannot read outside the trial.
|
|
108
|
+
* leaving out ignored paths. Symbolic links and special files are skipped,
|
|
109
|
+
* never followed, so a snapshot cannot read outside the trial. An ignored
|
|
110
|
+
* directory is walked only when a `!` pattern could re-include something in
|
|
111
|
+
* it, and is listed only if it ends up containing a snapshotted file.
|
|
79
112
|
*/
|
|
80
|
-
async function snapshotCandidateWorkspace(candidateRoot, writer,
|
|
113
|
+
async function snapshotCandidateWorkspace(candidateRoot, writer, ignore = DEFAULT_ARTIFACT_EXCLUDE) {
|
|
81
114
|
const artifacts = [];
|
|
82
|
-
const
|
|
83
|
-
|
|
115
|
+
const rules = ignoreRules(ignore);
|
|
116
|
+
/** Returns whether anything below `directory` was snapshotted. */
|
|
117
|
+
async function visit(directory, relativeDirectory, directoryIgnored) {
|
|
84
118
|
const entries = await readdir(directory, { withFileTypes: true });
|
|
85
119
|
entries.sort((left, right) => left.name.localeCompare(right.name));
|
|
120
|
+
let kept = false;
|
|
86
121
|
for (const entry of entries) {
|
|
87
122
|
const absolutePath = path.join(directory, entry.name);
|
|
88
123
|
const relativePath = path.join(relativeDirectory, entry.name);
|
|
89
|
-
|
|
124
|
+
const posixPath = relativePath.split(path.sep).join("/");
|
|
125
|
+
const ignored = rules.ignored(posixPath, directoryIgnored);
|
|
90
126
|
const stat = await lstat(absolutePath);
|
|
91
127
|
if (stat.isSymbolicLink()) continue;
|
|
92
128
|
if (stat.isDirectory()) {
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
129
|
+
if (ignored && !rules.mayReinclude(posixPath)) continue;
|
|
130
|
+
const entryIndex = artifacts.length;
|
|
131
|
+
const keptBelow = await visit(absolutePath, relativePath, ignored);
|
|
132
|
+
if (!ignored || keptBelow) {
|
|
133
|
+
artifacts.splice(entryIndex, 0, {
|
|
134
|
+
path: path.join("candidate", relativePath),
|
|
135
|
+
kind: "directory"
|
|
136
|
+
});
|
|
137
|
+
kept = true;
|
|
138
|
+
}
|
|
98
139
|
continue;
|
|
99
140
|
}
|
|
100
|
-
if (!stat.isFile()) continue;
|
|
141
|
+
if (ignored || !stat.isFile()) continue;
|
|
101
142
|
const data = await readFile(absolutePath);
|
|
102
143
|
const artifactPath = path.join("candidate", relativePath);
|
|
103
144
|
await writer.writeArtifact(artifactPath, data);
|
|
@@ -106,9 +147,11 @@ async function snapshotCandidateWorkspace(candidateRoot, writer, exclude = DEFAU
|
|
|
106
147
|
kind: "file",
|
|
107
148
|
size: data.byteLength
|
|
108
149
|
});
|
|
150
|
+
kept = true;
|
|
109
151
|
}
|
|
152
|
+
return kept;
|
|
110
153
|
}
|
|
111
|
-
await visit(candidateRoot, "");
|
|
154
|
+
await visit(candidateRoot, "", false);
|
|
112
155
|
return artifacts;
|
|
113
156
|
}
|
|
114
157
|
//#endregion
|
|
@@ -1202,6 +1245,7 @@ async function runTrial(definition, options) {
|
|
|
1202
1245
|
}
|
|
1203
1246
|
const candidateData = datasetValues(prepared, "candidate");
|
|
1204
1247
|
const scoringData = datasetValues(prepared, "all");
|
|
1248
|
+
const workspaceIgnore = [...options.artifactExclude ?? DEFAULT_ARTIFACT_EXCLUDE, ...definition.workspace?.ignore ?? []];
|
|
1205
1249
|
const events = [];
|
|
1206
1250
|
const runWriter = options.runWriter ?? await options.report.startRun(runMetadata(definition, options, runId, startedAt), { catalog: options.catalog });
|
|
1207
1251
|
const trialWriter = await runWriter.startTrial({
|
|
@@ -1216,6 +1260,7 @@ async function runTrial(definition, options) {
|
|
|
1216
1260
|
parameters: options.parameters,
|
|
1217
1261
|
matrix: options.matrix,
|
|
1218
1262
|
startedAt,
|
|
1263
|
+
workspaceIgnore,
|
|
1219
1264
|
...prepared.length ? { datasets: trialDatasets(prepared) } : {}
|
|
1220
1265
|
});
|
|
1221
1266
|
let writeChain = Promise.resolve();
|
|
@@ -1381,7 +1426,7 @@ async function runTrial(definition, options) {
|
|
|
1381
1426
|
scoring = scoreSummary(scoreResults, checkpoints, skippedScorers, metrics);
|
|
1382
1427
|
await trialWriter.writeScores(scoring);
|
|
1383
1428
|
if (workspace) try {
|
|
1384
|
-
artifacts = await snapshotCandidateWorkspace(workspace.artifacts.candidate.root, trialWriter,
|
|
1429
|
+
artifacts = await snapshotCandidateWorkspace(workspace.artifacts.candidate.root, trialWriter, workspaceIgnore);
|
|
1385
1430
|
} catch (captureError) {
|
|
1386
1431
|
if (!primaryError) primaryError = captureError;
|
|
1387
1432
|
await tryEmitError(captureError, emitRunner, now);
|