agent-inspect 6.25.0 → 6.26.0
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 +14 -6
- package/README.md +1 -1
- package/docs/CLI.md +2 -1
- package/docs/KNOWN-ISSUES.md +5 -0
- package/docs/LIMITATIONS.md +2 -0
- package/docs/SESSIONS-AND-OUTCOMES.md +22 -2
- package/docs/TRACE-CONTRACTS.md +4 -1
- package/package.json +1 -1
- package/packages/cli/dist/{chunk-NHVN5AYW.mjs → chunk-HPR2OJPU.mjs} +2 -2
- package/packages/cli/dist/{chunk-NHVN5AYW.mjs.map → chunk-HPR2OJPU.mjs.map} +1 -1
- package/packages/cli/dist/index.cjs +22 -6
- package/packages/cli/dist/index.cjs.map +1 -1
- package/packages/cli/dist/index.mjs +25 -9
- package/packages/cli/dist/index.mjs.map +1 -1
- package/packages/cli/dist/{src-SXWEZTIS.mjs → src-J43ECFHC.mjs} +3 -3
- package/packages/cli/dist/{src-SXWEZTIS.mjs.map → src-J43ECFHC.mjs.map} +1 -1
- package/packages/core/dist/advanced.cjs +13 -10
- package/packages/core/dist/advanced.cjs.map +1 -1
- package/packages/core/dist/advanced.d.cts +9 -3
- package/packages/core/dist/advanced.d.ts +9 -3
- package/packages/core/dist/advanced.mjs +14 -13
- package/packages/core/dist/advanced.mjs.map +1 -1
- package/packages/core/dist/checks.cjs +176 -51
- package/packages/core/dist/checks.cjs.map +1 -1
- package/packages/core/dist/checks.d.cts +10 -5
- package/packages/core/dist/checks.d.ts +10 -5
- package/packages/core/dist/checks.mjs +1 -1
- package/packages/core/dist/{chunk-VMDCDWBE.mjs → chunk-LHSXRIYU.mjs} +178 -53
- package/packages/core/dist/chunk-LHSXRIYU.mjs.map +1 -0
- package/packages/core/dist/chunk-VMDCDWBE.mjs.map +0 -1
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
|
-
import { createSafetyRawContentRule, createSafetyRedactionRule, createSafetySecretPatternRule, createSafetyOversizedAttributeRule, resolveTraceDir, parseDuration, buildSessionIndex, resolveBundleRunIds, getTraceFilePath, openTrace, sha256Hex, collectTraceSchemaVersions, aggregateBundleSafeStatus, bundleFailsOnSafety, bundleRunAssetRelativePath, buildPlaceholderArtifact, EVIDENCE_HTML_FILENAME, EVIDENCE_MANIFEST_FILENAME, buildBundleMetadata, buildBundleSummaryMarkdown, buildEvidenceHtmlShell, buildEvidenceProvenanceViewHtml, EVIDENCE_FORMAT_VERSION, inferEvidenceFileRole, buildEvidenceSafetyViewHtml, buildEvidenceDiffViewHtml, buildEvidenceCircuitViewHtml, buildEvidenceOutcomesViewHtml, buildEvidenceContractsViewHtml, buildEvidenceToolsLlmViewHtml, buildEvidenceCausalFailureViewHtml, buildEvidenceTimelineViewHtml, buildEvidenceTreeViewHtml, buildEvidenceManifest, serializeEvidenceManifest, buildZipArchive, assertBundlePathContained, enrichSessionRunRecord, normalizeBundleOutputPath, sanitizeBundleRunId, defaultBundleOutputPath, runTraceChecks, assertEvidenceRelativePath, TraceDirectory, loadTraceMetadataList, loadSessionRunRecords, TraceReadError, resolveRedactionProfile, applyProfileMetadataCaps, Redactor, flattenTree, escapeHtml, safeString, compactAttributes, stableJson, escapeMarkdown, truncateStringForProfile, extractMetadata, filterTraces, truncateName, buildRunSummary, formatDuration, formatTimestamp, renderErrorLine, isAgentInspectTrace, filterMetasBySessionScope, aggregateSessionCheckResults, createBaselineRegressionRule, summarizeSemanticParity, buildEvidenceCiPackage, verifyEvidenceDirectory, diffTraceEvents, renderRunDiff, buildRunTimeline, renderTimeline, buildTraceStats, renderTraceStats, parseDurationFilter, searchTraces, formatStepLabel, buildActivitySummary, renderActivitySummaryHuman, buildRunWhatSummary, renderRunWhat, buildLocalExplanation, resolveSuiteTemplate, defaultSuiteConfigTemplate, loadSuiteConfig, validateSuiteConfig, runSuite, renderSuiteReport, analyzeCohort, renderCohortReport, gateHasThresholds, runGate, renderGateReport, persistedInspectEventsToTraceEvents, renderStepLine, getIndent, TreeBuilder, zeroKinds, validateEvent, traceEventToPersistedInspectEvent, isPersistedInspectEvent, createRunStatusRule, createStructureOrphanRule, createStructureCycleRule, createRunDurationRule, createMaxStepDurationRule, createRequireCompletedRule, createStallDetectionRule, createObservedOutcomeRule, createRunDepthRule, createToolUsageRule, createLlmUsageRule, createStructureRelationshipRule, createStructureParallelWidthRule, summarizeObservedOutcomes, extractOutcomesFromTraceEvents, renderObservedOutcomesMarkdown, renderObservedOutcomesHtml, DEFAULT_SUITE_ARTIFACTS_DIR, parseCohortMetricList, COHORT_METRIC_IDS, parseGateList, nanoid, diffRuns, manualTraceEventsToComparableRun } from './chunk-
|
|
2
|
+
import { createSafetyRawContentRule, createSafetyRedactionRule, createSafetySecretPatternRule, createSafetyOversizedAttributeRule, resolveTraceDir, parseDuration, buildSessionIndex, resolveBundleRunIds, getTraceFilePath, openTrace, sha256Hex, collectTraceSchemaVersions, aggregateBundleSafeStatus, bundleFailsOnSafety, bundleRunAssetRelativePath, buildPlaceholderArtifact, EVIDENCE_HTML_FILENAME, EVIDENCE_MANIFEST_FILENAME, buildBundleMetadata, buildBundleSummaryMarkdown, buildEvidenceHtmlShell, buildEvidenceProvenanceViewHtml, EVIDENCE_FORMAT_VERSION, inferEvidenceFileRole, buildEvidenceSafetyViewHtml, buildEvidenceDiffViewHtml, buildEvidenceCircuitViewHtml, buildEvidenceOutcomesViewHtml, buildEvidenceContractsViewHtml, buildEvidenceToolsLlmViewHtml, buildEvidenceCausalFailureViewHtml, buildEvidenceTimelineViewHtml, buildEvidenceTreeViewHtml, buildEvidenceManifest, serializeEvidenceManifest, buildZipArchive, assertBundlePathContained, enrichSessionRunRecord, normalizeBundleOutputPath, sanitizeBundleRunId, defaultBundleOutputPath, runTraceChecks, assertEvidenceRelativePath, TraceDirectory, loadTraceMetadataList, loadSessionRunRecords, TraceReadError, resolveRedactionProfile, applyProfileMetadataCaps, Redactor, flattenTree, escapeHtml, safeString, compactAttributes, stableJson, escapeMarkdown, truncateStringForProfile, extractMetadata, filterTraces, truncateName, buildRunSummary, formatDuration, formatTimestamp, renderErrorLine, isAgentInspectTrace, filterMetasBySessionScope, aggregateSessionCheckResults, createBaselineRegressionRule, summarizeSemanticParity, buildEvidenceCiPackage, verifyEvidenceDirectory, diffTraceEvents, renderRunDiff, buildRunTimeline, renderTimeline, buildTraceStats, renderTraceStats, parseDurationFilter, searchTraces, formatStepLabel, buildActivitySummary, renderActivitySummaryHuman, buildRunWhatSummary, renderRunWhat, buildLocalExplanation, resolveSuiteTemplate, defaultSuiteConfigTemplate, loadSuiteConfig, validateSuiteConfig, runSuite, renderSuiteReport, analyzeCohort, renderCohortReport, gateHasThresholds, runGate, renderGateReport, persistedInspectEventsToTraceEvents, renderStepLine, getIndent, TreeBuilder, zeroKinds, validateEvent, traceEventToPersistedInspectEvent, isPersistedInspectEvent, createRunStatusRule, createStructureOrphanRule, createStructureCycleRule, createRunDurationRule, createMaxStepDurationRule, createRequireCompletedRule, createStallDetectionRule, createObservedOutcomeRule, createRunDepthRule, createToolUsageRule, createLlmUsageRule, createStructureRelationshipRule, createStructureParallelWidthRule, summarizeObservedOutcomes, extractOutcomesFromTraceEvents, renderObservedOutcomesMarkdown, renderObservedOutcomesHtml, DEFAULT_SUITE_ARTIFACTS_DIR, parseCohortMetricList, COHORT_METRIC_IDS, parseGateList, nanoid, diffRuns, manualTraceEventsToComparableRun } from './chunk-HPR2OJPU.mjs';
|
|
3
3
|
import { realpathSync, constants as constants$1, existsSync, readFileSync } from 'fs';
|
|
4
4
|
import path19 from 'path';
|
|
5
5
|
import { fileURLToPath, pathToFileURL } from 'url';
|
|
@@ -13,7 +13,7 @@ import { createServer } from 'http';
|
|
|
13
13
|
import { createRequire } from 'module';
|
|
14
14
|
|
|
15
15
|
// package.json
|
|
16
|
-
var version = "6.
|
|
16
|
+
var version = "6.26.0";
|
|
17
17
|
|
|
18
18
|
// packages/cli/src/trace-dir-scale.ts
|
|
19
19
|
var TRACE_COUNT_WARN = 1e3;
|
|
@@ -3397,7 +3397,7 @@ async function newestTraceMtimeMs(traceDir) {
|
|
|
3397
3397
|
}
|
|
3398
3398
|
async function loadSessionRuns(traceDir) {
|
|
3399
3399
|
try {
|
|
3400
|
-
const mod = await import('./src-
|
|
3400
|
+
const mod = await import('./src-J43ECFHC.mjs');
|
|
3401
3401
|
const dbPath = mod.resolveIndexDbPath(traceDir);
|
|
3402
3402
|
const status = mod.indexStatus(dbPath);
|
|
3403
3403
|
if (!status.healthy) {
|
|
@@ -6639,9 +6639,9 @@ function unionCheckSelect(input3) {
|
|
|
6639
6639
|
function resolvePreset(preset, context = {}) {
|
|
6640
6640
|
if (preset === void 0 || preset.trim() === "") return void 0;
|
|
6641
6641
|
const name = preset.trim().toLowerCase();
|
|
6642
|
-
if (name !== "trajectory" && name !== "safety" && name !== "comprehensive") {
|
|
6642
|
+
if (name !== "trajectory" && name !== "safety" && name !== "comprehensive" && name !== "behavioral-session") {
|
|
6643
6643
|
throw new Error(
|
|
6644
|
-
`Unknown --preset "${preset}". Use trajectory, safety, or
|
|
6644
|
+
`Unknown --preset "${preset}". Use trajectory, safety, comprehensive, or behavioral-session.`
|
|
6645
6645
|
);
|
|
6646
6646
|
}
|
|
6647
6647
|
const trajectorySelect = [
|
|
