@wix/evalforge-evaluator 0.114.0 → 0.116.0

package/build/index.mjs

@@ -1191,6 +1191,31 @@ function createTraceEventFromAnyMessage(message, context, stepNumber, isComplete
     outputPreview: `Message type: ${message.type}`
   };
 }
+async function prepareClaudeCodeEnvironment(cwd, skills, options) {
+  const { mkdir: mkdirAsync, writeFile: writeFile6 } = await import("fs/promises");
+  const claudeDir = `${cwd}/.claude`;
+  await mkdirAsync(claudeDir, { recursive: true });
+  await writeFile6(`${claudeDir}/settings.json`, "{}", {
+    flag: "wx"
+  }).catch(() => {
+  });
+  if (options.mcps && options.mcps.length > 0) {
+    await writeMcpToFilesystem(cwd, options.mcps);
+  }
+  if (options.subAgents && options.subAgents.length > 0) {
+    await writeSubAgentsToFilesystem(cwd, options.subAgents);
+  }
+  if (options.rules && options.rules.length > 0) {
+    await writeRulesToFilesystem(cwd, options.rules);
+  }
+  try {
+    await writeSkillsToFilesystem(cwd, skills);
+  } catch (writeError) {
+    throw new Error(
+      `Failed to write skills to filesystem: ${writeError instanceof Error ? writeError.message : String(writeError)}`
+    );
+  }
+}
 async function executeWithClaudeCode(skills, scenario, options) {
   const skillNames = skills.map((s) => s.name).join(", ");
   console.log("[executeWithClaudeCode] Starting execution", {
@@ -1214,29 +1239,6 @@ async function executeWithClaudeCode(skills, scenario, options) {
   }
   const startTime = /* @__PURE__ */ new Date();
   const allMessages = [];
-  const { mkdir: mkdirAsync, writeFile: writeFile6 } = await import("fs/promises");
-  const claudeDir = `${options.cwd}/.claude`;
-  await mkdirAsync(claudeDir, { recursive: true });
-  await writeFile6(`${claudeDir}/settings.json`, "{}", {
-    flag: "wx"
-  }).catch(() => {
-  });
-  if (options.mcps && options.mcps.length > 0) {
-    await writeMcpToFilesystem(options.cwd, options.mcps);
-  }
-  if (options.subAgents && options.subAgents.length > 0) {
-    await writeSubAgentsToFilesystem(options.cwd, options.subAgents);
-  }
-  if (options.rules && options.rules.length > 0) {
-    await writeRulesToFilesystem(options.cwd, options.rules);
-  }
-  try {
-    await writeSkillsToFilesystem(options.cwd, skills);
-  } catch (writeError) {
-    throw new Error(
-      `Failed to write skills to filesystem: ${writeError instanceof Error ? writeError.message : String(writeError)}`
-    );
-  }
   const sdkEnv = buildSdkEnvironment(options);
   let traceStepNumber = 0;
   const traceContext = options.traceContext;
@@ -2078,6 +2080,17 @@ var ClaudeCodeAdapter = class {
   id = "claude-code";
   name = "Claude Code";
   supportedCommands = [AgentRunCommand.CLAUDE];
+  /**
+   * Write infrastructure files (settings, MCPs, sub-agents, rules, skills)
+   * before the baseline snapshot is taken.
+   */
+  async prepareEnvironment(context) {
+    await prepareClaudeCodeEnvironment(context.cwd, context.skills, {
+      mcps: context.mcps,
+      subAgents: context.subAgents,
+      rules: context.rules
+    });
+  }
   /**
    * Execute a skill using the Claude Code SDK.
    *
@@ -2141,6 +2154,7 @@ defaultRegistry.register(claudeCodeAdapter);
 import { AgentRunCommand as AgentRunCommand2 } from "@wix/evalforge-types";
 
 // src/run-scenario/agents/opencode/execute.ts
+import { homedir as homedir2 } from "os";
 import {
   ClaudeModel as ClaudeModel3,
   DEFAULT_EVALUATOR_SYSTEM_PROMPT as DEFAULT_EVALUATOR_SYSTEM_PROMPT2,
@@ -2652,6 +2666,13 @@ function buildConversation2(messages) {
 
