@vercel/agent-eval 0.0.5 → 0.0.8

package/README.md

@@ -208,6 +208,7 @@ Choose your agent and authentication method:
 agent: 'vercel-ai-gateway/claude-code'  // Claude Code via AI Gateway
 agent: 'vercel-ai-gateway/codex'        // OpenAI Codex via AI Gateway
 agent: 'vercel-ai-gateway/opencode'     // OpenCode via AI Gateway
+agent: 'vercel-ai-gateway/ai-sdk-harness' // Simple AI SDK harness (any model)
 
 // Direct API (uses provider keys directly)
 agent: 'claude-code'  // requires ANTHROPIC_API_KEY
@@ -218,22 +219,28 @@ See the Environment Variables section below for setup instructions.
 
 ### OpenCode Model Configuration
 
-OpenCode uses Vercel AI Gateway exclusively. Models are specified with the `{provider}/{model}` format:
+OpenCode uses Vercel AI Gateway exclusively. Models **must** be specified with the `vercel/{provider}/{model}` format:
 
 ```typescript
 // Anthropic models
-model: 'anthropic/claude-sonnet-4'
-model: 'anthropic/claude-opus-4'
+model: 'vercel/anthropic/claude-sonnet-4'
+model: 'vercel/anthropic/claude-opus-4'
+
+// Minimax models
+model: 'vercel/minimax/minimax-m2.1'
+model: 'vercel/minimax/minimax-m2.1-lightning'
 
 // Moonshot AI (Kimi) models
-model: 'moonshotai/kimi-k2'
-model: 'moonshotai/kimi-k2-thinking'
+model: 'vercel/moonshotai/kimi-k2'
+model: 'vercel/moonshotai/kimi-k2-thinking'
 
 // OpenAI models
-model: 'openai/gpt-4o'
-model: 'openai/o3'
+model: 'vercel/openai/gpt-4o'
+model: 'vercel/openai/o3'
 ```
 
+> **Important:** The `vercel/` prefix is required. OpenCode's config sets up a `vercel` provider, so the model string must start with `vercel/` to route through Vercel AI Gateway correctly. Using just `anthropic/claude-sonnet-4` (without the `vercel/` prefix) will fail with a "provider not found" error.
+
 Under the hood, the agent creates an `opencode.json` config file that configures the Vercel provider:
 
 ```json
@@ -255,6 +262,28 @@ Under the hood, the agent creates an `opencode.json` config file that configures
 
 And runs: `opencode run "<prompt>" --model {provider}/{model} --format json`
 
+### AI SDK Harness Model Configuration
+
+The AI SDK harness (`vercel-ai-gateway/ai-sdk-harness`) is a lightweight agent that works with **any model** available on Vercel AI Gateway. Unlike OpenCode, it uses the standard `{provider}/{model}` format without a `vercel/` prefix:
+
+```typescript
+// Anthropic models
+model: 'anthropic/claude-sonnet-4'
+model: 'anthropic/claude-opus-4'
+
+// Moonshot AI (Kimi) models
+model: 'moonshotai/kimi-k2.5'
+model: 'moonshotai/kimi-k2-thinking'
+
+// Minimax models
+model: 'minimax/minimax-m2.1'
+
+// OpenAI models
+model: 'openai/gpt-4o'
+```
+
+The AI SDK harness includes these tools: `readFile`, `writeFile`, `editFile`, `listFiles`, `glob`, `grep`, and `bash`. It's ideal for evaluating models that may not be fully compatible with OpenCode.
+
 ### Full Configuration
 
 ```typescript
@@ -267,7 +296,8 @@ const config: ExperimentConfig = {
   // Model to use (defaults vary by agent)
   // - claude-code: 'opus'
   // - codex: 'openai/gpt-5.2-codex'
-  // - opencode: 'anthropic/claude-sonnet-4'
+  // - opencode: 'vercel/anthropic/claude-sonnet-4' (note: vercel/ prefix required)
+  // - ai-sdk-harness: 'anthropic/claude-sonnet-4' (works with any AI Gateway model)
   model: 'opus',
 
   // How many times to run each eval
@@ -279,8 +309,8 @@ const config: ExperimentConfig = {
   // npm scripts that must pass after agent finishes
   scripts: ['build', 'lint'],
 
-  // Timeout per run in seconds
-  timeout: 300,
+  // Timeout per run in seconds (default: 600)
+  timeout: 600,
 
