agent-inspect 6.31.8 → 6.31.9

package/packages/core/dist/advanced.d.cts

@@ -1269,7 +1269,7 @@ interface EvidenceCiPackageFiles {
 declare function buildEvidenceCiPackage(input: EvidenceCiPackageInput): EvidenceCiPackageFiles;
 
 type SuiteCaseStatus = "pass" | "fail" | "error" | "skipped";
-type SuiteDiagnosticCode = "AI_SUITE_CONFIG_INVALID" | "AI_SUITE_CONFIG_LOAD_FAILED" | "AI_SUITE_CASE_TRACE_MISSING" | "AI_SUITE_CASE_CHECK_FAILED" | "AI_SUITE_CASE_EVAL_FAILED" | "AI_SUITE_CASE_OBSERVATION_FAILED" | "AI_SUITE_TRACE_UNREADABLE";
+type SuiteDiagnosticCode = "AI_SUITE_CONFIG_INVALID" | "AI_SUITE_CONFIG_LOAD_FAILED" | "AI_SUITE_CASE_TRACE_MISSING" | "AI_SUITE_CASE_CHECK_FAILED" | "AI_SUITE_CASE_EVAL_FAILED" | "AI_SUITE_CASE_OBSERVATION_FAILED" | "AI_SUITE_NO_ASSERTIONS" | "AI_SUITE_UNKNOWN_SELECTOR" | "AI_SUITE_TRACE_UNREADABLE";
 interface SuiteDiagnostic {
     code: SuiteDiagnosticCode;
     severity: "error" | "warning" | "info";

package/packages/core/dist/advanced.d.ts

@@ -1269,7 +1269,7 @@ interface EvidenceCiPackageFiles {
 declare function buildEvidenceCiPackage(input: EvidenceCiPackageInput): EvidenceCiPackageFiles;
 
 type SuiteCaseStatus = "pass" | "fail" | "error" | "skipped";
-type SuiteDiagnosticCode = "AI_SUITE_CONFIG_INVALID" | "AI_SUITE_CONFIG_LOAD_FAILED" | "AI_SUITE_CASE_TRACE_MISSING" | "AI_SUITE_CASE_CHECK_FAILED" | "AI_SUITE_CASE_EVAL_FAILED" | "AI_SUITE_CASE_OBSERVATION_FAILED" | "AI_SUITE_TRACE_UNREADABLE";
+type SuiteDiagnosticCode = "AI_SUITE_CONFIG_INVALID" | "AI_SUITE_CONFIG_LOAD_FAILED" | "AI_SUITE_CASE_TRACE_MISSING" | "AI_SUITE_CASE_CHECK_FAILED" | "AI_SUITE_CASE_EVAL_FAILED" | "AI_SUITE_CASE_OBSERVATION_FAILED" | "AI_SUITE_NO_ASSERTIONS" | "AI_SUITE_UNKNOWN_SELECTOR" | "AI_SUITE_TRACE_UNREADABLE";
 interface SuiteDiagnostic {
     code: SuiteDiagnosticCode;
     severity: "error" | "warning" | "info";

