agent-inspect 5.2.0 → 5.4.0
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 -0
- package/docs/CLI.md +28 -1
- package/package.json +1 -1
- package/packages/cli/dist/{chunk-SJ5R2XBE.mjs → chunk-YS2I722C.mjs} +112 -3
- package/packages/cli/dist/chunk-YS2I722C.mjs.map +1 -0
- package/packages/cli/dist/index.cjs +2487 -2163
- package/packages/cli/dist/index.cjs.map +1 -1
- package/packages/cli/dist/index.mjs +2313 -2112
- package/packages/cli/dist/index.mjs.map +1 -1
- package/packages/cli/dist/{src-PK22BBMH.mjs → src-FVNE44UA.mjs} +3 -3
- package/packages/cli/dist/{src-PK22BBMH.mjs.map → src-FVNE44UA.mjs.map} +1 -1
- package/packages/core/dist/advanced.cjs +118 -0
- package/packages/core/dist/advanced.cjs.map +1 -1
- package/packages/core/dist/advanced.d.cts +14 -4
- package/packages/core/dist/advanced.d.ts +14 -4
- package/packages/core/dist/advanced.mjs +114 -1
- package/packages/core/dist/advanced.mjs.map +1 -1
- package/packages/core/dist/checks.d.cts +1 -1
- package/packages/core/dist/checks.d.ts +1 -1
- package/packages/core/dist/{context-Cp8NRbBX.d.ts → context-BJ5UQLDi.d.ts} +1 -1
- package/packages/core/dist/{context-BbfkMYjz.d.cts → context-DyKbkmd4.d.cts} +1 -1
- package/packages/core/dist/index.d.cts +2 -2
- package/packages/core/dist/index.d.ts +2 -2
- package/packages/core/dist/{types-CvsKGweN.d.cts → types-J3Xtuf9x.d.cts} +13 -1
- package/packages/core/dist/{types-CvsKGweN.d.ts → types-J3Xtuf9x.d.ts} +13 -1
- package/packages/cli/dist/chunk-SJ5R2XBE.mjs.map +0 -1
|
@@ -8446,6 +8446,118 @@ function renderSuiteReport(result, options = {}) {
|
|
|
8446
8446
|
return renderSuiteReportMarkdown(result);
|
|
8447
8447
|
}
|
|
8448
8448
|
|
|
8449
|
+
// packages/core/src/suite/templates.ts
|
|
8450
|
+
var SUITE_TEMPLATE_IDS = [
|
|
8451
|
+
"customer-support-agent",
|
|
8452
|
+
"refund-agent",
|
|
8453
|
+
"sales-assistant",
|
|
8454
|
+
"browser-task-agent",
|
|
8455
|
+
"mcp-tool-agent",
|
|
8456
|
+
"workflow-agent",
|
|
8457
|
+
"rag-answer-agent",
|
|
8458
|
+
"human-approval-agent"
|
|
8459
|
+
];
|
|
8460
|
+
function baseTemplate(name, cases) {
|
|
8461
|
+
return {
|
|
8462
|
+
name,
|
|
8463
|
+
traces: "./.agent-inspect",
|
|
8464
|
+
cases,
|
|
8465
|
+
checks: {
|
|
8466
|
+
select: ["run.status", "outcome.status"],
|
|
8467
|
+
run: { maxDurationMs: 6e4 }
|
|
8468
|
+
},
|
|
8469
|
+
redactionProfile: "local",
|
|
8470
|
+
artifacts: { outputDir: ".agent-inspect/suite-runs" }
|
|
8471
|
+
};
|
|
8472
|
+
}
|
|
8473
|
+
var TEMPLATES = {
|
|
8474
|
+
"customer-support-agent": baseTemplate("customer-support-agent", [
|
|
8475
|
+
{
|
|
8476
|
+
id: "support-ticket",
|
|
8477
|
+
runId: "support-ticket",
|
|
8478
|
+
requireTools: ["searchDocs"],
|
|
8479
|
+
forbidTools: ["deleteAccount"],
|
|
8480
|
+
expectedObservations: ["answerReady"],
