@wix/evalforge-evaluator 0.71.0 → 0.73.0

package/build/index.js

@@ -6477,165 +6477,9 @@ async function writeMcpToFilesystem(cwd, mcps) {
   console.log(`[MCP] Written to ${filePath}`);
 }
 
-// src/run-scenario/agents/claude-code/mcp-setup.ts
+// src/run-scenario/agents/claude-code/write-sub-agents.ts
 var import_promises5 = require("fs/promises");
-var import_child_process = require("child_process");
 var import_path7 = require("path");
-async function probeMcpServers(mcps, probeMs = 5e3) {
-  const results = [];
-  for (const mcp of mcps) {
-    const config = mcp.config;
-    for (const [name2, value] of Object.entries(config)) {
-      if (typeof value !== "object" || value === null) continue;
-      const cfg = value;
-      const command = cfg.command;
-      const args = cfg.args;
-      if (typeof command !== "string" || !Array.isArray(args)) continue;
-      const result = await probeOneServer(
-        name2,
-        command,
-        args,
-        probeMs
-      );
-      results.push(result);
-    }
-  }
-  return results;
-}
-function probeOneServer(name2, command, args, probeMs) {
-  return new Promise((resolve2) => {
-    const startMs = Date.now();
-    let stdout = "";
-    let stderr = "";
-    let settled = false;
-    const finish = (exitCode, signal) => {
-      if (settled) return;
-      settled = true;
-      resolve2({
-        name: name2,
-        command,
-        args,
-        exitCode,
-        signal,
-        stdout: stdout.slice(-2e3),
-        stderr: stderr.slice(-2e3),
-        durationMs: Date.now() - startMs
-      });
-    };
-    const child = (0, import_child_process.spawn)(command, args, {
-      stdio: ["pipe", "pipe", "pipe"],
-      env: process.env
-    });
-    child.stdout.on("data", (chunk) => {
-      stdout += chunk.toString();
-    });
-    child.stderr.on("data", (chunk) => {
-      stderr += chunk.toString();
-    });
-    child.on("error", (err) => {
-      stderr += `
-spawn error: ${err.message}`;
-      finish(null, null);
-    });
-    child.on("close", (code2, sig) => {
-      finish(code2, sig);
-    });
-    setTimeout(() => {
-      if (!settled) {
-        child.kill("SIGTERM");
-        finish(null, "PROBE_TIMEOUT");
-      }
-    }, probeMs);
-  });
-}
-async function probeNpmConfig(cwd) {
-  const homedir = process.env.HOME ?? "~";
-  const [wixAuthFile, npmInstallDryRun, mcpPackageInstalled] = await Promise.all([
-    (0, import_promises5.readFile)((0, import_path7.join)(homedir, ".wix", "auth", "api-key.json"), "utf8").catch(
-      (e) => `[not found: ${e.message}]`
-    ),
-    runShellCapture(
-      "npm install --dry-run @wix/mcp 2>&1 | tail -15",
-      cwd,
-      15e3
-    ),
-    runShellCapture(
-      'ls -la $(npm root -g)/@wix/mcp/package.json 2>/dev/null || echo "[not globally installed]" && npm list -g @wix/mcp 2>&1 | tail -5',
-      cwd,
-      5e3
-    )
-  ]);
-  const wixAuthParsed = (() => {
-    try {
-      const parsed = JSON.parse(wixAuthFile);
-      return JSON.stringify({
-        hasToken: typeof parsed.token === "string",
-        tokenLength: typeof parsed.token === "string" ? parsed.token.length : 0,
-        hasAccountId: typeof parsed.accountId === "string",
-        hasUserInfo: typeof parsed.userInfo === "object"
-      });
-    } catch {
-      return wixAuthFile.slice(0, 500);
-    }
-  })();
-  const npmEnvVars = {};
-  for (const [key, value] of Object.entries(process.env)) {
-    const lk = key.toLowerCase();
-    if (lk.startsWith("npm_config") || lk.includes("token") || lk.includes("auth") || lk.includes("registry") || lk.startsWith("wix_")) {
-      npmEnvVars[key] = lk.includes("token") || lk.includes("auth") || lk.includes("secret") ? `[REDACTED, length=${(value ?? "").length}]` : value;