   // Filter which evals to run (pick one)
   evals: '*',                                // all (default)

package/dist/cli.js

@@ -5,7 +5,8 @@
 import { Command } from 'commander';
 import { config as dotenvConfig } from 'dotenv';
 import { resolve, dirname, basename } from 'path';
-import { existsSync } from 'fs';
+import { existsSync, readFileSync } from 'fs';
+import { fileURLToPath } from 'url';
 import chalk from 'chalk';
 import { loadConfig, resolveEvalNames } from './lib/config.js';
 import { loadAllFixtures } from './lib/fixture.js';
@@ -13,13 +14,17 @@ import { runExperiment } from './lib/runner.js';
 import { initProject, getPostInitInstructions } from './lib/init.js';
 import { getAgent } from './lib/agents/index.js';
 import { getSandboxBackendInfo } from './lib/sandbox.js';
-// Load environment variables
+// Load environment variables (.env.local first, then .env as fallback)
+dotenvConfig({ path: '.env.local' });
 dotenvConfig();
+// Read version from package.json
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const pkg = JSON.parse(readFileSync(resolve(__dirname, '../package.json'), 'utf-8'));
 const program = new Command();
 program
     .name('agent-eval')
     .description('Framework for testing AI coding agents in isolated sandboxes')
-    .version('0.0.4');
+    .version(pkg.version);
 /**
  * Resolve config path shorthand.
  * - "cc" -> "experiments/cc.ts"
@@ -81,7 +86,7 @@ async function runExperimentCommand(configInput, options) {
         console.log(chalk.blue(`\nRunning ${evalNames.length} eval(s) x ${config.runs} run(s) = ${evalNames.length * config.runs} total runs`));
         console.log(chalk.blue(`Agent: ${config.agent}, Model: ${config.model}, Timeout: ${config.timeout}s, Early Exit: ${config.earlyExit}`));
         // Show which sandbox backend will be used
-        const sandboxInfo = getSandboxBackendInfo();
+        const sandboxInfo = getSandboxBackendInfo({ backend: config.sandbox });
         console.log(chalk.blue(`Sandbox: ${sandboxInfo.description}`));
         if (options.dry) {
             console.log(chalk.yellow('\n[DRY RUN] Would execute evals here'));