 // src/run-scenario/agents/opencode/execute.ts
 var DEFAULT_MODEL3 = `anthropic/${ClaudeModel3.CLAUDE_4_5_SONNET_1_0}`;
+function ensureOpenCodeInPath() {
+  const opencodeBin = `${homedir2()}/.opencode/bin`;
+  const currentPath = process.env.PATH || "";
+  if (!currentPath.includes(opencodeBin)) {
+    process.env.PATH = `${opencodeBin}:${currentPath}`;
+  }
+}
 function extractToolAction(toolName, args) {
   if (!toolName) return "Using tool...";
   const a = args;
@@ -2729,37 +2750,39 @@ function createTraceEventFromPart(part, context, stepNumber, isComplete) {
       return null;
   }
 }
-async function executeWithOpenCode(skills, scenario, options) {
-  const skillNames = skills.map((s) => s.name).join(", ");
-  console.log("[executeWithOpenCode] Starting execution", {
-    skillCount: skills.length,
-    skillNames,
-    scenarioId: scenario.id,
-    scenarioName: scenario.name,
-    cwd: options.cwd,
-    aiGatewayUrl: options.aiGatewayUrl,
-    hasAiGatewayHeaders: !!options.aiGatewayHeaders,
-    model: options.model
-  });
-  const startTime = /* @__PURE__ */ new Date();
+async function prepareOpenCodeEnvironment(cwd, skills, options) {
   if (options.mcps && options.mcps.length > 0) {
     console.log(
       `[MCP] ${options.mcps.length} MCP(s) will be configured inline`
     );
   }
   if (options.subAgents && options.subAgents.length > 0) {
-    await writeSubAgentsToFilesystem2(options.cwd, options.subAgents);
+    await writeSubAgentsToFilesystem2(cwd, options.subAgents);
   }
   if (options.rules && options.rules.length > 0) {
-    await writeRulesToFilesystem(options.cwd, options.rules);
+    await writeRulesToFilesystem(cwd, options.rules);
   }
   try {
-    await writeSkillsToFilesystem2(options.cwd, skills);
+    await writeSkillsToFilesystem2(cwd, skills);
   } catch (writeError) {
     throw new Error(
       `Failed to write skills to filesystem: ${writeError instanceof Error ? writeError.message : String(writeError)}`
     );
   }
+}
+async function executeWithOpenCode(skills, scenario, options) {
+  const skillNames = skills.map((s) => s.name).join(", ");
+  console.log("[executeWithOpenCode] Starting execution", {
+    skillCount: skills.length,
+    skillNames,
+    scenarioId: scenario.id,
+    scenarioName: scenario.name,
+    cwd: options.cwd,
+    aiGatewayUrl: options.aiGatewayUrl,
+    hasAiGatewayHeaders: !!options.aiGatewayHeaders,
+    model: options.model
+  });
+  const startTime = /* @__PURE__ */ new Date();
   const maxTurns = options.maxTurns ?? 10;
   const { config, providerID, modelID } = await buildOpenCodeConfig({
     model: options.model,
@@ -2807,6 +2830,7 @@ async function executeWithOpenCode(skills, scenario, options) {
   }
   let server;
   try {
+    ensureOpenCodeInPath();
     console.log("[SDK-DEBUG] Starting OpenCode server...");
     server = await createOpencodeServer({
       config,
@@ -3138,6 +3162,13 @@ var OpenCodeAdapter = class {
   id = "opencode";
   name = "OpenCode";
   supportedCommands = [AgentRunCommand2.OPENCODE];
+  async prepareEnvironment(context) {
+    await prepareOpenCodeEnvironment(context.cwd, context.skills, {
+      mcps: context.mcps,
+      subAgents: context.subAgents,
+      rules: context.rules
+    });
+  }
   async execute(context) {
     const {
       skills,
@@ -4273,6 +4304,15 @@ arrayDiff.join = arrayDiff.removeEmpty = function(value) {
 };
 