|
|
8481
|
+
maxDurationMs: 45e3
|
|
8482
|
+
}
|
|
8483
|
+
]),
|
|
8484
|
+
"refund-agent": baseTemplate("refund-agent", [
|
|
8485
|
+
{
|
|
8486
|
+
id: "refund-flow",
|
|
8487
|
+
runId: "refund-flow",
|
|
8488
|
+
requireTools: ["searchDocs"],
|
|
8489
|
+
forbidTools: ["deleteAccount"],
|
|
8490
|
+
expectedObservations: ["policyShown"],
|
|
8491
|
+
maxDurationMs: 3e4
|
|
8492
|
+
}
|
|
8493
|
+
]),
|
|
8494
|
+
"sales-assistant": baseTemplate("sales-assistant", [
|
|
8495
|
+
{
|
|
8496
|
+
id: "quote-request",
|
|
8497
|
+
runId: "quote-request",
|
|
8498
|
+
requireTools: ["lookupPricing"],
|
|
8499
|
+
maxDurationMs: 3e4
|
|
8500
|
+
}
|
|
8501
|
+
]),
|
|
8502
|
+
"browser-task-agent": baseTemplate("browser-task-agent", [
|
|
8503
|
+
{
|
|
8504
|
+
id: "browser-nav",
|
|
8505
|
+
runId: "browser-nav",
|
|
8506
|
+
requireTools: ["browserNavigate"],
|
|
8507
|
+
maxDurationMs: 12e4
|
|
8508
|
+
}
|
|
8509
|
+
]),
|
|
8510
|
+
"mcp-tool-agent": baseTemplate("mcp-tool-agent", [
|
|
8511
|
+
{
|
|
8512
|
+
id: "mcp-call",
|
|
8513
|
+
runId: "mcp-call",
|
|
8514
|
+
requireTools: ["mcp:tools/list"],
|
|
8515
|
+
maxDurationMs: 3e4
|
|
8516
|
+
}
|
|
8517
|
+
]),
|
|
8518
|
+
"workflow-agent": baseTemplate("workflow-agent", [
|
|
8519
|
+
{
|
|
8520
|
+
id: "workflow-handoff",
|
|
8521
|
+
runId: "workflow-handoff",
|
|
8522
|
+
requireTools: ["handoff"],
|
|
8523
|
+
maxDurationMs: 6e4
|
|
8524
|
+
}
|
|
8525
|
+
]),
|
|
8526
|
+
"rag-answer-agent": baseTemplate("rag-answer-agent", [
|
|
8527
|
+
{
|
|
8528
|
+
id: "rag-answer",
|
|
8529
|
+
runId: "rag-answer",
|
|
8530
|
+
requireTools: ["retrieve", "answer"],
|
|
8531
|
+
expectedObservations: ["citationShown"],
|
|
8532
|
+
maxDurationMs: 45e3
|
|
8533
|
+
}
|
|
8534
|
+
]),
|
|
8535
|
+
"human-approval-agent": baseTemplate("human-approval-agent", [
|
|
8536
|
+
{
|
|
8537
|
+
id: "approval-gate",
|
|
8538
|
+
runId: "approval-gate",
|
|
8539
|
+
expectedObservations: ["approvalRequested", "approvalGranted"],
|
|
8540
|
+
maxDurationMs: 6e4
|
|
8541
|
+
}
|
|
8542
|
+
])
|
|
8543
|
+
};
|
|
8544
|
+
function listSuiteTemplates() {
|
|
8545
|
+
return [...SUITE_TEMPLATE_IDS];
|
|
8546
|
+
}
|
|
8547
|
+
function getSuiteTemplate(id) {
|
|
8548
|
+
if (!SUITE_TEMPLATE_IDS.includes(id)) return void 0;
|
|
8549
|
+
return structuredClone(TEMPLATES[id]);
|
|
8550
|
+
}
|
|
8551
|
+
function resolveSuiteTemplate(id) {
|
|
8552
|
+
const template = getSuiteTemplate(id);
|
|
8553
|
+
if (template === void 0) {
|
|
8554
|
+
throw new Error(
|
|
8555
|
+
`Unknown suite template "${id}". Available: ${SUITE_TEMPLATE_IDS.join(", ")}`
|
|
8556
|
+
);
|
|
8557
|
+
}
|
|
8558
|
+
return template;
|
|
8559
|
+
}
|
|
8560
|
+
|
|
8449
8561
|
// packages/core/src/cohort/types.ts
|
|
8450
8562
|
