@hue-run/sdk 0.9.0 → 0.10.0
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/CLI.md +100 -29
- package/ENVIRONMENTS.md +44 -9
- package/EVALUATIONS.md +39 -19
- package/README.md +51 -22
- package/dist/cli/eval-direct.d.ts +12 -1
- package/dist/cli/eval-direct.js +145 -33
- package/dist/cli/eval.d.ts +28 -1
- package/dist/cli/eval.js +469 -72
- package/dist/evals/checkpoint.d.ts +15 -1
- package/dist/evals/checkpoint.js +34 -3
- package/dist/evals/client.d.ts +19 -3
- package/dist/evals/client.js +82 -34
- package/dist/evals/environment-target.d.ts +18 -3
- package/dist/evals/environment-target.js +22 -10
- package/dist/evals/exit-cleanup.d.ts +4 -0
- package/dist/evals/exit-cleanup.js +26 -0
- package/dist/evals/files.d.ts +59 -1
- package/dist/evals/files.js +187 -27
- package/dist/evals/local-worker.d.ts +21 -3
- package/dist/evals/local-worker.js +14 -0
- package/dist/evals/runner.d.ts +45 -0
- package/dist/evals/runner.js +164 -13
- package/dist/evals/scenarios.d.ts +7 -5
- package/dist/evals/scenarios.js +71 -14
- package/dist/evals/simulation.d.ts +17 -4
- package/dist/evals/simulation.js +8 -0
- package/dist/evals/types.d.ts +4 -0
- package/dist/evals/verdicts.d.ts +15 -2
- package/dist/evals/verdicts.js +43 -9
- package/dist/evals.d.ts +4 -3
- package/dist/evals.js +2 -2
- package/dist/inline-files.d.ts +4 -0
- package/dist/inline-files.js +40 -17
- package/dist/snapshot.d.ts +2 -2
- package/dist/snapshot.js +47 -12
- package/dist/transport.js +2 -2
- package/dist/version.d.ts +1 -1
- package/dist/version.js +1 -1
- package/package.json +1 -1
package/dist/evals.d.ts
CHANGED
|
@@ -1,9 +1,10 @@
|
|
|
1
|
-
export { createEvaluationClient, EvaluationClient, HueApiError } from "./evals/client.js";
|
|
1
|
+
export { ArtifactSizeError, createEvaluationClient, EvaluationClient, HueApiError, } from "./evals/client.js";
|
|
2
2
|
export type { EvaluationClientOptions } from "./evals/client.js";
|
|
3
3
|
export { runExperiment, rescore, UncertainExecutionError, OutcomeSerializationError, TargetCancelledError, TargetOutcomeUncertainError, } from "./evals/runner.js";
|
|
4
|
-
export type { RunExperimentOptions, RunExperimentTargetContext, RescoreOptions, RunnerReport, } from "./evals/runner.js";
|
|
4
|
+
export type { RunExperimentOptions, RunExperimentTargetContext, RescoreOptions, RunnerReport, TelemetryIssueCount, TelemetryNotAccepted, } from "./evals/runner.js";
|
|
5
5
|
export { TargetResult, withFiles } from "./evals/types.js";
|
|
6
|
-
export { outputContentTypes, outputFileLimits, OutputFileError, safeFilename, targetFileRoles, } from "./evals/files.js";
|
|
6
|
+
export { CaseFileError, outputContentTypes, outputFileLimits, OutputFileError, safeFilename, targetFileRoles, } from "./evals/files.js";
|
|
7
|
+
export type { CaseFileErrorCode } from "./evals/files.js";
|
|
7
8
|
export { runSimulation } from "./evals/simulation.js";
|
|
8
9
|
export type { RepositorySimulationCase, RepositorySimulationScorer, RunSimulationOptions, SimulationDefinition, SimulationProgress, SimulationReport, SimulationScenario, SimulationTargetContext, } from "./evals/simulation.js";
|
|
9
10
|
export { actualAgentManifestV2, agentManifestDigestV2, attemptBaselineV2, attemptBindingRead, attemptConnectionBundleV2, attemptIdentityV2, dependencyManifestV2, dependencyProviderV2, expectedAgentManifestV2, executionManifestDigestV2, parityEvidenceV2, preflightFindingV2, preflightReportV2, prepareAttemptInputV2, projectMcpConnectionV2, secretFreeBindingV2, surfaceBindingV2, } from "./evals/attempt.js";
|
package/dist/evals.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
|
-
export { createEvaluationClient, EvaluationClient, HueApiError } from "./evals/client.js";
