bitfab-cli 0.2.246 → 0.2.247

package/dist/index.js

@@ -28304,6 +28304,24 @@ var removeGradersFromDataset = {
     graderIds: external_exports.preprocess(parseJsonString, external_exports.array(external_exports.uuid()).min(1).max(100)).describe("Grader ids to remove (1-100)")
   }
 };
+var rerunGradersOnDataset = {
+  name: "rerun_graders_on_dataset",
+  title: "Re-run Graders on Dataset",
+  description: "Re-score every trace in a dataset with its attached graders, overwriting their previous verdicts. This is how you grade traces that were added before a grader existed, or re-score everything after editing a grader's criteria: attaching a grader does not itself grade anything. Omit graderIds to re-run every grader attached to the dataset, or pass a subset; ids that are not attached are rejected rather than silently skipped. Waits up to 90 seconds for the run to finish and reports how many traces were graded; if it is still running, call again with the same arguments to check progress (it reports the running job rather than starting a second one). Read the resulting verdicts with get_grader_labels.",
+  inputSchema: {
+    datasetId: external_exports.uuid().describe("The dataset id (from save_dataset or list_datasets)"),
+    graderIds: external_exports.preprocess(parseJsonString, external_exports.array(external_exports.uuid()).min(1).max(100)).optional().describe("Grader ids to re-run (1-100). Omit to re-run every grader attached to the dataset.")
+  }
+};
+var rerunGradersOnExperiment = {
+  name: "rerun_graders_on_experiment",
+  title: "Re-run Graders on Experiment",
+  description: "Re-score an experiment's completed replay traces with its attached graders, overwriting their previous verdicts. Use this after attaching a grader to an already-completed experiment (which persists the assignment but does not grade the existing traces) or after editing a grader's criteria. Only the replays belonging to this experiment are graded, never the originals they were compared against. Omit graderIds to re-run every grader attached to the experiment, or pass a subset; ids that are not attached are rejected. Waits up to 90 seconds and reports how many traces were graded; if it is still running, call again with the same arguments to check progress. Read the resulting verdicts with get_grader_labels.",
+  inputSchema: {
+    testRunId: external_exports.uuid().describe("The experiment (test run) id, from list_experiments"),
+    graderIds: external_exports.preprocess(parseJsonString, external_exports.array(external_exports.uuid()).min(1).max(100)).optional().describe("Grader ids to re-run (1-100). Omit to re-run every grader attached to the experiment.")
+  }
+};
 var addGradersToExperiment = {
   name: "add_graders_to_experiment",
   title: "Add Graders to Experiment",
@@ -28489,6 +28507,8 @@ var ALL_TOOL_CONTRACTS = [
   removeTracesFromDataset,
   addGradersToDataset,
   removeGradersFromDataset,
+  rerunGradersOnDataset,
+  rerunGradersOnExperiment,
   addGradersToExperiment,
   removeGradersFromExperiment,
   addGradersToExperimentGroup,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bitfab-cli",
-  "version": "0.2.246",
+  "version": "0.2.247",
   "description": "Install and configure the Bitfab plugin in Claude Code, Codex, or Cursor.",
   "type": "module",
   "main": "dist/index.js",