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AAEA;;GAEG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,MAAM,IAAI,YAAY,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,MAAM,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AAChC,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,uBAAuB,EAAE,MAAM,eAAe,CAAC;AACrE,OAAO,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AACjD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AAEzD,6BAA6B;AAC7B,YAAY,EAAE,CAAC;AAEf,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;AAE9B,OAAO;KACJ,IAAI,CAAC,YAAY,CAAC;KAClB,WAAW,CAAC,8DAA8D,CAAC;KAC3E,OAAO,CAAC,OAAO,CAAC,CAAC;AAEpB;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,KAAa;IACtC,6DAA6D;IAC7D,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAClG,OAAO,KAAK,CAAC;IACf,CAAC;IACD,6DAA6D;IAC7D,OAAO,eAAe,KAAK,KAAK,CAAC;AACnC,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,oBAAoB,CAAC,WAAmB,EAAE,OAA0B;IACjF,IAAI,CAAC;QACH,MAAM,UAAU,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC;QAClD,MAAM,kBAAkB,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,UAAU,CAAC,CAAC;QAE9D,IAAI,CAAC,UAAU,CAAC,kBAAkB,CAAC,EAAE,CAAC;YACpC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,0BAA0B,kBAAkB,EAAE,CAAC,CAAC,CAAC;YACzE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,uBAAuB,UAAU,KAAK,CAAC,CAAC,CAAC;QAChE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAEpD,mDAAmD;QACnD,kEAAkE;QAClE,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC;QACxD,MAAM,QAAQ,GAAG,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,QAAQ,EAAE,CAAC,CAAC,CAAC;YACnE,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,yDAAyD,CAAC,CAAC,CAAC;YACrF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,wBAAwB,QAAQ,KAAK,CAAC,CAAC,CAAC;QAC/D,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;QAEvD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,cAAc,MAAM,CAAC,MAAM,sBAAsB,CAAC,CAAC,CAAC;YAC7E,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;gBAC3B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,KAAK,CAAC,WAAW,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;QAED,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC,CAAC;YACzD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,6BAA6B;QAC7B,MAAM,cAAc,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QACnD,MAAM,SAAS,GAAG,gBAAgB,CAAC,MAAM,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC;QAEjE,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,6BAA6B,CAAC,CAAC,CAAC;YACxD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,WAAW,QAAQ,CAAC,MAAM,+BAA+B,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACvG,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;YAC7B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAC1C,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,SAAS,CAAC,MAAM,cAAc,MAAM,CAAC,IAAI,aAAa,SAAS,CAAC,MAAM,GAAG,MAAM,CAAC,IAAI,aAAa,CAAC,CAAC,CAAC;QACxI,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,KAAK,YAAY,MAAM,CAAC,KAAK,cAAc,MAAM,CAAC,OAAO,kBAAkB,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAExI,0CAA0C;QAC1C,MAAM,WAAW,GAAG,qBAAqB,EAAE,CAAC;QAC5C,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,YAAY,WAAW,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;QAE/D,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAChB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,sCAAsC,CAAC,CAAC,CAAC;YAClE,OAAO;QACT,CAAC;QAED,8CAA8C;QAC9C,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACrC,MAAM,YAAY,GAAG,KAAK,CAAC,eAAe,EAAE,CAAC;QAC7C,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;QACzC,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,YAAY,mCAAmC,CAAC,CAAC,CAAC;YAC7E,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,iEAAiE,CAAC,CAAC,CAAC;YAC7F,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,kDAAkD;QAClD,MAAM,gBAAgB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;QAE5E,uCAAuC;QACvC,MAAM,cAAc,GAAG,QAAQ,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACxE,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,SAAS,CAAC,CAAC;QAErD,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC,CAAC;QAEpD,qBAAqB;QACrB,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC;YAClC,MAAM;YACN,QAAQ,EAAE,gBAAgB;YAC1B,MAAM;YACN,UAAU;YACV,cAAc;YACd,UAAU,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC;SACtC,CAAC,CAAC;QAEH,6BAA6B;QAC7B,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,SAAS,CAAC,CAAC;QAC3E,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAClC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC;AAED;;GAEG;AACH,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,QAAQ,CAAC,QAAQ,EAAE,+BAA+B,CAAC;KACnD,WAAW,CAAC,iDAAiD,CAAC;KAC9D,MAAM,CAAC,KAAK,EAAE,IAAY,EAAE,EAAE;IAC7B,IAAI,CAAC;QACH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,8BAA8B,IAAI,EAAE,CAAC,CAAC,CAAC;QAE9D,MAAM,UAAU,GAAG,WAAW,CAAC;YAC7B,IAAI;YACJ,SAAS,EAAE,OAAO,CAAC,GAAG,EAAE;SACzB,CAAC,CAAC;QAEH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,+BAA+B,CAAC,CAAC,CAAC;QAC1D,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC,CAAC;IACzD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC,CAAC,CAAC;AAEL;;;GAGG;AACH,OAAO;KACJ,QAAQ,CAAC,UAAU,EAAE,sCAAsC,CAAC;KAC5D,MAAM,CAAC,OAAO,EAAE,0CAA0C,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,WAA+B,EAAE,OAA0B,EAAE,EAAE;IAC5E,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO,CAAC,IAAI,EAAE,CAAC;QACf,OAAO;IACT,CAAC;IACD,MAAM,oBAAoB,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;AACnD,CAAC,CAAC,CAAC;AAEL,OAAO,CAAC,KAAK,EAAE,CAAC"}
