vskill 1.0.0 → 1.0.1

package/agents.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "generatedAt": "2026-04-27T07:19:05.521Z",
+  "generatedAt": "2026-04-27T13:45:59.139Z",
   "agentPrefixes": [
     ".adal",
     ".agent",

package/dist/eval-server/api-routes.js

@@ -951,6 +951,17 @@ function getEffectiveRawModel() {
 function getClient() {
     return createLlmClient(currentOverrides);
 }
+// Per-request client: prefer body-supplied provider/model, then session globals.
+// Lets the frontend send the user's selected model with each request so historical
+// runs reliably reflect the picker value rather than a stale session default.
+function clientFromBody(body) {
+    const reqProvider = typeof body?.provider === "string" ? body.provider : undefined;
+    const reqModel = typeof body?.model === "string" ? body.model : undefined;
+    const provider = (reqProvider || currentOverrides.provider || "claude-cli");
+    const model = reqModel || currentOverrides.model;
+    const client = createLlmClient({ provider, model });
+    return { client, provider };
+}
 /** Derive sidebar badge status from benchmark + current eval IDs. */
 function computeBenchmarkStatus(benchmark, evalIds, hasEvals) {
     if (!benchmark)
@@ -2549,7 +2560,7 @@ export function registerRoutes(router, root, projectName) {
             const evals = loadAndValidateEvals(skillDir);
             const skillMdPath = join(skillDir, "SKILL.md");
             const skillContent = existsSync(skillMdPath) ? readFileSync(skillMdPath, "utf-8") : "";
-            const client = getClient();
+            const { client, provider: effectiveProvider } = clientFromBody(body);
             const systemPrompt = buildEvalSystemPrompt(skillContent);
             // Create separate judge client if judgeModel is specified
             let judgeClient;
@@ -2570,7 +2581,7 @@ export function registerRoutes(router, root, projectName) {
             }
             await runBenchmarkSSE({
                 res, skillDir, skillName: evals.skill_name, systemPrompt,
-                runType: "benchmark", provider: currentOverrides.provider || "claude-cli",
+                runType: "benchmark", provider: effectiveProvider,
                 evalCases: evals.evals, filterIds, client, judgeClient, judgeCache,
                 isAborted: () => aborted, concurrency,
             });
@@ -2592,11 +2603,11 @@ export function registerRoutes(router, root, projectName) {
         initSSE(res, req);
         try {
             const evals = loadAndValidateEvals(skillDir);
-            const client = getClient();
+            const { client, provider: effectiveProvider } = clientFromBody(body);
             await runBenchmarkSSE({
                 res, skillDir, skillName: evals.skill_name,
                 systemPrompt: "You are a helpful AI assistant.",
-                runType: "baseline", provider: currentOverrides.provider || "claude-cli",
+                runType: "baseline", provider: effectiveProvider,
                 evalCases: evals.evals, filterIds, client, isAborted: () => aborted,
             });
         }
@@ -2636,14 +2647,14 @@ export function registerRoutes(router, root, projectName) {
             }
             const skillMdPath = join(skillDir, "SKILL.md");
             const skillContent = existsSync(skillMdPath) ? readFileSync(skillMdPath, "utf-8") : "";
-            const client = getClient();
+            const { client, provider: effectiveProvider } = clientFromBody(body);
             const systemPrompt = isBaseline
                 ? buildBaselineSystemPrompt()
                 : buildEvalSystemPrompt(skillContent);
             await sem.acquire();
             const benchCase = await runSingleCaseSSE({
                 res, evalCase, systemPrompt, client, isAborted: () => aborted,
-                provider: currentOverrides.provider || "claude-cli",
+                provider: effectiveProvider,
             });
             if (!released) {
                 released = true;
@@ -2659,7 +2670,7 @@ export function registerRoutes(router, root, projectName) {
                         cases: [benchCase],
                         overall_pass_rate: benchCase.pass_rate,
                         type: isBaseline ? "baseline" : "benchmark",
-                        provider: currentOverrides.provider || "claude-cli",
+                        provider: effectiveProvider,
                         totalDurationMs: benchCase.durationMs ?? 0,
                         totalInputTokens: benchCase.inputTokens ?? null,
                         totalOutputTokens: benchCase.outputTokens ?? null,