agent-eval-opencode 0.10.0

package/dist/lib/agents/cursor.js

@@ -0,0 +1,204 @@
+/**
+ * Cursor CLI agent implementation.
+ * Uses direct Cursor API access.
+ */
+import { createSandbox, collectLocalFiles, splitTestFiles, verifyNoTestFiles, } from '../sandbox.js';
+import { runValidation, captureGeneratedFiles, createVitestConfig, CURSOR_DIRECT, initGitAndCommit, injectTranscriptContext, } from './shared.js';
+/**
+ * Extract transcript from Cursor CLI stream-json output.
+ * When run with --output-format stream-json, Cursor outputs JSONL (newline-delimited JSON).
+ */
+function extractTranscriptFromOutput(output) {
+    if (!output || !output.trim()) {
+        return undefined;
+    }
+    // The --output-format stream-json output contains JSON events, one per line
+    // Filter to only include lines that look like JSON objects
+    const lines = output.split('\n').filter(line => {
+        const trimmed = line.trim();
+        return trimmed.startsWith('{') && trimmed.endsWith('}');
+    });
+    if (lines.length === 0) {
+        return undefined;
+    }
+    return lines.join('\n');
+}
+/**
+ * Create Cursor CLI agent with direct API authentication.
+ */
+export function createCursorAgent() {
+    return {
+        name: 'cursor',
+        displayName: 'Cursor CLI',
+        getApiKeyEnvVar() {
+            return CURSOR_DIRECT.apiKeyEnvVar;
+        },
+        getDefaultModel() {
+            return 'composer-1.5';
+        },
+        async run(fixturePath, options) {
+            const startTime = Date.now();
+            let sandbox = null;
+            let agentOutput = '';
+            let transcript;
+            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 (abort may have fired during create)
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: '',
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox.sandboxId,
+                    };
+                }
+                // Upload workspace files (excluding tests)
+                await sandbox.uploadFiles(workspaceFiles);
+                await initGitAndCommit(sandbox);
+                // Run setup function if provided
+                if (options.setup) {
+                    await options.setup(sandbox);
+                }
+                // Install dependencies
+                let installResult = await sandbox.runCommand('npm', ['install']);
+                if (installResult.exitCode !== 0) {
+                    installResult = await sandbox.runCommand('npm', ['install']);
+                }
+                if (installResult.exitCode !== 0) {
+                    const output = (installResult.stdout + installResult.stderr).trim().split('\n').slice(-10).join('\n');
+                    throw new Error(`npm install failed (exit code ${installResult.exitCode}):\n${output}`);
+                }
+                // Install Cursor CLI globally using official installation script
+                const cliInstall = await sandbox.runShell('curl https://cursor.com/install -fsSL | bash');
+                if (cliInstall.exitCode !== 0) {
+                    throw new Error(`Cursor CLI install failed: ${cliInstall.stderr}`);
+                }
+                // Verify no test files in sandbox
+                await verifyNoTestFiles(sandbox);
+                // Run Cursor CLI with direct API access
+                // --print: non-interactive mode (required for scripts/headless)
+                // --force: auto-approve all tool operations
+                // --output-format stream-json: structured JSONL transcript (only works with --print)
+                const cursorResult = await sandbox.runCommand('agent', [
+                    options.prompt,
+                    '--print',
+                    '--force',
+                    '--model',
+                    options.model,
+                    '--output-format',
+                    'stream-json',
+                ], {
+                    env: {
+                        [CURSOR_DIRECT.apiKeyEnvVar]: options.apiKey,
+                    },
+                });
+                agentOutput = cursorResult.stdout + cursorResult.stderr;
+                transcript = extractTranscriptFromOutput(agentOutput);
+                if (cursorResult.exitCode !== 0) {
+                    // Extract meaningful error from output (last few lines usually contain the error)
+                    const errorLines = agentOutput.trim().split('\n').slice(-5).join('\n');
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        transcript,
+                        error: errorLines || `Cursor CLI exited with code ${cursorResult.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);
+                // Inject transcript context so EVAL.ts tests can assert on agent behavior
+                await injectTranscriptContext(sandbox, transcript, 'cursor', options.model);
+                // Run validation scripts
+                const validationResults = await runValidation(sandbox, options.scripts ?? []);
+                // Capture generated files
+                const { generatedFiles, deletedFiles } = await captureGeneratedFiles(sandbox);
+                return {
+                    success: validationResults.allPassed,
+                    output: agentOutput,
+                    transcript,
+                    duration: Date.now() - startTime,
+                    testResult: validationResults.test,
+                    scriptsResults: validationResults.scripts,
+                    sandboxId: sandbox.sandboxId,
+                    generatedFiles,
+                    deletedFiles,
+                };
+            }
+            catch (error) {
+                // Check if this was an abort
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        transcript,
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox?.sandboxId,
+                    };
+                }
+                return {
+                    success: false,
+                    output: agentOutput,
+                    transcript,
+                    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=cursor.js.map

