agent-inspect 6.17.5 → 6.17.7

package/packages/cli/dist/index.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-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, 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, renderErrorLine, 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-GOHYSE7W.mjs';
+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, 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, renderErrorLine, 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-25LUDS4T.mjs';
 import { realpathSync, constants as constants$1, existsSync, readFileSync } from 'fs';
 import path18 from 'path';
 import { fileURLToPath, pathToFileURL } from 'url';
@@ -7,12 +7,13 @@ import { Command, Option } from 'commander';
 import { readFile, stat, mkdir, writeFile, unlink, appendFile, access, rm, open, readdir, constants } from 'fs/promises';
 import process2, { stdin, stdout } from 'process';
 import crypto, { createHash } from 'crypto';
+import { execFile, spawn } from 'child_process';
+import { promisify } from 'util';
 import { createServer } from 'http';
-import { spawn } from 'child_process';
 import { createRequire } from 'module';
 
 // package.json
-var version = "6.17.5";
+var version = "6.17.7";
 
 // packages/cli/src/trace-dir-scale.ts
 var TRACE_COUNT_WARN = 1e3;
@@ -3352,7 +3353,7 @@ async function newestTraceMtimeMs(traceDir) {
 }
 async function loadSessionRuns(traceDir) {
   try {
-    const mod = await import('./src-OYGG7QAZ.mjs');
+    const mod = await import('./src-2FS6BLR7.mjs');
     const dbPath = mod.resolveIndexDbPath(traceDir);
     const status = mod.indexStatus(dbPath);
     if (!status.healthy) {
@@ -3874,6 +3875,17 @@ function isCredentialSensitiveKey(key, sensitiveKeys) {
   return false;
 }
 
+// packages/redact/src/key-value-secret.ts
+var KEY_VALUE_SECRET_PATTERN_SOURCE = String.raw`\b(?:api[_-]?key|internal[_-]?token|access[_-]?token|auth[_-]?token|password|secret|token)=([^\s"'\\]{8,})`;
+var KEY_VALUE_SECRET_PATTERN = new RegExp(
+  KEY_VALUE_SECRET_PATTERN_SOURCE,
+  "i"
+);
+function valueContainsKeyValueSecret(value) {
+  KEY_VALUE_SECRET_PATTERN.lastIndex = 0;
+  return KEY_VALUE_SECRET_PATTERN.test(value);
+}
+
 // packages/redact/src/index.ts
 var DEFAULT_REDACT_KEYS2 = [
   "authorization",
@@ -4080,6 +4092,15 @@ var credentialDetectors = [
     pattern: /-----BEGIN [A-Z ]*PRIVATE KEY-----[\s\S]*-----END [A-Z ]*PRIVATE KEY-----/,
     severity: "error"
   }),
+  {
+    id: "value.keyValueSecret",
+    severity: "error",
+    matchKind: "value",
+    detect(input3) {
+      if (typeof input3.value !== "string") return [];
+      return valueContainsKeyValueSecret(input3.value) ? [{ action: "replace", severity: "error", matchKind: "value" }] : [];
+    }
+  },
   {
     id: "value.creditCard",
     severity: "error",
@@ -5786,6 +5807,7 @@ function mergeSafetyExtensions(result, read, options) {
 }
 