|
|
1
|
+
export { ArtifactSizeError, createEvaluationClient, EvaluationClient, HueApiError, } from "./evals/client.js";
|
|
2
2
|
export { runExperiment, rescore, UncertainExecutionError, OutcomeSerializationError, TargetCancelledError, TargetOutcomeUncertainError, } from "./evals/runner.js";
|
|
3
3
|
export { TargetResult, withFiles } from "./evals/types.js";
|
|
4
|
-
export { outputContentTypes, outputFileLimits, OutputFileError, safeFilename, targetFileRoles, } from "./evals/files.js";
|
|
4
|
+
export { CaseFileError, outputContentTypes, outputFileLimits, OutputFileError, safeFilename, targetFileRoles, } from "./evals/files.js";
|
|
5
5
|
export { runSimulation } from "./evals/simulation.js";
|
|
6
6
|
export { actualAgentManifestV2, agentManifestDigestV2, attemptBaselineV2, attemptBindingRead, attemptConnectionBundleV2, attemptIdentityV2, dependencyManifestV2, dependencyProviderV2, expectedAgentManifestV2, executionManifestDigestV2, parityEvidenceV2, preflightFindingV2, preflightReportV2, prepareAttemptInputV2, projectMcpConnectionV2, secretFreeBindingV2, surfaceBindingV2, } from "./evals/attempt.js";
|
|
7
7
|
export { builtins, defineLocalScorer, scoreLocally } from "./evals/scorers.js";
|
package/dist/inline-files.d.ts
CHANGED
|
@@ -1,5 +1,9 @@
|
|
|
1
1
|
/** Inline file content larger than this many UTF-8 bytes is exported as its digest instead. */
|
|
2
2
|
export declare const INLINE_FILE_LIMIT: number;
|
|
3
|
+
/** Message text one record inspects for inline files in all, in UTF-16 code units. */
|
|
4
|
+
export declare const INLINE_FILE_TEXT_PER_RECORD: number;
|
|
5
|
+
/** Whether an attribute is one of the recorded message attributes that can inline files. */
|
|
6
|
+
export declare function isMessageKey(key: string): boolean;
|
|
3
7
|
/**
|
|
4
8
|
* Replaces inline files longer than {@link INLINE_FILE_LIMIT} in a recorded message attribute
|
|
5
9
|
* with their SHA-256 and byte size, so a span that inlines a large file exports the file's
|
package/dist/inline-files.js
CHANGED
|
@@ -3,32 +3,56 @@ import { MAX_BODY_BYTES } from "./config.js";
|
|
|
3
3
|
/** Inline file content larger than this many UTF-8 bytes is exported as its digest instead. */
|
|
4
4
|
export const INLINE_FILE_LIMIT = 64 * 1024;
|
|
5
5
|
const MAX_INLINE_FILE_TEXT = 8 * MAX_BODY_BYTES;
|
|
6
|
+
/** Message text one record inspects for inline files in all, in UTF-16 code units. */
|
|
7
|
+
export const INLINE_FILE_TEXT_PER_RECORD = 2 * MAX_INLINE_FILE_TEXT;
|
|
6
8
|
/** Message attributes whose JSON can inline files: GenAI blob parts and AI SDK 6 file parts. */
|
|
7
9
|
const messageKeys = new Set([
|
|
8
10
|
"gen_ai.input.messages",
|
|
9
11
|
"gen_ai.output.messages",
|
|
10
12
|
"ai.prompt.messages",
|
|
11
13
|
]);
|
|
14
|
+
/** Whether an attribute is one of the recorded message attributes that can inline files. */
|
|
15
|
+
export function isMessageKey(key) {
|
|
16
|
+
return messageKeys.has(key);
|
|
17
|
+
}
|
|
12
18
|
/** Strict base64: alphabet characters only, padded to a multiple of four. */
|
|
13
19
|
const base64 = /^[A-Za-z0-9+/]*={0,2}$/;
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
20
|
+
/** An RFC 2397 `data:` URL's media type and parameters, up to the comma before its data. */
|
|
21
|
+
const dataUrl = /^data:([^;,]*)((?:;[^;,]*)*),/;
|
|
22
|
+
const hexPair = /^[0-9A-Fa-f]{2}$/;
|
|
23
|
+
/** RFC 2397 data without `;base64`: percent-escaped octets, other characters as UTF-8. Undefined
|
|
24
|
+
* when a `%` does not start an escape. */
|
|
25
|
+
function percentDecoded(payload) {
|
|
26
|
+
// Escapes only shrink the text, so its UTF-8 length bounds the bytes.