package/packages/core/dist/advanced.mjs

@@ -1,6 +1,6 @@
-import { extractSessionWorkflowMetadata, sessionKeyForRun, defineTraceContract, runTraceChecks, createRunStatusRule, createToolUsageRule, createRunDurationRule, createLlmUsageRule, createObservedOutcomeRule } from './chunk-2CCRUI42.mjs';
+import { extractSessionWorkflowMetadata, sessionKeyForRun, defineTraceContract, runTraceChecks, createRunStatusRule, createToolUsageRule, createRunDurationRule, createRunDepthRule, createToolFailureRule, createLlmUsageRule, createObservedOutcomeRule } from './chunk-2CCRUI42.mjs';
 export { extractSessionWorkflowMetadata, sessionKeyForRun } from './chunk-2CCRUI42.mjs';
-export { MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, TERMINAL_INDENT, TraceDirectory, createInspector, createInspectorRuntime, formatTerminalName, getCurrentContext, getCurrentCorrelationMetadata, getCurrentDepth, getCurrentRunId, getCurrentRunName, getCurrentStepId, getIndent, getParentStepId, getTraceDirFromContext, getTraceSafetyFromContext, hasActiveContext, isAgentInspectEnabled, isSilentContext, maybeInspectRun, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, renderErrorLine, renderRunSummary, renderStepLine, resolveTraceDir, runWithContext, runWithStepContext } from './chunk-VQUB3WRI.mjs';
+export { MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, TERMINAL_INDENT, TraceDirectory, createInspector, createInspectorRuntime, formatTerminalName, getCurrentContext, getCurrentCorrelationMetadata, getCurrentDepth, getCurrentRunId, getCurrentRunName, getCurrentStepId, getIndent, getParentStepId, getTraceDirFromContext, getTraceSafetyFromContext, hasActiveContext, isAgentInspectEnabled, isSilentContext, maybeInspectRun, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, renderErrorLine, renderRunSummary, renderStepLine, resolveTraceDir, runWithContext, runWithStepContext } from './chunk-3SPBELJB.mjs';
 import { buildRunTimeline, filterTraces, extractMetadata, buildRunSummary } from './chunk-7YYS742B.mjs';
 export { buildRunSummary, buildRunTimeline, buildRunWhatSummary, buildTraceStats, extractMetadata, filterTraces, formatStepLabel, renderRunWhat, renderTimeline, renderTraceStats } from './chunk-7YYS742B.mjs';
 import { extractOutcomesFromTraceEvents, parseObservationFilter, summarizeObservedOutcomes, extractOutcomesFromPersistedEvents, renderObservedOutcomesHtml } from './chunk-N5PZAI6Q.mjs';
@@ -3626,21 +3626,48 @@ async function resolveSuiteCaseTrace(suiteCase, options) {
     reason: `no trace found for run id "${runKey}" under ${options.tracesDir}`
   };
 }
+var KNOWN_SUITE_SELECT_IDS = /* @__PURE__ */ new Set([
+  "run.status",
+  "run.duration",
+  "run.depth",
+  "outcome.status",
+  "tool.usage",
+  "tool.failures",
+  "llm.usage"
+]);
 function diagnostic3(code, message, severity = "error", caseId) {
   return { code, message, severity, ...caseId !== void 0 ? { caseId } : {} };
 }
-function buildCaseRules(suiteCase, config) {
+function buildCaseAssertions(suiteCase, config) {
   const rules = [];
   const select = new Set(config.checks?.select ?? []);
-  if (select.has("run.status")) {
+  const configDiagnostics = [];
+  const evalConfig = config.eval;
+  const declaredSelect = [...select];
+  for (const id of declaredSelect) {
+    if (!KNOWN_SUITE_SELECT_IDS.has(id)) {
+      configDiagnostics.push(
+        diagnostic3(
+          "AI_SUITE_UNKNOWN_SELECTOR",
+          `Unknown or unsupported suite check selector "${id}".`,
+          "error",
+          suiteCase.id
+        )
+      );
+    }
+  }
+  if (select.has("run.status") || evalConfig?.requireSuccess === true) {
     rules.push(createRunStatusRule());
+    select.add("run.status");
   }
   const requiredTools = [
     ...config.checks?.tool?.required ?? [],
+    ...evalConfig?.requiredTools ?? [],
     ...suiteCase.requireTools ?? []
   ];
   const forbiddenTools = [
     ...config.checks?.tool?.forbidden ?? [],
+    ...evalConfig?.forbiddenTools ?? [],
     ...suiteCase.forbidTools ?? []
   ];
   if (requiredTools.length > 0 || forbiddenTools.length > 0) {
