openmates 0.12.0-alpha.12 → 0.12.0-alpha.13

package/dist/{chunk-R5Z4FBJJ.js → chunk-YHOUQRWZ.js}

@@ -41830,6 +41830,7 @@ async function handleBenchmark(client, subcommand, rest, flags) {
   const runs = parseRuns(flags.runs);
   const extensiveSize = parseExtensiveSize(flags["extensive-size"]);
   const parallel = parseParallel(flags.parallel);
+  const caseIds = parseCaseIds(flags.case);
   const dryRun = flags["dry-run"] === true;
   const output = typeof flags.output === "string" ? flags.output : void 0;
   const runId = typeof flags["run-id"] === "string" ? flags["run-id"] : randomUUID3();
@@ -41839,7 +41840,7 @@ async function handleBenchmark(client, subcommand, rest, flags) {
       "Benchmark runs spend real credits from the logged-in account. Rerun with --confirm-spend-credits, or use --dry-run to preview the plan."
     );
   }
-  const cases = expandCases(suites, runs, extensiveSize);
+  const cases = filterCases(expandCases(suites, runs, extensiveSize), caseIds);
   const pricing = loadPricingForModels([...targetModels, judgeModel]);
   const estimate = estimateCredits(cases, targetModels, judgeModel, pricing);
   const result = makeBaseResult({
@@ -41893,6 +41894,7 @@ Options:
   --dry-run                     Preview the benchmark plan without inference or spend
   --compare                     Compare two or more target models
   --suite <list>                Comma-separated suites: quick, extensive, all (default: quick)
+  --case <id[,id...]>           Run only specific case id(s) from the selected suites
   --extensive-size <n>          Extensive cases to run: 5, 10, or 20 (default: ${DEFAULT_EXTENSIVE_SIZE})
   --runs <n>                    Repeat each selected case (default: 1)
   --parallel <n>                Concurrent target case requests (default: ${DEFAULT_PARALLEL})
@@ -41941,6 +41943,24 @@ function parseParallel(value) {
   }
   return parsed;
 }
+function parseCaseIds(value) {
+  if (value === void 0 || value === false) return [];
+  if (value === true) throw new Error("--case requires a case id");
+  const caseIds = value.split(",").map((caseId) => caseId.trim()).filter(Boolean);
+  if (caseIds.length === 0) throw new Error("--case requires at least one case id");
+  return [...new Set(caseIds)];
+}
+function filterCases(cases, caseIds) {
+  if (caseIds.length === 0) return cases;
+  const availableIds = new Set(cases.map((benchmarkCase) => benchmarkCase.id));
+  const missing = caseIds.filter((caseId) => !availableIds.has(caseId));
+  if (missing.length > 0) {
+    throw new Error(
+      `Unknown benchmark case id(s): ${missing.join(", ")}. Available in selected suite(s): ${[...availableIds].sort().join(", ")}`
+    );
+  }
+  return cases.filter((benchmarkCase) => caseIds.includes(benchmarkCase.id));
+}
 function expandCases(suites, runs, extensiveSize) {
   const selected = [];
   if (suites.includes("quick")) selected.push(...QUICK_CASES);
@@ -42145,7 +42165,14 @@ async function uploadBenchmarkImage(client, fileEmbed) {
   fileEmbed.embed.status = "finished";
   fileEmbed.embed.contentHash = uploadResult.content_hash;
   fileEmbed.embed.embedId = uploadResult.embed_id;
-  fileEmbed.referenceBlock = createEmbedReferenceBlock(embedRef);
+  fileEmbed.referenceBlock = createBenchmarkEmbedReferenceBlock(fileEmbed.embed.embedId, fileEmbed.embed.type);
+}
+function createBenchmarkEmbedReferenceBlock(embedId, embedType) {
+  return `
+
+\`\`\`json
+${JSON.stringify({ type: embedType, embed_id: embedId })}
+\`\`\``;
 }
 async function judgeCase(params) {
   const startedAt = Date.now();

package/dist/cli.js

@@ -2,7 +2,7 @@
 import {
   getExtForLang,
   serializeToYaml
-} from "./chunk-R5Z4FBJJ.js";
+} from "./chunk-YHOUQRWZ.js";
 import "./chunk-AXNRPVLE.js";
 export {
   getExtForLang,

package/dist/index.js

@@ -9,7 +9,7 @@ import {
   deriveAppUrl,
   getExtForLang,
   serializeToYaml
-} from "./chunk-R5Z4FBJJ.js";
+} from "./chunk-YHOUQRWZ.js";
 import "./chunk-AXNRPVLE.js";
 export {
   ASSISTANT_FEEDBACK_REPORT_TITLE,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openmates",
-  "version": "0.12.0-alpha.12",
+  "version": "0.12.0-alpha.13",
   "description": "OpenMates CLI and SDK",
   "type": "module",
   "main": "dist/index.js",