var COHORT_METRIC_IDS = [
|
|
8451
8563
|
"errorRate",
|
|
@@ -9506,6 +9618,7 @@ exports.MAX_TERMINAL_DEPTH = MAX_TERMINAL_DEPTH;
|
|
|
9506
9618
|
exports.MAX_TERMINAL_NAME_LENGTH = MAX_TERMINAL_NAME_LENGTH;
|
|
9507
9619
|
exports.RUNS_DIR_NAME = RUNS_DIR_NAME;
|
|
9508
9620
|
exports.SESSION_WORKFLOW_KEYS = SESSION_WORKFLOW_KEYS;
|
|
9621
|
+
exports.SUITE_TEMPLATE_IDS = SUITE_TEMPLATE_IDS;
|
|
9509
9622
|
exports.TERMINAL_INDENT = TERMINAL_INDENT;
|
|
9510
9623
|
exports.TraceDirectory = TraceDirectory;
|
|
9511
9624
|
exports.aggregateBundleSafeStatus = aggregateBundleSafeStatus;
|
|
@@ -9534,6 +9647,8 @@ exports.enrichSessionRunRecord = enrichSessionRunRecord;
|
|
|
9534
9647
|
exports.enrichSessionSummary = enrichSessionSummary;
|
|
9535
9648
|
exports.ensureTraceDir = ensureTraceDir;
|
|
9536
9649
|
exports.extractMetadata = extractMetadata;
|
|
9650
|
+
exports.extractOutcomesFromPersistedEvents = extractOutcomesFromPersistedEvents;
|
|
9651
|
+
exports.extractOutcomesFromTraceEvents = extractOutcomesFromTraceEvents;
|
|
9537
9652
|
exports.extractSessionWorkflowMetadata = extractSessionWorkflowMetadata;
|
|
9538
9653
|
exports.filterMetasBySessionScope = filterMetasBySessionScope;
|
|
9539
9654
|
exports.filterTraces = filterTraces;
|
|
@@ -9552,6 +9667,7 @@ exports.getDefaultTraceDir = getDefaultTraceDir;
|
|
|
9552
9667
|
exports.getIndent = getIndent;
|
|
9553
9668
|
exports.getParentStepId = getParentStepId;
|
|
9554
9669
|
exports.getRunIdFromTraceFileName = getRunIdFromTraceFileName;
|
|
9670
|
+
exports.getSuiteTemplate = getSuiteTemplate;
|
|
9555
9671
|
exports.getTraceDirFromContext = getTraceDirFromContext;
|
|
9556
9672
|
exports.getTraceFilePath = getTraceFilePath;
|
|
9557
9673
|
exports.getTraceSafetyFromContext = getTraceSafetyFromContext;
|
|
@@ -9564,6 +9680,7 @@ exports.isSilentContext = isSilentContext;
|
|
|
9564
9680
|
exports.isStepStatus = isStepStatus;
|
|
9565
9681
|
exports.isStepType = isStepType;
|
|
9566
9682
|
exports.isTraceEvent = isTraceEvent;
|
|
9683
|
+
exports.listSuiteTemplates = listSuiteTemplates;
|
|
9567
9684
|
exports.listTraceFiles = listTraceFiles;
|
|
9568
9685
|
exports.loadSessionRunRecords = loadSessionRunRecords;
|
|
9569
9686
|
exports.loadSuiteConfig = loadSuiteConfig;
|
|
@@ -9607,6 +9724,7 @@ exports.resolveBundleRunIds = resolveBundleRunIds;
|
|
|
9607
9724
|
exports.resolveRedactionProfile = resolveRedactionProfile;
|
|
9608
9725
|
exports.resolveSuiteCaseTrace = resolveSuiteCaseTrace;
|
|
9609
9726
|
exports.resolveSuiteConfigPath = resolveSuiteConfigPath;
|
|
9727
|
+
exports.resolveSuiteTemplate = resolveSuiteTemplate;
|
|
9610
9728
|
exports.resolveTraceDir = resolveTraceDir;
|
|
9611
9729
|
exports.resolveTraceSafetyOptions = resolveTraceSafetyOptions;
|
|
9612
9730
|
exports.runGate = runGate;
|