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AAEA;;GAEG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,MAAM,IAAI,YAAY,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,MAAM,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,KAAK,MAAM,OAAO,CAAC;AAC1B,OAAO,EAAE,UAAU,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AAChD,OAAO,EAAE,WAAW,EAAE,uBAAuB,EAAE,MAAM,eAAe,CAAC;AACrE,OAAO,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AACjD,OAAO,EAAE,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AAEzD,uEAAuE;AACvE,YAAY,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;AACrC,YAAY,EAAE,CAAC;AAEf,iCAAiC;AACjC,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,OAAO,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC;AAErF,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;AAE9B,OAAO;KACJ,IAAI,CAAC,YAAY,CAAC;KAClB,WAAW,CAAC,8DAA8D,CAAC;KAC3E,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAExB;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,KAAa;IACtC,6DAA6D;IAC7D,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAClG,OAAO,KAAK,CAAC;IACf,CAAC;IACD,6DAA6D;IAC7D,OAAO,eAAe,KAAK,KAAK,CAAC;AACnC,CAAC;AAED;;GAEG;AACH,KAAK,UAAU,oBAAoB,CAAC,WAAmB,EAAE,OAA0B;IACjF,IAAI,CAAC;QACH,MAAM,UAAU,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC;QAClD,MAAM,kBAAkB,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,UAAU,CAAC,CAAC;QAE9D,IAAI,CAAC,UAAU,CAAC,kBAAkB,CAAC,EAAE,CAAC;YACpC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,0BAA0B,kBAAkB,EAAE,CAAC,CAAC,CAAC;YACzE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,uBAAuB,UAAU,KAAK,CAAC,CAAC,CAAC;QAChE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAEpD,mDAAmD;QACnD,kEAAkE;QAClE,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC;QACxD,MAAM,QAAQ,GAAG,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,QAAQ,EAAE,CAAC,CAAC,CAAC;YACnE,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,yDAAyD,CAAC,CAAC,CAAC;YACrF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,wBAAwB,QAAQ,KAAK,CAAC,CAAC,CAAC;QAC/D,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;QAEvD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,cAAc,MAAM,CAAC,MAAM,sBAAsB,CAAC,CAAC,CAAC;YAC7E,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;gBAC3B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,KAAK,CAAC,WAAW,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;QAED,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,CAAC,CAAC,CAAC;YACzD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,6BAA6B;QAC7B,MAAM,cAAc,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QACnD,MAAM,SAAS,GAAG,gBAAgB,CAAC,MAAM,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC;QAEjE,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,6BAA6B,CAAC,CAAC,CAAC;YACxD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,WAAW,QAAQ,CAAC,MAAM,+BAA+B,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACvG,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;YAC7B,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAC1C,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,SAAS,CAAC,MAAM,cAAc,MAAM,CAAC,IAAI,aAAa,SAAS,CAAC,MAAM,GAAG,MAAM,CAAC,IAAI,aAAa,CAAC,CAAC,CAAC;QACxI,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,KAAK,YAAY,MAAM,CAAC,KAAK,cAAc,MAAM,CAAC,OAAO,kBAAkB,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAExI,0CAA0C;QAC1C,MAAM,WAAW,GAAG,qBAAqB,CAAC,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;QACvE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,YAAY,WAAW,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;QAE/D,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;YAChB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,sCAAsC,CAAC,CAAC,CAAC;YAClE,OAAO;QACT,CAAC;QAED,8CAA8C;QAC9C,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACrC,MAAM,YAAY,GAAG,KAAK,CAAC,eAAe,EAAE,CAAC;QAC7C,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;QACzC,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,YAAY,mCAAmC,CAAC,CAAC,CAAC;YAC7E,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,iEAAiE,CAAC,CAAC,CAAC;YAC7F,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,kDAAkD;QAClD,MAAM,gBAAgB,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;QAE5E,uCAAuC;QACvC,MAAM,cAAc,GAAG,QAAQ,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;QACxE,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,SAAS,CAAC,CAAC;QAErD,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC,CAAC;QAEpD,qBAAqB;QACrB,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC;YAClC,MAAM;YACN,QAAQ,EAAE,gBAAgB;YAC1B,MAAM;YACN,UAAU;YACV,cAAc;YACd,UAAU,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC;SACtC,CAAC,CAAC;QAEH,6BAA6B;QAC7B,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,SAAS,CAAC,CAAC;QAC3E,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAClC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC;AAED;;GAEG;AACH,OAAO;KACJ,OAAO,CAAC,MAAM,CAAC;KACf,QAAQ,CAAC,QAAQ,EAAE,+BAA+B,CAAC;KACnD,WAAW,CAAC,iDAAiD,CAAC;KAC9D,MAAM,CAAC,KAAK,EAAE,IAAY,EAAE,EAAE;IAC7B,IAAI,CAAC;QACH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,8BAA8B,IAAI,EAAE,CAAC,CAAC,CAAC;QAE9D,MAAM,UAAU,GAAG,WAAW,CAAC;YAC7B,IAAI;YACJ,SAAS,EAAE,OAAO,CAAC,GAAG,EAAE;SACzB,CAAC,CAAC;QAEH,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,+BAA+B,CAAC,CAAC,CAAC;QAC1D,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC,CAAC;IACzD,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,KAAK,EAAE,CAAC;YAC3B,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACtD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,2BAA2B,CAAC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC,CAAC,CAAC;AAEL;;;GAGG;AACH,OAAO;KACJ,QAAQ,CAAC,UAAU,EAAE,sCAAsC,CAAC;KAC5D,MAAM,CAAC,OAAO,EAAE,0CAA0C,CAAC;KAC3D,MAAM,CAAC,KAAK,EAAE,WAA+B,EAAE,OAA0B,EAAE,EAAE;IAC5E,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO,CAAC,IAAI,EAAE,CAAC;QACf,OAAO;IACT,CAAC;IACD,MAAM,oBAAoB,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;AACnD,CAAC,CAAC,CAAC;AAEL,OAAO,CAAC,KAAK,EAAE,CAAC"}