@@ -6676,6 +6676,15 @@ function resolvePreset(preset, context = {}) {
|
|
|
6676
6676
|
select: safetySelect
|
|
6677
6677
|
};
|
|
6678
6678
|
}
|
|
6679
|
+
if (name === "behavioral-session") {
|
|
6680
|
+
return {
|
|
6681
|
+
requireCompleted: true,
|
|
6682
|
+
enableSafetyRedaction: false,
|
|
6683
|
+
enableStructureRelationshipDefaults: false,
|
|
6684
|
+
select: ["run.requireCompleted", "outcome.status", "structure.orphan"],
|
|
6685
|
+
failOnObservation: "failed"
|
|
6686
|
+
};
|
|
6687
|
+
}
|
|
6679
6688
|
const select = [.../* @__PURE__ */ new Set([...trajectorySelect, ...safetySelect])];
|
|
6680
6689
|
return {
|
|
6681
6690
|
requireCompleted: true,
|
|
@@ -7109,7 +7118,8 @@ function applyResolvedPreset(config, options, resolved) {
|
|
|
7109
7118
|
config: { checks: checks2 },
|
|
7110
7119
|
options: {
|
|
7111
7120
|
...options,
|
|
7112
|
-
...resolved.requireCompleted ? { requireCompleted: true } : {}
|
|
7121
|
+
...resolved.requireCompleted ? { requireCompleted: true } : {},
|
|
7122
|
+
...resolved.failOnObservation !== void 0 && (options.failOnObservation === void 0 || options.failOnObservation.trim() === "") ? { failOnObservation: resolved.failOnObservation } : {}
|
|
7113
7123
|
}
|
|
7114
7124
|
};
|
|
7115
7125
|
}
|
|
@@ -7255,7 +7265,13 @@ function isSafetyFinding(ruleId) {
|
|
|
7255
7265
|
}
|
|
7256
7266
|
function printPresetClassSummary(result, preset) {
|
|
7257
7267
|
const name = preset?.trim().toLowerCase();
|
|
7258
|
-
if (name !== "trajectory" && name !== "safety" && name !== "comprehensive") {
|
|
7268
|
+
if (name !== "trajectory" && name !== "safety" && name !== "comprehensive" && name !== "behavioral-session") {
|
|
7269
|
+
return;
|
|
7270
|
+
}
|
|
7271
|
+
if (name === "behavioral-session") {
|
|
7272
|
+
console.log(
|
|
7273
|
+
`Behavioral session: ${result.status === "pass" ? "PASS" : "FAIL"} (outcomes scored; tool errors may be expected)`
|
|
7274
|
+
);
|
|
7259
7275
|
return;
|
|
7260
7276
|
}
|
|
7261
7277
|
const hasSafetyFindings = result.findings.some(
|
|
@@ -12063,7 +12079,7 @@ function isModuleNotFound4(e) {
|
|
|
12063
12079
|
}
|
|
12064
12080
|
async function loadIndexSqlite() {
|
|
12065
12081
|
try {
|
|
12066
|
-
return await import('./src-
|
|
12082
|
+
return await import('./src-J43ECFHC.mjs');
|
|
12067
12083
|
} catch (e) {
|
|
12068
12084
|
if (isModuleNotFound4(e)) {
|
|
12069
12085
|
console.error(
|
|
@@ -12991,7 +13007,7 @@ function createCliProgram() {
|
|
|
12991
13007
|
"run optional circuit rules (repeatable): same-tool-repetition, max-retries, ...",
|
|
12992
13008
|
(value, previous = []) => [...previous, value]
|
|
12993
13009
|
).addOption(
|
|
12994
|
-
new Option("--preset <name>", "additive check preset").choices(["trajectory", "safety", "comprehensive"])
|
|
13010
|
+
new Option("--preset <name>", "additive check preset").choices(["trajectory", "safety", "comprehensive", "behavioral-session"])
|
|
12995
13011
|
).addOption(
|
|
12996
13012
|
new Option(
|
|
12997
13013
|
"--evidence-on <mode>",
|