|
|
27
|
+
const bytes = Buffer.alloc(Buffer.byteLength(payload, "utf8"));
|
|
28
|
+
let length = 0;
|
|
29
|
+
let start = 0;
|
|
30
|
+
for (let index = payload.indexOf("%"); index !== -1; index = payload.indexOf("%", start)) {
|
|
31
|
+
const hex = payload.slice(index + 1, index + 3);
|
|
32
|
+
if (!hexPair.test(hex))
|
|
33
|
+
return undefined;
|
|
34
|
+
length += bytes.write(payload.slice(start, index), length, "utf8");
|
|
35
|
+
bytes[length++] = Number.parseInt(hex, 16);
|
|
36
|
+
start = index + 3;
|
|
37
|
+
}
|
|
38
|
+
length += bytes.write(payload.slice(start), length, "utf8");
|
|
39
|
+
return bytes.subarray(0, length);
|
|
19
40
|
}
|
|
20
41
|
/**
|
|
21
|
-
* The bytes an inline file part carries
|
|
22
|
-
*
|
|
42
|
+
* The bytes an inline file part carries, the file's own bytes whatever its media type: a `data:`
|
|
43
|
+
* URL decoded by its own encoding (base64 with `;base64`, percent-escapes otherwise), content in
|
|
44
|
+
* the base64 alphabet decoded, and anything else, such as a text file's own text, as UTF-8. A
|
|
45
|
+
* `data:` URL whose data does not decode is taken as UTF-8 too.
|
|
23
46
|
*/
|
|
24
|
-
function fileBytes(content
|
|
25
|
-
const
|
|
26
|
-
const payload = content.slice(
|
|
27
|
-
|
|
28
|
-
payload
|
|
29
|
-
base64.test(payload)
|
|
30
|
-
|
|
31
|
-
|
|
47
|
+
function fileBytes(content) {
|
|
48
|
+
const url = dataUrl.exec(content);
|
|
49
|
+
const payload = url ? content.slice(url[0].length) : content;
|
|
50
|
+
const decoded = url && !url[2].split(";").includes("base64")
|
|
51
|
+
? percentDecoded(payload)
|
|
52
|
+
: payload.length % 4 === 0 && base64.test(payload)
|
|
53
|
+
? Buffer.from(payload, "base64")
|
|
54
|
+
: undefined;
|
|
55
|
+
return decoded ?? Buffer.from(content, "utf8");
|
|
32
56
|
}
|
|
33
57
|
/** The key holding a part's inline content: GenAI `blob` parts and AI SDK 6 `file` parts. */
|
|
34
58
|
function contentKey(part) {
|
|
@@ -44,9 +68,8 @@ function hashNode(value, state, depth) {
|
|
|
44
68
|
const part = value;
|
|
45
69
|
const key = contentKey(part);
|
|
46
70
|
const inline = key === undefined ? undefined : part[key];
|
|
47
|
-
const mimeType = part.mime_type ?? part.mediaType;
|
|
48
71
|
if (key !== undefined && typeof inline === "string") {
|
|
49
|
-
const bytes = fileBytes(inline
|
|
72
|
+
const bytes = fileBytes(inline);
|
|
50
73
|
if (bytes.byteLength > INLINE_FILE_LIMIT) {
|
|
51
74
|
const { [key]: _omitted, ...rest } = part;
|
|
52
75
|
state.changed = true;
|
package/dist/snapshot.d.ts
CHANGED
|
@@ -1,11 +1,11 @@
|
