@workbench-ai/workbench-built-in-adapters 0.0.131 → 0.0.133

package/dist/execute.js

@@ -708,6 +708,7 @@ async function writeAgentSkillOutput(request, workload, adapter, options = {}) {
     if (request.operation !== "skill.run") {
         throw new Error("Agent skill execution results can only complete skill.run operations.");
     }
+    await sealProviderSkillRunWorkspace(request);
     const traceRoot = path.join(request.paths.output, ".workbench", "internal", "agent-skill");
     const agentResult = await executeBuiltInAgentTurn(options.agentExecutor, {
         role: "runner",
@@ -740,7 +741,11 @@ async function writeAgentSkillOutput(request, workload, adapter, options = {}) {
             metadata: agentResult.metadata,
         }, null, 2)}\n`,
     };
-    await writeSurfaceFiles(request.paths.output, [trace, ...agentResult.traceFiles]);
+    await writeSurfaceFiles(request.paths.output, [
+        ...agentSessionEvidenceFiles(workload.job.id, adapter.agent.use, agentResult.metadata),
+        trace,
+        ...agentResult.traceFiles,
+    ]);
     const runtime = await importWorkbenchRuntime();
     const usage = runtime.assignUsageRole("runner", agentResult.usage);
     await writeWorkbenchAdapterOperationResult(request.paths.output, {
@@ -756,6 +761,47 @@ async function writeAgentSkillOutput(request, workload, adapter, options = {}) {
         ...(usage ? { usage } : {}),
     });
 }
+async function sealProviderSkillRunWorkspace(request) {
+    await Promise.all([
+        fs.rm(path.join(request.paths.workspace, ".workbench"), { recursive: true, force: true }),
+        request.paths.enginePrivate
+            ? fs.rm(request.paths.enginePrivate, { recursive: true, force: true })
+            : Promise.resolve(),
+        request.paths.traces
+            ? fs.rm(request.paths.traces, { recursive: true, force: true })
+            : Promise.resolve(),
+    ]);
+}
+function agentSessionEvidenceFiles(jobId, provider, metadata) {
+    const record = metadata && typeof metadata === "object" && !Array.isArray(metadata)
+        ? metadata
+        : {};
+    const sessionId = typeof record.sessionId === "string" && record.sessionId.trim()
+        ? record.sessionId.trim()
+        : undefined;
+    const providerId = typeof record.providerId === "string" && record.providerId.trim()
+        ? record.providerId.trim()
+        : provider;
+    const model = typeof record.model === "string" && record.model.trim()
+        ? record.model.trim()
+        : undefined;
+    if (!sessionId && !providerId && !model) {
+        return [];
+    }
+    return [
+        jsonSurfaceFile("agent-session.json", {
+            schema: "workbench.agent.session.v1",
+            jobId,
+            provider,
+            providerId,
+            ...(model ? { model } : {}),
+            ...(sessionId ? {
+                sessionId,
+                ref: `${provider}:${sessionId}`,
+            } : {}),
+        }),
+    ];
+}
 function buildAgentSkillPrompt(workload, adapter) {
     return [
         ...(adapter.instructions ? ["Instructions:", adapter.instructions, ""] : []),
@@ -937,6 +983,7 @@ async function writeRubricEvidenceFiles(args) {
         ...(args.usage ? { usage: args.usage } : {}),
     };
     await writeSurfaceFiles(args.request.paths.output, [
+        jsonSurfaceFile("rubric-scorecard.json", scorecard),
         jsonSurfaceFile(`${root}/scorecard.json`, scorecard),
         ...args.criterionRuns.map((run) => jsonSurfaceFile(`${root}/criteria/${safeInternalPathSegment(run.result.criterion_id)}/result.json`, {
             schema: "workbench.engine.rubric.criterion-evidence.v1",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@workbench-ai/workbench-built-in-adapters",
-  "version": "0.0.131",
+  "version": "0.0.133",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/workbench-ai/workbench.git",
@@ -33,11 +33,11 @@
   "dependencies": {
     "yaml": "^2.8.2",
     "@workbench-ai/agent-driver-anthropic-claude-code": "0.0.49",
-    "@workbench-ai/agent-driver-openai-codex": "0.0.49",
-    "@workbench-ai/workbench-core": "0.0.131",
-    "@workbench-ai/workbench-protocol": "0.0.131",
     "@workbench-ai/agent-driver": "0.0.49",
-    "@workbench-ai/workbench-contract": "0.0.131"
+    "@workbench-ai/workbench-contract": "0.0.133",
+    "@workbench-ai/workbench-protocol": "0.0.133",
+    "@workbench-ai/agent-driver-openai-codex": "0.0.49",
+    "@workbench-ai/workbench-core": "0.0.133"
   },
   "devDependencies": {
     "@types/node": "^24.3.1",