package/dist/lib/agents/ai-sdk-agent.d.ts

@@ -0,0 +1,10 @@
+/**
+ * AI SDK Agent - A simple coding agent using the Vercel AI SDK.
+ * Works with any model available on Vercel AI Gateway.
+ */
+import type { Agent } from './types.js';
+/**
+ * Create AI SDK agent with Vercel AI Gateway authentication.
+ */
+export declare function createAiSdkAgent(): Agent;
+//# sourceMappingURL=ai-sdk-agent.d.ts.map

package/dist/lib/agents/ai-sdk-agent.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ai-sdk-agent.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/ai-sdk-agent.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AA8QzE;;GAEG;AACH,wBAAgB,gBAAgB,IAAI,KAAK,CAiMxC"}

package/dist/lib/agents/ai-sdk-agent.js

@@ -0,0 +1,427 @@
+/**
+ * AI SDK Agent - A simple coding agent using the Vercel AI SDK.
+ * Works with any model available on Vercel AI Gateway.
+ */
+import { createSandbox, collectLocalFiles, splitTestFiles, verifyNoTestFiles, } from '../sandbox.js';
+import { runValidation, captureGeneratedFiles, createVitestConfig, AI_GATEWAY, } from './shared.js';
+/**
+ * The CLI script source code that runs inside the sandbox.
+ * This is a self-contained script that uses the AI SDK.
+ */
+const CLI_SCRIPT = `
+import { generateText, tool, stepCountIs } from 'ai';
+import { createGateway } from '@ai-sdk/gateway';
+import { z } from 'zod';
+import { readFileSync, writeFileSync, readdirSync, existsSync, statSync } from 'fs';
+import { execSync } from 'child_process';
+import { join, dirname } from 'path';
+import { mkdirSync } from 'fs';
+
+// Parse command line arguments
+const args = process.argv.slice(2);
+let prompt = '';
+let model = '';
+
+for (let i = 0; i < args.length; i++) {
+  if (args[i] === '--prompt' && args[i + 1]) {
+    prompt = args[++i];
+  } else if (args[i] === '--model' && args[i + 1]) {
+    model = args[++i];
+  }
+}
+
+if (!prompt || !model) {
+  console.error('Usage: ai-sdk-agent --prompt "..." --model "provider/model"');
+  process.exit(1);
+}
+
+// Create AI Gateway client
+const gateway = createGateway({
+  apiKey: process.env.AI_GATEWAY_API_KEY,
+});
+
+// Transcript events
+const events = [];
+
+function logEvent(type, data) {
+  const event = { type, timestamp: Date.now(), ...data };
+  events.push(event);
+  console.log(JSON.stringify(event));
+}
+
+// Define coding tools
+const tools = {
+  readFile: tool({
+    description: 'Read the contents of a file at the given path',
+    inputSchema: z.object({
+      path: z.string().describe('The file path to read'),
+    }),
+    execute: async ({ path }) => {
+      try {
+        const content = readFileSync(path, 'utf-8');
+        logEvent('tool_result', { tool: 'readFile', path, success: true });
+        return content;
+      } catch (error) {
+        logEvent('tool_result', { tool: 'readFile', path, success: false, error: error.message });
+        return \`Error reading file: \${error.message}\`;
+      }
+    },
+  }),
+
+  writeFile: tool({
+    description: 'Write content to a file at the given path. Creates directories if needed.',
+    inputSchema: z.object({
+      path: z.string().describe('The file path to write'),
+      content: z.string().describe('The content to write'),
+    }),
+    execute: async ({ path, content }) => {
+      try {
+        const dir = dirname(path);
+        if (dir && dir !== '.') {
+          mkdirSync(dir, { recursive: true });
+        }
+        writeFileSync(path, content);
+        logEvent('tool_result', { tool: 'writeFile', path, success: true });
+        return 'File written successfully';
+      } catch (error) {
+        logEvent('tool_result', { tool: 'writeFile', path, success: false, error: error.message });
+        return \`Error writing file: \${error.message}\`;
+      }
+    },
+  }),
+
+  editFile: tool({
+    description: 'Edit a file by replacing a specific string with new content',
+    inputSchema: z.object({
+      path: z.string().describe('The file path to edit'),
+      oldString: z.string().describe('The exact string to find and replace'),
+      newString: z.string().describe('The replacement string'),
+    }),
+    execute: async ({ path, oldString, newString }) => {
+      try {
+        const content = readFileSync(path, 'utf-8');
+        if (!content.includes(oldString)) {
+          logEvent('tool_result', { tool: 'editFile', path, success: false, error: 'String not found' });
+          return 'Error: The specified string was not found in the file';
+        }
+        const newContent = content.replace(oldString, newString);
+        writeFileSync(path, newContent);
+        logEvent('tool_result', { tool: 'editFile', path, success: true });
+        return 'File edited successfully';
+      } catch (error) {
+        logEvent('tool_result', { tool: 'editFile', path, success: false, error: error.message });
+        return \`Error editing file: \${error.message}\`;
+      }
+    },
+  }),
+
+  listFiles: tool({
+    description: 'List files in a directory. Call with path="." to list current directory.',
+    inputSchema: z.object({
+      path: z.string().describe('The directory path to list (use "." for current directory)'),
+      recursive: z.boolean().describe('Whether to list recursively').optional(),
+    }),
+    execute: async ({ path, recursive }) => {
+      const targetPath = path || '.';
+      const isRecursive = recursive || false;
+      try {
+        if (isRecursive) {
+          const result = execSync(\`find \${targetPath} -type f | head -100\`, { encoding: 'utf-8' });
+          logEvent('tool_result', { tool: 'listFiles', path: targetPath, recursive: isRecursive, success: true });