-    }
-  }
-  return {
-    cwd,
-    processCwd: process.cwd(),
-    wixAuthFile: wixAuthParsed,
-    wixAuthFileExists: !wixAuthFile.startsWith("[not found"),
-    npmInstallDryRun: npmInstallDryRun.trim().slice(0, 1500),
-    mcpPackageInstalled: mcpPackageInstalled.trim().slice(0, 500),
-    npmEnvVars,
-    homedir
-  };
-}
-function runShellCapture(cmd, cwd, timeoutMs) {
-  return new Promise((resolve2) => {
-    let output = "";
-    let settled = false;
-    const child = (0, import_child_process.spawn)("sh", ["-c", cmd], {
-      stdio: ["pipe", "pipe", "pipe"],
-      cwd,
-      env: process.env
-    });
-    child.stdout.on("data", (chunk) => {
-      output += chunk.toString();
-    });
-    child.stderr.on("data", (chunk) => {
-      output += chunk.toString();
-    });
-    child.on("close", () => {
-      if (!settled) {
-        settled = true;
-        resolve2(output);
-      }
-    });
-    child.on("error", (err) => {
-      if (!settled) {
-        settled = true;
-        resolve2(`[error: ${err.message}]`);
-      }
-    });
-    setTimeout(() => {
-      if (!settled) {
-        settled = true;
-        child.kill("SIGTERM");
-        resolve2(output + "\n[timed out]");
-      }
-    }, timeoutMs);
-  });
-}
-
-// src/run-scenario/agents/claude-code/write-sub-agents.ts
-var import_promises6 = require("fs/promises");
-var import_path8 = require("path");
 var AGENTS_DIR = ".claude/agents";
 function toAgentFilename(name2, index, nameCount) {
   const base = (name2 || "").toLowerCase().replace(/\s+/g, "-").replace(/[^a-z0-9-]/g, "").replace(/^-+|-+$/g, "") || `sub-agent-${index}`;
@@ -6645,13 +6489,13 @@ function toAgentFilename(name2, index, nameCount) {
 }
 async function writeSubAgentsToFilesystem(cwd, subAgents) {
   if (subAgents.length === 0) return;
-  const agentsDir = (0, import_path8.join)(cwd, AGENTS_DIR);
-  await (0, import_promises6.mkdir)(agentsDir, { recursive: true });
+  const agentsDir = (0, import_path7.join)(cwd, AGENTS_DIR);
+  await (0, import_promises5.mkdir)(agentsDir, { recursive: true });
   const nameCount = /* @__PURE__ */ new Map();
   for (const [i, agent] of subAgents.entries()) {
     const filename = toAgentFilename(agent.name, i, nameCount);
-    const filePath = (0, import_path8.join)(agentsDir, `${filename}.md`);
-    await (0, import_promises6.writeFile)(filePath, agent.subAgentMd, "utf8");
+    const filePath = (0, import_path7.join)(agentsDir, `${filename}.md`);
+    await (0, import_promises5.writeFile)(filePath, agent.subAgentMd, "utf8");
   }
   console.log(`[SubAgents] Written to ${agentsDir}`);
 }
@@ -6886,56 +6730,6 @@ async function executeWithClaudeCode(skills, scenario, options) {
   const allMessages = [];
   if (options.mcps && options.mcps.length > 0) {
     await writeMcpToFilesystem(options.cwd, options.mcps);
-    const probeResults = await probeMcpServers(options.mcps);
-    if (options.traceContext) {
-      emitTraceEvent(
-        {
-          evalRunId: options.traceContext.evalRunId,
-          scenarioId: options.traceContext.scenarioId,
-          scenarioName: options.traceContext.scenarioName,
-          targetId: options.traceContext.targetId,
-          targetName: options.traceContext.targetName,
-          stepNumber: 0,
-          type: import_evalforge_types3.LiveTraceEventType.DIAGNOSTIC,
-          outputPreview: JSON.stringify({
-            event: "mcp-probe",
-            results: probeResults
-          }).slice(0, 2e3),
-          timestamp: (/* @__PURE__ */ new Date()).toISOString(),
-          isComplete: false
-        },
-        options.traceContext.tracePushUrl,
-        options.traceContext.routeHeader,
-        options.traceContext.authToken
-      );
-    }