|
|
1
1
|
import type { ReadableSpan } from "@opentelemetry/sdk-trace";
|
|
2
2
|
import type { ReadableLogRecord, ReadWriteLogRecord } from "@opentelemetry/sdk-logs";
|
|
3
|
-
export declare function snapshotSpan(source: ReadableSpan, limit: number): {
|
|
3
|
+
export declare function snapshotSpan(source: ReadableSpan, limit: number, captureContent?: boolean): {
|
|
4
4
|
record: ReadableSpan;
|
|
5
5
|
bytes: number;
|
|
6
6
|
unresolvedResource: boolean;
|
|
7
7
|
};
|
|
8
|
-
export declare function snapshotLog(source: ReadableLogRecord, limit: number): {
|
|
8
|
+
export declare function snapshotLog(source: ReadableLogRecord, limit: number, captureContent?: boolean): {
|
|
9
9
|
record: ReadWriteLogRecord;
|
|
10
10
|
bytes: number;
|
|
11
11
|
unresolvedResource: boolean;
|
package/dist/snapshot.js
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import { createTraceState } from "@opentelemetry/api";
|
|
2
2
|
import { types as utilTypes } from "node:util";
|
|
3
3
|
import { resourceFromAttributes } from "@opentelemetry/resources";
|
|
4
|
+
import { hashInlineFiles, INLINE_FILE_LIMIT, INLINE_FILE_TEXT_PER_RECORD, isMessageKey, } from "./inline-files.js";
|
|
4
5
|
// Intrinsic accessors are captured once and invoked with an explicit receiver so a
|
|
5
6
|
// hostile object cannot override them; the unbound reference is the point.
|
|
6
7
|
// eslint-disable-next-line @typescript-eslint/unbound-method
|
|
@@ -10,20 +11,52 @@ const typedArraySet = Uint8Array.prototype.set;
|
|
|
10
11
|
/** Copies only exported data, with the same finite budget used for admission. */
|
|
11
12
|
class Snapshot {
|
|
12
13
|
limit;
|
|
14
|
+
captureContent;
|
|
13
15
|
bytes = 512;
|
|
14
16
|
unresolvedResource = false;
|
|
15
17
|
nodes = 0;
|
|
16
18
|
ancestors = new Set();
|
|
17
19
|
copied = new Map();
|
|
18
|
-
|
|
20
|
+
hashed = new Map();
|
|
21
|
+
inspected = 0;
|
|
22
|
+
/** In metadata-only mode the recorded messages, which export strips, are not copied at all. */
|
|
23
|
+
constructor(limit, captureContent = true) {
|
|
19
24
|
this.limit = limit;
|
|
25
|
+
this.captureContent = captureContent;
|
|
20
26
|
}
|
|
21
27
|
charge(bytes) {
|
|
22
28
|
this.bytes += bytes;
|
|
23
29
|
if (this.bytes > this.limit)
|
|
24
30
|
throw new RangeError("Telemetry byte budget exceeded");
|
|
25
31
|
}
|
|
26
|
-
|
|
32
|
+
/**
|
|
33
|
+
* An attribute map. A large inline file in a message attribute shrinks to its digest before it
|
|
34
|
+
* is charged: export replaces it anyway, and charging the file could drop the whole record.