@@ -3652,20 +3679,51 @@ function buildCaseRules(suiteCase, config) {
     );
     select.add("tool.usage");
   }
-  const maxDurationMs = suiteCase.maxDurationMs ?? config.checks?.run?.maxDurationMs ?? config.eval?.maxDurationMs;
+  const maxDurationMs = suiteCase.maxDurationMs ?? config.checks?.run?.maxDurationMs ?? evalConfig?.maxDurationMs;
   if (maxDurationMs !== void 0) {
     rules.push(createRunDurationRule({ maxDurationMs }));
     select.add("run.duration");
   }
-  const llm = config.checks?.llm;
-  if (llm?.allowedModels !== void 0 || llm?.maxTotalTokens !== void 0) {
-    rules.push(createLlmUsageRule(llm));
+  const maxDepth = config.checks?.run?.maxDepth ?? evalConfig?.maxDepth;
+  if (maxDepth !== void 0) {
+    rules.push(createRunDepthRule({ maxDepth }));
+    select.add("run.depth");
+  }
+  if (evalConfig?.maxRetries !== void 0) {
+    rules.push(createToolFailureRule({ maxRetries: evalConfig.maxRetries }));
+    select.add("tool.failures");
+  }
+  const allowedModels = config.checks?.llm?.allowedModels;
+  const maxTotalTokens = config.checks?.llm?.maxTotalTokens ?? evalConfig?.maxTotalTokens;
+  if (allowedModels !== void 0 || maxTotalTokens !== void 0) {
+    rules.push(
+      createLlmUsageRule({
+        ...allowedModels !== void 0 ? { allowedModels } : {},
+        ...maxTotalTokens !== void 0 ? { maxTotalTokens } : {}
+      })
+    );
     select.add("llm.usage");
   }
   if (select.has("outcome.status")) {
     rules.push(createObservedOutcomeRule({ failOn: ["failed"] }));
   }
-  return { rules, select: [...select] };
+  const observationCount = suiteCase.expectedObservations?.length ?? 0;
+  if (rules.length === 0 && observationCount === 0 && configDiagnostics.length === 0) {
+    configDiagnostics.push(
+      diagnostic3(
+        "AI_SUITE_NO_ASSERTIONS",
+        `Suite case "${suiteCase.id}" declares no effective checks, eval controls, or expected observations.`,
+        "error",
+        suiteCase.id
+      )
+    );
+  }
+  return {
+    rules,
+    select: [...select],
+    observationCount,
+    configDiagnostics
+  };
 }
 function outcomesFromRead(read) {
   const persistedOutcomes = extractOutcomesFromPersistedEvents(
@@ -3746,8 +3804,20 @@ async function runSuiteCase(suiteCase, config, options) {
       ]
     };
   }
-  const { rules, select } = buildCaseRules(suiteCase, config);
-  const checkResult = rules.length > 0 ? runTraceChecks({ read }, { rules, select }) : {
+  const compiled = buildCaseAssertions(suiteCase, config);
+  if (compiled.configDiagnostics.length > 0) {
+    return {
+      id: suiteCase.id,
+      status: "error",
+      tracePath: resolved.tracePath,
+      ...resolved.runId !== void 0 ? { runId: resolved.runId } : {},
+      checkOk: false,
+      ...config.eval?.requireSuccess === true ? { evalOk: false } : {},
+      message: compiled.configDiagnostics.map((item) => item.message).join("; "),
+      diagnostics: compiled.configDiagnostics
+    };
+  }
+  const checkResult = compiled.rules.length > 0 ? runTraceChecks({ read }, { rules: compiled.rules, select: compiled.select }) : {
     ok: true,
     status: "pass",
     format: read.format,
@@ -3775,6 +3845,9 @@ async function runSuiteCase(suiteCase, config, options) {
   ];
   const checkOk = checkResult.ok;
   const observationsOk = observationResult.ok;
+  const evalOk = config.eval?.requireSuccess === true ? checkResult.findings.every(
+    (finding) => finding.ruleId !== "run.status" || finding.status !== "fail"
+  ) && checkResult.diagnostics.every((item) => item.severity !== "error") : void 0;
   const ok = checkOk && observationsOk;
   const status = ok ? "pass" : checkResult.status === "error" ? "error" : "fail";
   return {
@@ -3783,6 +3856,7 @@ async function runSuiteCase(suiteCase, config, options) {
     tracePath: resolved.tracePath,
     ...resolved.runId !== void 0 ? { runId: resolved.runId } : {},
     checkOk,
+    ...evalOk !== void 0 ? { evalOk } : {},
     observationsOk,
     diagnostics,
     ...ok ? {} : {