+          return result;
+        }
+        const files = readdirSync(targetPath);
+        logEvent('tool_result', { tool: 'listFiles', path: targetPath, recursive: isRecursive, success: true });
+        return files.join('\\n');
+      } catch (error) {
+        logEvent('tool_result', { tool: 'listFiles', path: targetPath, success: false, error: error.message });
+        return \`Error listing files: \${error.message}\`;
+      }
+    },
+  }),
+
+  glob: tool({
+    description: 'Find files matching a pattern (e.g., "*.ts" for TypeScript files)',
+    inputSchema: z.object({
+      pattern: z.string().describe('The file pattern (e.g., "*.ts", "*.js")'),
+    }),
+    execute: async ({ pattern }) => {
+      try {
+        // Extract just the file pattern, remove any path prefix
+        const filePattern = pattern.replace(/^\\*\\*\\//, '').replace(/^\\.\\//, '');
+        const result = execSync(\`find . -name "\${filePattern}" -type f 2>/dev/null | grep -v node_modules | head -50\`, { encoding: 'utf-8' });
+        logEvent('tool_result', { tool: 'glob', pattern, success: true });
+        return result.trim() || 'No files found';
+      } catch (error) {
+        logEvent('tool_result', { tool: 'glob', pattern, success: false, error: error.message });
+        return 'No files found';
+      }
+    },
+  }),
+
+  grep: tool({
+    description: 'Search for a text pattern in files',
+    inputSchema: z.object({
+      pattern: z.string().describe('The search pattern'),
+      path: z.string().describe('The file or directory to search in').optional(),
+    }),
+    execute: async ({ pattern, path }) => {
+      const targetPath = path || '.';
+      try {
+        const result = execSync(\`grep -rn "\${pattern}" \${targetPath} 2>/dev/null | grep -v node_modules | head -50\`, { encoding: 'utf-8' });
+        logEvent('tool_result', { tool: 'grep', pattern, path: targetPath, success: true });
+        return result.trim() || 'No matches found';
+      } catch (error) {
+        logEvent('tool_result', { tool: 'grep', pattern, path: targetPath, success: false });
+        return 'No matches found';
+      }
+    },
+  }),
+
+  bash: tool({
+    description: 'Run a bash command',
+    inputSchema: z.object({
+      command: z.string().describe('The command to run'),
+    }),
+    execute: async ({ command }) => {
+      try {
+        const result = execSync(command, { encoding: 'utf-8', timeout: 30000 });
+        logEvent('tool_result', { tool: 'bash', command, success: true });
+        return result;
+      } catch (error) {
+        logEvent('tool_result', { tool: 'bash', command, success: false, error: error.message });
+        return \`Error: \${error.message}\\n\${error.stdout || ''}\\n\${error.stderr || ''}\`;
+      }
+    },
+  }),
+};
+
+// System prompt for the coding agent
+const systemPrompt = \`You are an expert coding agent. Your job is to complete programming tasks by reading, writing, and modifying files.
+
+Available tools:
+- readFile(path): Read a file's contents
+- writeFile(path, content): Write/create a file (creates directories if needed)
+- editFile(path, oldString, newString): Replace a specific string in a file
+- listFiles(path): List files in a directory (use path="." for current directory)
+- glob(pattern): Find files by pattern (e.g., "*.ts")
+- grep(pattern, path): Search for text in files
+- bash(command): Run shell commands
+
+IMPORTANT WORKFLOW:
+1. First, list files to understand the project structure: listFiles(path=".")
+2. Read any relevant existing files to understand the context
+3. Make the necessary code changes using writeFile or editFile
+4. If needed, run build/test commands with bash to verify
+
+RULES:
+- Always check what files exist before modifying them
+- Create complete, working code - not placeholders
+- Put files in the correct directories (e.g., src/ for source files)
+- Be thorough but efficient\`;
+
+// Run the agent
+async function main() {
+  logEvent('start', { model, prompt });
+
+  try {
+    const result = await generateText({
+      model: gateway(model),
+      tools,
+      stopWhen: stepCountIs(100), // Allow up to 100 steps
+      system: systemPrompt,
+      prompt,
+      onStepFinish: ({ stepType, text, toolCalls, toolResults }) => {
+        logEvent('step', { stepType, text, toolCalls: toolCalls?.length, toolResults: toolResults?.length });
+      },
+    });
+
+    logEvent('complete', {
+      success: true,
+      steps: result.steps.length,
+      text: result.text,
+    });
+  } catch (error) {
+    logEvent('error', {
+      success: false,
+      error: error.message,
+      name: error.name,
+    });
+    process.exit(1);
+  }
+}
+
+main();
+`;
+/**
+ * Create AI SDK agent with Vercel AI Gateway authentication.
+ */
+export function createAiSdkAgent() {
+    return {
+        name: 'vercel-ai-gateway/ai-sdk-harness',
+        displayName: 'AI SDK Harness (Vercel AI Gateway)',
+        getApiKeyEnvVar() {
+            return AI_GATEWAY.apiKeyEnvVar;
+        },
+        getDefaultModel() {
+            return 'anthropic/claude-sonnet-4';
+        },
+        async run(fixturePath, options) {
+            const startTime = Date.now();
+            let sandbox = null;
+            let agentOutput = '';
+            let aborted = false;
+            let sandboxStopped = false;
+            // Handle abort signal
+            const abortHandler = () => {
+                aborted = true;
+                if (sandbox && !sandboxStopped) {
+                    sandboxStopped = true;
+                    sandbox.stop().catch(() => { });