|
|
35
|
+
*/
|
|
36
|
+
attributes(value) {
|
|
37
|
+
return this.copy(value, 1, "attributes");
|
|
38
|
+
}
|
|
39
|
+
/** A message attribute with its large inline files hashed. The same text is hashed once per
|
|
40
|
+
* record, and a record inspects at most {@link INLINE_FILE_TEXT_PER_RECORD} of it in all;
|
|
41
|
+
* beyond that, messages are charged as recorded. */
|
|
42
|
+
inlineFiles(key, value) {
|
|
43
|
+
// A UTF-16 unit is at most three UTF-8 bytes, so shorter text cannot hold a file to hash.
|
|
44
|
+
if (typeof value !== "string" || !isMessageKey(key) || value.length * 3 <= INLINE_FILE_LIMIT)
|
|
45
|
+
return value;
|
|
46
|
+
if (this.hashed.has(value))
|
|
47
|
+
return this.hashed.get(value);
|
|
48
|
+
if (this.inspected + value.length > INLINE_FILE_TEXT_PER_RECORD)
|
|
49
|
+
return value;
|
|
50
|
+
this.inspected += value.length;
|
|
51
|
+
const result = hashInlineFiles(key, value);
|
|
52
|
+
this.hashed.set(value, result);
|
|
53
|
+
return result;
|
|
54
|
+
}
|
|
55
|
+
/** Span events, whose attributes are attribute maps. */
|
|
56
|
+
events(value) {
|
|
57
|
+
return this.copy(value, 1, "events");
|
|
58
|
+
}
|
|
59
|
+
copy(value, depth = 0, shape = "value") {
|
|
27
60
|
if (++this.nodes > 16384 || depth > 32)
|
|
28
61
|
throw new RangeError("Telemetry complexity limit exceeded");
|
|
29
62
|
this.charge(16);
|
|
@@ -69,7 +102,7 @@ class Snapshot {
|
|
|
69
102
|
const descriptor = Object.getOwnPropertyDescriptor(value, index);
|
|
70
103
|
if (descriptor && !("value" in descriptor))
|
|
71
104
|
throw new TypeError("Telemetry accessors are unsupported");
|
|
72
|
-
copy.push(this.copy(descriptor?.value, depth + 1));
|
|
105
|
+
copy.push(this.copy(descriptor?.value, depth + 1, shape === "events" ? "event" : "value"));
|
|
73
106
|
}
|
|
74
107
|
}
|
|
75
108
|
else {
|
|
@@ -79,8 +112,10 @@ class Snapshot {
|
|
|
79
112
|
const descriptor = Object.getOwnPropertyDescriptor(value, key);
|
|
80
113
|
if (!descriptor || !("value" in descriptor))
|
|
81
114
|
throw new TypeError("Telemetry accessors are unsupported");
|
|
115
|
+
if (shape === "attributes" && !this.captureContent && isMessageKey(key))
|
|
116
|
+
continue;
|
|
82
117
|
this.charge(key.length * 2 + 16);
|
|
83
|
-
copy[key] = this.copy(descriptor.value, depth + 1);
|
|
118
|
+
copy[key] = this.copy(shape === "attributes" ? this.inlineFiles(key, descriptor.value) : descriptor.value, depth + 1, shape === "event" && key === "attributes" ? "attributes" : "value");
|
|
84
119
|
}
|
|
85
120
|
}
|
|
86
121
|
this.ancestors.delete(value);
|
|
@@ -126,14 +161,14 @@ class Snapshot {
|
|
|
126
161
|
function contextReader(context) {
|
|
127
162
|
return () => context;
|
|
128
163
|
}
|
|
129
|
-
export function snapshotSpan(source, limit) {
|
|
130
|
-
const snapshot = new Snapshot(limit);
|
|
164
|
+
export function snapshotSpan(source, limit, captureContent = true) {
|
|
165
|
+
const snapshot = new Snapshot(limit, captureContent);
|
|
131
166
|
const context = snapshot.context(source.spanContext());
|
|
132
167
|
if (source.links.length > 16384)
|
|
133
168
|
throw new RangeError("Link complexity limit exceeded");
|
|
134
169
|
const links = source.links.map((link) => ({
|
|
135
170
|
context: snapshot.context(link.context),
|
|
136
|
-
attributes: snapshot.
