More fixes

Author: cte

apps/web-evals/src/lib/server/runs.ts

@@ -56,15 +56,11 @@ export async function createRun({ suite, exercises = [], systemPrompt, ...values
 			? { ...process.env, FOOTGUN_SYSTEM_PROMPT: systemPrompt }
 			: process.env
 
-		const childProcess = spawn(
-			"pnpm",
-			["--filter", "@roo-code/evals", "dev", "run", "all", "--runId", run.id.toString()],
-			{
-				detached: true,
-				stdio: ["ignore", logFile, logFile],
-				env,
-			},
-		)
+		const childProcess = spawn("pnpm", ["--filter", "@roo-code/evals", "cli", run.id.toString()], {
+			detached: true,
+			stdio: ["ignore", logFile, logFile],
+			env,
+		})
 
 		childProcess.unref()
 		await _updateRun(run.id, { pid: childProcess.pid })

packages/evals/Dockerfile

@@ -7,7 +7,6 @@ RUN npm install -g npm-run-all
 # Install dependencies
 RUN apt update && apt install -y sudo curl git vim jq
 
-
 # Create a `vscode` user
 RUN useradd -m vscode -s /bin/bash && \
   echo "vscode ALL=(ALL) NOPASSWD:ALL" > /etc/sudoers.d/vscode && \
@@ -72,7 +71,7 @@ RUN echo 'source $HOME/.cargo/env' >> $HOME/.bashrc
  COPY --chown=vscode:vscode ./evals/.env ./
 
  # Push database schema
- RUN pnpm --filter @roo-code/evals db:push
+ RUN pnpm --filter @roo-code/evals db:push --force
 
  EXPOSE 3000
  CMD ["pnpm", "web"]

packages/evals/README.md

@@ -14,7 +14,7 @@ cd Roo-Code
 Run the setup script:
 
 ```sh
-cd evals
+cd packages/evals
 ./scripts/setup.sh
 ```
 

packages/evals/package.json

@@ -9,7 +9,7 @@
 		"check-types": "tsc --noEmit",
 		"_test": "dotenvx run -f .env.test -- vitest run",
 		"clean": "rimraf dist .turbo",
-		"cli": "dotenvx run -f .env.development -- tsx src/cli/index.ts",
+		"cli": "dotenvx run -f .env.development .env.local -- tsx src/cli/index.ts",
 		"drizzle-kit": "dotenvx run -f .env.development -- tsx node_modules/drizzle-kit/bin.cjs",
 		"drizzle-kit:test": "dotenvx run -f .env.test -- tsx node_modules/drizzle-kit/bin.cjs",
 		"db:start": "docker compose up -d",

packages/evals/scripts/copy-run.mts

@@ -318,19 +318,19 @@ else
   echo "✅ Done"
 fi
 
-if [[ ! -s .env ]]; then
-  cp .env.sample .env || exit 1
+if [[ ! -s .env.local ]]; then
+  touch .env.local || exit 1
 fi
 
 echo -n "🗄️ Syncing Roo Code evals database... "
 pnpm --filter @roo-code/evals db:push --force &>/dev/null || exit 1
 echo "✅ Done"
 
-if ! grep -q "OPENROUTER_API_KEY" .env; then
+if ! grep -q "OPENROUTER_API_KEY" .env.local; then
   read -p "🔐 Enter your OpenRouter API key (sk-or-v1-...): " openrouter_api_key
   echo "🔑 Validating..."
   curl --silent --fail https://openrouter.ai/api/v1/key -H "Authorization: Bearer $openrouter_api_key" &>/dev/null || exit 1
-  echo "OPENROUTER_API_KEY=$openrouter_api_key" >> .env || exit 1
+  echo "OPENROUTER_API_KEY=$openrouter_api_key" >> .env.local || exit 1
 fi
 
 current_version=$(code --list-extensions --show-versions 2>/dev/null | grep roo)
@@ -346,9 +346,9 @@ if ! nc -z localhost 3000; then
   read -p "🌐 Would you like to start the evals web app? (Y/n): " start_evals
 
   if [[ "$start_evals" =~ ^[Yy]|^$ ]]; then
-    pnpm web
+    pnpm --filter @roo-code/web-evals dev
   else
-    echo "💡 You can start it anytime with 'pnpm web'."
+    echo "💡 You can start it anytime with 'pnpm --filter @roo-code/web-evals dev'."
   fi
 else
   echo "👟 The evals web app is running at http://localhost:3000"