nex-code 0.4.41 → 0.5.1

package/dist/skills/autoresearch.js

@@ -14,9 +14,10 @@
  * - No iteration cap by default — runs until stopped
  */
 
-const { execSync, spawn } = require("child_process");
+const { execSync, execFileSync, spawn } = require("child_process");
 const fs = require("fs");
 const path = require("path");
+const { T: _T } = require("../theme");
 
 // Lazy-load agent to reset read guards between experiments
 function resetReadGuards() {
@@ -473,7 +474,7 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
           // Check if branch already exists
           let branchExists = false;
           try {
-            execSync(`git rev-parse --verify ${branchName}`, {
+            execFileSync("git", ["rev-parse", "--verify", branchName], {
               cwd: process.cwd(),
               stdio: ["pipe", "pipe", "pipe"],
             });
@@ -492,7 +493,7 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
             } catch {
               // Ignore stash errors (nothing to stash)
             }
-            execSync(`git checkout ${branchName}`, {
+            execFileSync("git", ["checkout", branchName], {
               cwd: process.cwd(),
               stdio: ["pipe", "pipe", "pipe"],
             });
@@ -504,7 +505,7 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
           }
 
           const sourceBranch = gitBranch() || "unknown";
-          execSync(`git checkout -b ${branchName}`, {
+          execFileSync("git", ["checkout", "-b", branchName], {
             cwd: process.cwd(),
             stdio: ["pipe", "pipe", "pipe"],
           });
@@ -554,8 +555,7 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
           }).trim();
 
           if (hasChanges) {
-            execSync(
-              `git commit -m "autoresearch: checkpoint before: ${(args.message || "experiment").replace(/"/g, '\\"')}"`,
+            execFileSync("git", ["commit", "-m", `autoresearch: checkpoint before: ${args.message || "experiment"}`],
               { cwd: process.cwd(), stdio: "pipe" },
             );
           }
@@ -960,7 +960,7 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
           args.metric < sessionBaselineScore
         ) {
           console.log(
-            `\x1b[31m   ⚠ Score ${args.metric} < baseline ${sessionBaselineScore} — overriding kept=true to kept=false\x1b[0m`,
+            `${_T.red}   ⚠ Score ${args.metric} < baseline ${sessionBaselineScore} — overriding kept=true to kept=false${_T.reset}`,
           );
           args.kept = false;
           args.status = "discard";
@@ -995,8 +995,8 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
         const keptCount = experiments.filter((e) => e.kept).length;
         const revertedCount = experiments.filter((e) => !e.kept).length;
         const statusIcon = args.kept
-          ? "\x1b[32m✔ KEPT\x1b[0m"
-          : "\x1b[31m✘ REVERTED\x1b[0m";
+          ? `${_T.green}✔ KEPT${_T.reset}`
+          : `${_T.red}✘ REVERTED${_T.reset}`;
         const delta =
           prev != null && typeof args.metric === "number"
             ? ` (${args.metric > prev ? "+" : ""}${(args.metric - prev).toFixed(1)} pts)`
@@ -1086,7 +1086,7 @@ Use ar_run_experiment with output_file to redirect, then ar_extract_metric to re
 
           const expNum = experiments.length + 1;
           console.log(
-            `\x1b[33m   ↩ Reverted\x1b[0m  ${currentHash.slice(0, 7)} → ${newHash.slice(0, 7)}${args.reason ? `  (${args.reason})` : ""}`,
+            `${_T.yellow}   ↩ Reverted${_T.reset}  ${currentHash.slice(0, 7)} → ${newHash.slice(0, 7)}${args.reason ? `  (${args.reason})` : ""}`,
           );
 
           return JSON.stringify({
@@ -1288,7 +1288,7 @@ function _runWatchTest(testCommand, changedFile) {
     } catch {
       // Agent not in active conversation — just log
       process.stderr.write(
-        `\n\x1b[33m⚠ Watch: tests failed after ${changedFile} changed\x1b[0m\n`,
+        `\n${_T.yellow}⚠ Watch: tests failed after ${changedFile} changed${_T.reset}\n`,
       );
     }
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nex-code",
-  "version": "0.4.41",
+  "version": "0.5.1",
   "description": "Run 400B+ open coding models on your codebase without the hardware bill. Ollama Cloud first — OpenAI, Anthropic, and Gemini when you need them.",
   "bin": {
     "nex-code": "./dist/nex-code.js"
@@ -18,7 +18,7 @@
     "start": "node dist/nex-code.js",
     "build": "node scripts/build.js",
     "dev": "node scripts/build.js --dev",
-    "test": "jest --forceExit",
+    "test": "jest --forceExit --maxWorkers=50%",
     "test:orchestrator": "jest tests/orchestrator.test.js --forceExit",
     "coverage": "jest --coverage --forceExit",
     "test:watch": "jest --watch",
@@ -30,7 +30,11 @@
     "improve": "node scripts/improve.js",
     "extract-examples": "node scripts/extract-examples.js",
     "benchmark:realworld": "node scripts/benchmark-realworld.js",
-    "release": "npm version patch && git push --follow-tags && npm publish"
+    "benchmark:reallife": "node scripts/benchmark-reallife.js",
+    "benchmark:report": "node scripts/benchmark-reallife-report.js",
+    "improve:reallife": "node scripts/improve-reallife.js",
+    "benchmark:gate": "node scripts/benchmark-gate.js",
+    "release": "bash scripts/release.sh"
   },
   "keywords": [
     "ai",
@@ -69,7 +73,9 @@
   "license": "MIT",
   "dependencies": {
     "axios": "^1.7.0",
-    "dotenv": "^16.4.0"
+    "dotenv": "^16.4.0",
+    "pixelmatch": "^7.1.0",
+    "pngjs": "^7.0.0"
   },
   "devDependencies": {
     "esbuild": "^0.27.3",