|
|
171
|
+
attributes: snapshot.attributes(link.attributes),
|
|
137
172
|
droppedAttributesCount: snapshot.copy(link.droppedAttributesCount),
|
|
138
173
|
}));
|
|
139
174
|
const record = {
|
|
@@ -145,13 +180,13 @@ export function snapshotSpan(source, limit) {
|
|
|
145
180
|
duration: source.duration,
|
|
146
181
|
ended: source.ended,
|
|
147
182
|
status: source.status,
|
|
148
|
-
attributes: source.attributes,
|
|
149
|
-
events: source.events,
|
|
150
183
|
instrumentationScope: source.instrumentationScope,
|
|
151
184
|
droppedAttributesCount: source.droppedAttributesCount,
|
|
152
185
|
droppedEventsCount: source.droppedEventsCount,
|
|
153
186
|
droppedLinksCount: source.droppedLinksCount,
|
|
154
187
|
}),
|
|
188
|
+
attributes: snapshot.attributes(source.attributes),
|
|
189
|
+
events: snapshot.events(source.events),
|
|
155
190
|
spanContext: contextReader(context),
|
|
156
191
|
parentSpanContext: snapshot.context(source.parentSpanContext),
|
|
157
192
|
links,
|
|
@@ -159,8 +194,8 @@ export function snapshotSpan(source, limit) {
|
|
|
159
194
|
};
|
|
160
195
|
return { record, bytes: snapshot.bytes, unresolvedResource: snapshot.unresolvedResource };
|
|
161
196
|
}
|
|
162
|
-
export function snapshotLog(source, limit) {
|
|
163
|
-
const snapshot = new Snapshot(limit);
|
|
197
|
+
export function snapshotLog(source, limit, captureContent = true) {
|
|
198
|
+
const snapshot = new Snapshot(limit, captureContent);
|
|
164
199
|
// Only our batching processor sees this copy. Its writer methods deliberately
|
|
165
200
|
// cannot mutate the admitted snapshot or invalidate its charged byte count.
|
|
166
201
|
const record = {
|
|
@@ -171,10 +206,10 @@ export function snapshotLog(source, limit) {
|
|
|
171
206
|
severityNumber: source.severityNumber,
|
|
172
207
|
eventName: source.eventName,
|
|
173
208
|
body: source.body,
|
|
174
|
-
attributes: source.attributes,
|
|
175
209
|
instrumentationScope: source.instrumentationScope,
|
|
176
210
|
droppedAttributesCount: source.droppedAttributesCount,
|
|
177
211
|
}),
|
|
212
|
+
attributes: snapshot.attributes(source.attributes),
|
|
178
213
|
spanContext: snapshot.context(source.spanContext),
|
|
179
214
|
resource: snapshot.resource(source.resource),
|
|
180
215
|
setAttribute() {
|
package/dist/transport.js
CHANGED
|
@@ -191,8 +191,8 @@ export class HueTransport {
|
|
|
191
191
|
const remaining = (advisory ? this.options.maxQueueBytes / 4 : this.options.maxQueueBytes) -
|
|
192
192
|
this.pendingBytes;
|
|
193
193
|
const snapshot = signal === "traces"
|
|
194
|
-
? snapshotSpan(record, remaining)
|
|
195
|
-
: snapshotLog(record, remaining);
|
|
194
|
+
? snapshotSpan(record, remaining, this.options.captureContent)
|
|
195
|
+
: snapshotLog(record, remaining, this.options.captureContent);
|
|
196
196
|
this.pendingBytes += snapshot.bytes;
|
|
197
197
|
if (signal === "traces")
|
|
198
198
|
this.spans.set(snapshot.record, snapshot.bytes);
|
package/dist/version.d.ts
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
1
|
/** Package version shared by the instrumentation scope and the export User-Agent. */
|
|
2
|
-
export declare const sdkVersion = "0.
|
|
2
|
+
export declare const sdkVersion = "0.10.0";
|
package/dist/version.js
CHANGED