+                }
+            };
+            if (options.signal) {
+                if (options.signal.aborted) {
+                    return {
+                        success: false,
+                        output: '',
+                        error: 'Aborted before start',
+                        duration: 0,
+                    };
+                }
+                options.signal.addEventListener('abort', abortHandler);
+            }
+            try {
+                // Collect files from fixture
+                const allFiles = await collectLocalFiles(fixturePath);
+                const { workspaceFiles, testFiles } = splitTestFiles(allFiles);
+                // Check for abort before expensive operations
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: '',
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                    };
+                }
+                // Create sandbox
+                sandbox = await createSandbox({
+                    timeout: options.timeout,
+                    runtime: 'node24',
+                    backend: options.sandbox,
+                });
+                // Check for abort after sandbox creation
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: '',
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox.sandboxId,
+                    };
+                }
+                // Upload workspace files (excluding tests)
+                await sandbox.uploadFiles(workspaceFiles);
+                // Run setup function if provided
+                if (options.setup) {
+                    await options.setup(sandbox);
+                }
+                // Install dependencies
+                const installResult = await sandbox.runCommand('npm', ['install']);
+                if (installResult.exitCode !== 0) {
+                    throw new Error(`npm install failed: ${installResult.stderr}`);
+                }
+                // Install AI SDK dependencies
+                const aiInstall = await sandbox.runCommand('npm', [
+                    'install',
+                    'ai@^5.0.11',
+                    '@ai-sdk/gateway@^1.0.0',
+                    'zod@^3.23.8',
+                ]);
+                if (aiInstall.exitCode !== 0) {
+                    throw new Error(`AI SDK install failed: ${aiInstall.stderr}`);
+                }
+                // Write the CLI script to the sandbox
+                await sandbox.writeFiles({
+                    'ai-sdk-agent.mjs': CLI_SCRIPT,
+                });
+                // Verify no test files in sandbox
+                await verifyNoTestFiles(sandbox);
+                // Run the AI SDK agent
+                const agentResult = await sandbox.runCommand('node', [
+                    'ai-sdk-agent.mjs',
+                    '--prompt',
+                    options.prompt,
+                    '--model',
+                    options.model,
+                ], {
+                    env: {
+                        [AI_GATEWAY.apiKeyEnvVar]: options.apiKey,
+                    },
+                });
+                agentOutput = agentResult.stdout + agentResult.stderr;
+                if (agentResult.exitCode !== 0) {
+                    // Extract meaningful error from output
+                    const errorLines = agentOutput.trim().split('\n').slice(-5).join('\n');
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        error: errorLines || `AI SDK agent exited with code ${agentResult.exitCode}`,
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox.sandboxId,
+                    };
+                }
+                // Upload test files for validation
+                await sandbox.uploadFiles(testFiles);
+                // Create vitest config for EVAL.ts/tsx
+                await createVitestConfig(sandbox);
+                // The agent outputs JSON events, use that as transcript
+                const transcript = agentOutput;
+                // Run validation scripts
+                const validationResults = await runValidation(sandbox, options.scripts ?? []);
+                // Capture generated files
+                const generatedFiles = await captureGeneratedFiles(sandbox);
+                return {
+                    success: validationResults.allPassed,
+                    output: agentOutput,
+                    transcript,
+                    duration: Date.now() - startTime,
+                    testResult: validationResults.test,
+                    scriptsResults: validationResults.scripts,
+                    sandboxId: sandbox.sandboxId,
+                    generatedFiles,
+                };
+            }
+            catch (error) {
+                // Check if this was an abort
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox?.sandboxId,
+                    };
+                }
+                return {
+                    success: false,
+                    output: agentOutput,
+                    error: error instanceof Error ? error.message : String(error),
+                    duration: Date.now() - startTime,
+                    sandboxId: sandbox?.sandboxId,
+                };
+            }
+            finally {
+                // Clean up abort listener
+                if (options.signal) {
+                    options.signal.removeEventListener('abort', abortHandler);
+                }
+                if (sandbox && !sandboxStopped) {
+                    sandboxStopped = true;
+                    await sandbox.stop();
+                }
+            }
+        },
+    };
+}
+//# sourceMappingURL=ai-sdk-agent.js.map

