bitfab-cli 0.2.459 → 0.2.461

package/dist/index.js

@@ -29339,18 +29339,19 @@ var listGraders = {
 var saveDataset = {
   name: "save_dataset",
   title: "Save Dataset",
-  description: "Create or edit a labeled dataset for a traced function. Datasets are named buckets of traces (e.g. 'Hallucination failures', 'Edge cases for grader X') that humans review and that experiments replay against. To create, pass traceFunctionKey, a short human-readable name, and an optional description. Saving with the name of an existing dataset on the same trace function updates that dataset instead of creating a duplicate, so re-running a flow is safe; pass a new description to change it, or omit it to leave the current one. To rename a dataset or change its description in place, pass its datasetId with a new name and/or description; its id, traces, and graders are kept, and a name already used by another dataset on the same trace function is refused. Returns the dataset id; use it with add_traces_to_dataset to populate. Call list_datasets first when you need to see what already exists.",
+  description: "Create or edit a labeled dataset for a traced function. Datasets are named buckets of traces (e.g. 'Hallucination failures', 'Edge cases for grader X') that humans review and that experiments replay against. To create, pass traceFunctionKey, a short human-readable name, and an optional description. Saving with the name of an existing dataset on the same trace function updates that dataset instead of creating a duplicate, so re-running a flow is safe; pass a new description to change it, or omit it to leave the current one. To rename a dataset or change its description in place, pass its datasetId with a new name and/or description; its id, traces, and graders are kept, and a name already used by another dataset on the same trace function is refused. Pass golden: true to mark the dataset as golden, a trusted reference set to replay and grade changes against (golden datasets are listed first, for the whole organization, in the Bitfab app and in list_datasets), or golden: false to unmark it; omit golden to leave it as it is. This works on create, on a save by name, and with datasetId. Returns the dataset id; use it with add_traces_to_dataset to populate. Call list_datasets first when you need to see what already exists.",
   inputSchema: {
     datasetId: external_exports.uuid().optional().describe("Existing dataset id (from list_datasets) to rename or redescribe in place. Omit to create a dataset or save one by name."),
     traceFunctionKey: external_exports.string().min(1).optional().describe("The trace function key the dataset is scoped to (must match traces you'll add). Required without datasetId."),
     name: external_exports.string().min(1).optional().describe("Short human-readable name shown in the labeling UI (e.g. 'Hallucination failures'). Required without datasetId. With datasetId, the new name; omit to keep the current one."),
-    description: external_exports.string().optional().describe("Optional longer description of what's in the dataset. With datasetId, omit to keep the current one or pass an empty string to clear it.")
+    description: external_exports.string().optional().describe("Optional longer description of what's in the dataset. With datasetId, omit to keep the current one or pass an empty string to clear it."),
+    golden: external_exports.boolean().optional().describe("true marks the dataset as golden, false unmarks it. Omit to leave it as it is.")
   }
 };
 var listDatasets = {
   name: "list_datasets",
   title: "List Datasets",
-  description: "List all datasets for a traced function, with id, name, description, current trace count, and assigned graders for each. Call this before changing trace or grader assignments to discover dataset ids and current membership, or before save_dataset to avoid duplicates.",
+  description: "List all datasets for a traced function, golden datasets first, with id, name, whether it is golden, description, current trace count, and assigned graders for each. Golden datasets are the ones the team marked as trusted reference sets, so prefer them when choosing a dataset to replay or grade. Call this before changing trace or grader assignments to discover dataset ids and current membership, or before save_dataset to avoid duplicates.",
   inputSchema: {
     traceFunctionKey: external_exports.string().min(1).describe("The trace function key to list datasets for")
   }
@@ -30378,10 +30379,10 @@ var semver3 = __toESM(require_semver2(), 1);
 
 // ../bitfab-plugin-lib/dist/bakedSdkVersions.js
 var BAKED_SDK_VERSIONS = {
-  typescript: "0.65.2",
-  python: "0.65.2",
-  ruby: "0.65.2",
-  go: "0.65.2"
+  typescript: "0.65.4",
+  python: "0.65.3",
+  ruby: "0.65.3",
+  go: "0.65.3"
 };
 
 // ../bitfab-plugin-lib/dist/installedSdk.js

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bitfab-cli",
-  "version": "0.2.459",
+  "version": "0.2.461",
   "description": "Install and configure the Bitfab plugin in Claude Code, Codex, or Cursor, or the Bitfab skill pack in Amp (Alpha).",
   "type": "module",
   "main": "dist/index.js",