-    const npmDiag = await probeNpmConfig(options.cwd);
-    console.log(
-      "[DEBUG-d744ca] npm-config-diagnostic",
-      JSON.stringify(npmDiag)
-    );
-    if (options.traceContext) {
-      emitTraceEvent(
-        {
-          evalRunId: options.traceContext.evalRunId,
-          scenarioId: options.traceContext.scenarioId,
-          scenarioName: options.traceContext.scenarioName,
-          targetId: options.traceContext.targetId,
-          targetName: options.traceContext.targetName,
-          stepNumber: 0,
-          type: import_evalforge_types3.LiveTraceEventType.DIAGNOSTIC,
-          outputPreview: JSON.stringify({
-            event: "npm-config-diagnostic",
-            ...npmDiag
-          }).slice(0, 8e3),
-          timestamp: (/* @__PURE__ */ new Date()).toISOString(),
-          isComplete: false
-        },
-        options.traceContext.tracePushUrl,
-        options.traceContext.routeHeader,
-        options.traceContext.authToken
-      );
-    }
   }
   if (options.subAgents && options.subAgents.length > 0) {
     await writeSubAgentsToFilesystem(options.cwd, options.subAgents);
@@ -7196,41 +6990,6 @@ IMPORTANT: This is an automated evaluation run. Follow these guidelines:
             })
           );
         }
-        const sdkMsg = message;
-        if (sdkMsg.type === "system" && sdkMsg.subtype === "init") {
-          const initData = sdkMsg;
-          const mcpInfo = {
-            mcp_servers: initData.mcp_servers,
-            tools: initData.tools,
-            cwd: options.cwd
-          };
-          console.error(
-            "[MCP-DIAG] Init message MCP status:",
-            JSON.stringify(mcpInfo, null, 2)
-          );
-          if (traceContext) {
-            emitTraceEvent(
-              {
-                evalRunId: traceContext.evalRunId,
-                scenarioId: traceContext.scenarioId,
-                scenarioName: traceContext.scenarioName,
-                targetId: traceContext.targetId,
-                targetName: traceContext.targetName,
-                stepNumber: traceStepNumber,
-                type: import_evalforge_types3.LiveTraceEventType.DIAGNOSTIC,
-                outputPreview: JSON.stringify({
-                  event: "mcp-init-status",
-                  ...mcpInfo
-                }).slice(0, 2e3),
-                timestamp: (/* @__PURE__ */ new Date()).toISOString(),
-                isComplete: false
-              },
-              traceContext.tracePushUrl,
-              traceContext.routeHeader,
-              traceContext.authToken
-            );
-          }
-        }
         if (traceContext) {
           traceStepNumber++;
           const traceEvent = createTraceEventFromAnyMessage(
@@ -7756,7 +7515,7 @@ defaultRegistry.register(claudeCodeAdapter);
 
 // src/run-scenario/file-diff.ts
 var import_fs6 = require("fs");
-var import_path9 = require("path");
+var import_path8 = require("path");
 
 // ../../node_modules/diff/lib/index.mjs
 function Diff() {
@@ -7932,7 +7691,7 @@ Diff.prototype = {
   tokenize: function tokenize(value) {
     return Array.from(value);
   },
-  join: function join6(chars) {
+  join: function join5(chars) {
     return chars.join("");
   },
   postProcess: function postProcess(changeObjects) {
@@ -8372,8 +8131,8 @@ function snapshotDirectory(dir, baseDir) {
   }
   const entries = (0, import_fs6.readdirSync)(dir, { withFileTypes: true });
   for (const entry of entries) {
-    const fullPath = (0, import_path9.join)(dir, entry.name);
-    const relativePath = (0, import_path9.relative)(base, fullPath);
+    const fullPath = (0, import_path8.join)(dir, entry.name);
+    const relativePath = (0, import_path8.relative)(base, fullPath);
     if (shouldIgnore(entry.name)) {
       continue;
     }