package/dist/lib/agents/ai-sdk-agent.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ai-sdk-agent.js","sourceRoot":"","sources":["../../../src/lib/agents/ai-sdk-agent.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,OAAO,EACL,aAAa,EACb,iBAAiB,EACjB,cAAc,EACd,iBAAiB,GAElB,MAAM,eAAe,CAAC;AAEvB,OAAO,EACL,aAAa,EACb,qBAAqB,EACrB,kBAAkB,EAClB,UAAU,GACX,MAAM,aAAa,CAAC;AAKrB;;;GAGG;AACH,MAAM,UAAU,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAoPlB,CAAC;AAEF;;GAEG;AACH,MAAM,UAAU,gBAAgB;IAC9B,OAAO;QACL,IAAI,EAAE,kCAAkC;QACxC,WAAW,EAAE,oCAAoC;QAEjD,eAAe;YACb,OAAO,UAAU,CAAC,YAAY,CAAC;QACjC,CAAC;QAED,eAAe;YACb,OAAO,2BAA2B,CAAC;QACrC,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACrD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,OAAO,GAAsB,IAAI,CAAC;YACtC,IAAI,WAAW,GAAG,EAAE,CAAC;YACrB,IAAI,OAAO,GAAG,KAAK,CAAC;YACpB,IAAI,cAAc,GAAG,KAAK,CAAC;YAE3B,sBAAsB;YACtB,MAAM,YAAY,GAAG,GAAG,EAAE;gBACxB,OAAO,GAAG,IAAI,CAAC;gBACf,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;gBACjC,CAAC;YACH,CAAC,CAAC;YAEF,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACnB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC3B,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,sBAAsB;wBAC7B,QAAQ,EAAE,CAAC;qBACZ,CAAC;gBACJ,CAAC;gBACD,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;YACzD,CAAC;YAED,IAAI,CAAC;gBACH,6BAA6B;gBAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAAC;gBACtD,MAAM,EAAE,cAAc,EAAE,SAAS,EAAE,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;gBAE/D,8CAA8C;gBAC9C,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;qBACjC,CAAC;gBACJ,CAAC;gBAED,iBAAiB;gBACjB,OAAO,GAAG,MAAM,aAAa,CAAC;oBAC5B,OAAO,EAAE,OAAO,CAAC,OAAO;oBACxB,OAAO,EAAE,QAAQ;oBACjB,OAAO,EAAE,OAAO,CAAC,OAAO;iBACzB,CAAC,CAAC;gBAEH,yCAAyC;gBACzC,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,2CAA2C;gBAC3C,MAAM,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC;gBAE1C,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,MAAM,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACnE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,IAAI,KAAK,CAAC,uBAAuB,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC;gBACjE,CAAC;gBAED,8BAA8B;gBAC9B,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBAChD,SAAS;oBACT,YAAY;oBACZ,wBAAwB;oBACxB,aAAa;iBACd,CAAC,CAAC;gBACH,IAAI,SAAS,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC7B,MAAM,IAAI,KAAK,CAAC,0BAA0B,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC;gBAChE,CAAC;gBAED,sCAAsC;gBACtC,MAAM,OAAO,CAAC,UAAU,CAAC;oBACvB,kBAAkB,EAAE,UAAU;iBAC/B,CAAC,CAAC;gBAEH,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,uBAAuB;gBACvB,MAAM,WAAW,GAAG,MAAM,OAAO,CAAC,UAAU,CAC1C,MAAM,EACN;oBACE,kBAAkB;oBAClB,UAAU;oBACV,OAAO,CAAC,MAAM;oBACd,SAAS;oBACT,OAAO,CAAC,KAAK;iBACd,EACD;oBACE,GAAG,EAAE;wBACH,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;qBAC1C;iBACF,CACF,CAAC;gBAEF,WAAW,GAAG,WAAW,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC;gBAEtD,IAAI,WAAW,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC/B,uCAAuC;oBACvC,MAAM,UAAU,GAAG,WAAW,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACvE,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,UAAU,IAAI,iCAAiC,WAAW,CAAC,QAAQ,EAAE;wBAC5E,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,mCAAmC;gBACnC,MAAM,OAAO,CAAC,WAAW,CAAC,SAAS,CAAC,CAAC;gBAErC,uCAAuC;gBACvC,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;gBAElC,wDAAwD;gBACxD,MAAM,UAAU,GAAG,WAAW,CAAC;gBAE/B,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE5D,OAAO;oBACL,OAAO,EAAE,iBAAiB,CAAC,SAAS;oBACpC,MAAM,EAAE,WAAW;oBACnB,UAAU;oBACV,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,UAAU,EAAE,iBAAiB,CAAC,IAAI;oBAClC,cAAc,EAAE,iBAAiB,CAAC,OAAO;oBACzC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,cAAc;iBACf,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,6BAA6B;gBAC7B,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;qBAC9B,CAAC;gBACJ,CAAC;gBACD,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE,WAAW;oBACnB,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;oBAC7D,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;iBAC9B,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,0BAA0B;gBAC1B,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;oBACnB,OAAO,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;gBAC5D,CAAC;gBACD,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;gBACvB,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC"}

package/dist/lib/agents/claude-code.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"claude-code.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/claude-code.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AAmDzE;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,KAAK,CA4LpG"}
+{"version":3,"file":"claude-code.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/claude-code.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AAmDzE;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,KAAK,CA6LpG"}

package/dist/lib/agents/claude-code.js

@@ -80,10 +80,11 @@ export function createClaudeCodeAgent({ useVercelAiGateway }) {
                         duration: Date.now() - startTime,
                     };
                 }
-                // Create sandbox (auto-detects backend based on env)
+                // Create sandbox
                 sandbox = await createSandbox({
                     timeout: options.timeout,
                     runtime: 'node24',
+                    backend: options.sandbox,
                 });
                 // Check for abort after sandbox creation (abort may have fired during create)
                 if (aborted) {