 // packages/cli/src/check.ts
+var execFileAsync = promisify(execFile);
 var DEFAULT_SELECT = ["run.status"];
 function uniqueSelectIds(ids) {
   const seen = /* @__PURE__ */ new Set();
@@ -5798,12 +5820,30 @@ function uniqueSelectIds(ids) {
   }
   return out;
 }
+function resolveForbiddenToolOptions(options) {
+  const merged = [...options.forbiddenTool ?? [], ...options.forbidTool ?? []];
+  const seen = /* @__PURE__ */ new Set();
+  const out = [];
+  for (const name of merged) {
+    if (seen.has(name)) continue;
+    seen.add(name);
+    out.push(name);
+  }
+  return out;
+}
+function withResolvedForbiddenTools(options) {
+  return {
+    ...options,
+    forbiddenTool: resolveForbiddenToolOptions(options),
+    forbidTool: void 0
+  };
+}
 function cliShorthandSelectIds(options) {
   const ids = [];
   if (options.failOnObservation !== void 0 && options.failOnObservation.trim() !== "") {
     ids.push("outcome.status");
   }
-  if ((options.requiredTool?.length ?? 0) > 0 || (options.forbiddenTool?.length ?? 0) > 0) {
+  if ((options.requiredTool?.length ?? 0) > 0 || (options.forbiddenTool?.length ?? 0) > 0 || (options.forbidTool?.length ?? 0) > 0) {
     ids.push("tool.usage");
   }
   if ((options.allowedModel?.length ?? 0) > 0 || options.maxTotalTokens !== void 0) {
@@ -6232,8 +6272,7 @@ async function loadConfig(configPath) {
       const raw = await readFile(absolute, "utf-8");
       return parseCheckConfig(JSON.parse(raw));
     }
-    const mod = await import(pathToFileURL(absolute).href);
-    return parseCheckConfig("default" in mod ? mod.default : mod);
+    return parseCheckConfig(await importJsCheckConfig(absolute));
   } catch (error) {
     if (error instanceof CheckConfigError) throw error;
     const message = error instanceof Error ? error.message : String(error);
@@ -6243,6 +6282,46 @@ async function loadConfig(configPath) {
     );
   }
 }
+async function importJsCheckConfig(absolute) {
+  const href = pathToFileURL(absolute).href;
+  if (!process.env.VITEST) {
+    const mod = await import(
+      /* @vite-ignore */
+      href
+    );
+    return "default" in mod ? mod.default : mod;
+  }
+  const script = `
+const mod = await import(${JSON.stringify(href)});
+const value = "default" in mod ? mod.default : mod;
+process.stdout.write(JSON.stringify({ ok: true, value }));
+`;
+  const env = { ...process.env };
+  delete env.NODE_OPTIONS;
+  delete env.VITEST;
+  delete env.VITEST_WORKER_ID;
+  const { stdout, stderr } = await execFileAsync(
+    process.execPath,
+    ["--input-type=module", "--eval", script],
+    {
+      encoding: "utf8",
+      maxBuffer: 4 * 1024 * 1024,
+      env
+    }
+  );
+  try {
+    const parsed = JSON.parse(stdout);
+    if (parsed.ok !== true) {
+      throw new Error(stderr.trim() || "Check config module did not return a value");
+    }
+    return parsed.value;
+  } catch (error) {
+    if (stderr.trim()) {
+      throw new Error(stderr.trim());
+    }
+    throw error instanceof Error ? error : new Error(String(error));
+  }
+}
 function normalizeConfig(config) {
   if (config.checks === void 0) return {};
   if (typeof config.checks !== "object" || Array.isArray(config.checks)) {
@@ -6485,6 +6564,7 @@ function readErrorResult(error) {
   );
 }
 async function checkCommand(target, options = {}, stdin = process.stdin) {
+  options = withResolvedForbiddenTools(options);
   let result;
   let phase = "config";
   let evidenceRead;
@@ -11591,7 +11671,7 @@ function isModuleNotFound4(e) {
 }
 async function loadIndexSqlite() {
   try {
-    return await import('./src-OYGG7QAZ.mjs');
+    return await import('./src-2FS6BLR7.mjs');
   } catch (e) {
     if (isModuleNotFound4(e)) {
       console.error(
@@ -12480,6 +12560,9 @@ function createCliProgram() {
   ]).option("--forbidden-tool <name>", "forbid a tool name (repeatable)", (value, previous = []) => [
     ...previous,
     value
+  ]).option("--forbid-tool <name>", "alias for --forbidden-tool (repeatable)", (value, previous = []) => [
+    ...previous,
+    value
   ]).option("--allowed-model <model>", "allow an LLM model (repeatable)", (value, previous = []) => [
     ...previous,
     value