 // src/run-scenario/file-diff.ts
+function deriveInfrastructurePaths(prePrep, postPrep) {
+  const infraPaths = /* @__PURE__ */ new Set();
+  for (const path2 of Object.keys(postPrep)) {
+    if (prePrep[path2] === void 0 || prePrep[path2] !== postPrep[path2]) {
+      infraPaths.add(path2);
+    }
+  }
+  return infraPaths;
+}
 var IGNORED_PATTERNS = [
   "node_modules",
   ".git",
@@ -4376,7 +4416,7 @@ function generateDiffLines(before, after) {
   }
   return result;
 }
-function diffSnapshots(before, after) {
+function diffSnapshots(before, after, infrastructurePaths) {
   const diffs = [];
   const allPaths = /* @__PURE__ */ new Set([...Object.keys(before), ...Object.keys(after)]);
   for (const path2 of allPaths) {
@@ -4390,7 +4430,8 @@ function diffSnapshots(before, after) {
       path: path2,
       expected: beforeContent,
       actual: afterContent,
-      diffLines: diffLines2
+      diffLines: diffLines2,
+      ...infrastructurePaths?.has(path2) && { isInfrastructure: true }
     });
   }
   const deletedPaths = [...allPaths].filter((p) => after[p] === void 0);
@@ -4414,7 +4455,7 @@ function diffSnapshots(before, after) {
   result.sort((a, b) => a.path.localeCompare(b.path));
   return result;
 }
-function extractTemplateFiles(before, after) {
+function extractTemplateFiles(before, after, infrastructurePaths) {
   const files = [];
   const allPaths = /* @__PURE__ */ new Set([...Object.keys(before), ...Object.keys(after)]);
   for (const path2 of allPaths) {
@@ -4434,7 +4475,8 @@ function extractTemplateFiles(before, after) {
     files.push({
       path: path2,
       content: afterContent,
-      status
+      status,
+      ...infrastructurePaths?.has(path2) && { isInfrastructure: true }
     });
   }
   files.sort((a, b) => a.path.localeCompare(b.path));
@@ -4450,7 +4492,6 @@ async function runAgentWithContext(config, evalRunId2, scenario, evalData, workD
   const identifier = isSDK ? simpleAgentAdapter.id : agent?.runCommand ?? DEFAULT_AGENT_COMMAND;
   const adapter = getAdapter(identifier);
   const startedAt = (/* @__PURE__ */ new Date()).toISOString();
-  const beforeSnapshot = workDir ? snapshotDirectory(workDir) : {};
   const targetId = evalData.evalRun.presetId ?? agent?.id ?? evalData.evalRun.id;
   const targetName = evalData.presetName || agent?.name || "";
   const executionContext = {
@@ -4475,11 +4516,22 @@ async function runAgentWithContext(config, evalRunId2, scenario, evalData, workD
     rules: evalData.rules?.length > 0 ? evalData.rules : void 0,
     systemPrompt: agent?.systemPrompt
   };
+  const hasPrepare = !!adapter.prepareEnvironment;
+  const prePrepSnapshot = hasPrepare && workDir ? snapshotDirectory(workDir) : {};
+  if (hasPrepare) {
+    await adapter.prepareEnvironment(executionContext);
+  }
+  const beforeSnapshot = workDir ? snapshotDirectory(workDir) : {};
+  const infrastructurePaths = hasPrepare ? deriveInfrastructurePaths(prePrepSnapshot, beforeSnapshot) : /* @__PURE__ */ new Set();
   const { outputText, durationMs, llmTrace, conversation } = await adapter.execute(executionContext);
   const completedAt = (/* @__PURE__ */ new Date()).toISOString();
   const afterSnapshot = workDir ? snapshotDirectory(workDir) : {};
-  const fileDiffs = diffSnapshots(beforeSnapshot, afterSnapshot);
-  const templateFiles = workDir ? extractTemplateFiles(beforeSnapshot, afterSnapshot) : void 0;
+  const fileDiffs = diffSnapshots(
+    beforeSnapshot,
+    afterSnapshot,
+    infrastructurePaths
+  );
+  const templateFiles = workDir ? extractTemplateFiles(beforeSnapshot, afterSnapshot, infrastructurePaths) : void 0;
   return {
     id: randomUUID4(),
     targetId,