package/dist/lib/agents/cursor.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cursor.js","sourceRoot":"","sources":["../../../src/lib/agents/cursor.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,aAAa,EACb,gBAAgB,EAChB,uBAAuB,GACxB,MAAM,aAAa,CAAC;AAKrB;;;GAGG;AACH,SAAS,2BAA2B,CAAC,MAAc;IACjD,IAAI,CAAC,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QAC9B,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,4EAA4E;IAC5E,2DAA2D;IAC3D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;QAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB;IAC/B,OAAO;QACL,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,YAAY;QAEzB,eAAe;YACb,OAAO,aAAa,CAAC,YAAY,CAAC;QACpC,CAAC;QAED,eAAe;YACb,OAAO,cAAc,CAAC;QACxB,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,UAA8B,CAAC;YACnC,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,8EAA8E;gBAC9E,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,MAAM,gBAAgB,CAAC,OAAO,CAAC,CAAC;gBAEhC,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,IAAI,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACjE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBAC/D,CAAC;gBACD,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,MAAM,GAAG,CAAC,aAAa,CAAC,MAAM,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACtG,MAAM,IAAI,KAAK,CAAC,iCAAiC,aAAa,CAAC,QAAQ,OAAO,MAAM,EAAE,CAAC,CAAC;gBAC1F,CAAC;gBAED,iEAAiE;gBACjE,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CACvC,8CAA8C,CAC/C,CAAC;gBACF,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,8BAA8B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACrE,CAAC;gBAED,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,wCAAwC;gBACxC,gEAAgE;gBAChE,4CAA4C;gBAC5C,qFAAqF;gBACrF,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAC3C,OAAO,EACP;oBACE,OAAO,CAAC,MAAM;oBACd,SAAS;oBACT,SAAS;oBACT,SAAS;oBACT,OAAO,CAAC,KAAK;oBACb,iBAAiB;oBACjB,aAAa;iBACd,EACD;oBACE,GAAG,EAAE;wBACH,CAAC,aAAa,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;qBAC7C;iBACF,CACF,CAAC;gBAEF,WAAW,GAAG,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC;gBACxD,UAAU,GAAG,2BAA2B,CAAC,WAAW,CAAC,CAAC;gBAEtD,IAAI,YAAY,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAChC,kFAAkF;oBAClF,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,UAAU;wBACV,KAAK,EAAE,UAAU,IAAI,+BAA+B,YAAY,CAAC,QAAQ,EAAE;wBAC3E,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,0EAA0E;gBAC1E,MAAM,uBAAuB,CAAC,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;gBAE5E,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,EAAE,cAAc,EAAE,YAAY,EAAE,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE9E,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;oBACd,YAAY;iBACb,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,UAAU;wBACV,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,UAAU;oBACV,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/gemini.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Gemini CLI agent implementation.
+ * Uses direct Google Gemini API access.
+ */
+import type { Agent } from './types.js';
+/**
+ * Create Gemini CLI agent with direct API authentication.
+ */
+export declare function createGeminiAgent(): Agent;
+//# sourceMappingURL=gemini.d.ts.map

package/dist/lib/agents/gemini.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"gemini.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/gemini.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AA6CzE;;GAEG;AACH,wBAAgB,iBAAiB,IAAI,KAAK,CA2MzC"}

package/dist/lib/agents/gemini.js

@@ -0,0 +1,207 @@
+/**
+ * Gemini CLI agent implementation.
+ * Uses direct Google Gemini API access.
+ */
+import { createSandbox, collectLocalFiles, splitTestFiles, verifyNoTestFiles, } from '../sandbox.js';
+import { runValidation, captureGeneratedFiles, createVitestConfig, GEMINI_DIRECT, initGitAndCommit, injectTranscriptContext, } from './shared.js';
+/**
+ * Extract transcript from Gemini stream-json output.
+ * When run with --output-format stream-json, Gemini outputs JSONL (newline-delimited JSON).
+ */
+function extractTranscriptFromOutput(output) {
+    if (!output || !output.trim()) {
+        return undefined;
+    }
+    // The --output-format stream-json output contains JSON events, one per line
+    // Filter to only include lines that look like JSON objects
+    const lines = output.split('\n').filter(line => {
+        const trimmed = line.trim();
+        return trimmed.startsWith('{') && trimmed.endsWith('}');
+    });
+    if (lines.length === 0) {
+        return undefined;
+    }
+    return lines.join('\n');
+}
+/**
+ * Create Gemini CLI agent with direct API authentication.
+ */
+export function createGeminiAgent() {
+    return {
+        name: 'gemini',
+        displayName: 'Gemini CLI',
+        getApiKeyEnvVar() {
+            return GEMINI_DIRECT.apiKeyEnvVar;
+        },
+        getDefaultModel() {
+            return 'gemini-3-pro-preview';
+        },
+        async run(fixturePath, options) {
+            const startTime = Date.now();
+            let sandbox = null;
+            let agentOutput = '';
+            let transcript;
+            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 (abort may have fired during create)
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: '',
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox.sandboxId,
+                    };
+                }
+                // Upload workspace files (excluding tests)
+                await sandbox.uploadFiles(workspaceFiles);
+                await initGitAndCommit(sandbox);
+                // Run setup function if provided
+                if (options.setup) {
+                    await options.setup(sandbox);
+                }
+                // Install dependencies
+                let installResult = await sandbox.runCommand('npm', ['install']);
+                if (installResult.exitCode !== 0) {
+                    installResult = await sandbox.runCommand('npm', ['install']);
+                }
+                if (installResult.exitCode !== 0) {
+                    const output = (installResult.stdout + installResult.stderr).trim().split('\n').slice(-10).join('\n');
+                    throw new Error(`npm install failed (exit code ${installResult.exitCode}):\n${output}`);
+                }
+                // Install Gemini CLI globally
+                const cliInstall = await sandbox.runCommand('npm', [
+                    'install',
+                    '-g',
+                    '@google/gemini-cli',
+                ]);
+                if (cliInstall.exitCode !== 0) {
+                    throw new Error(`Gemini CLI install failed: ${cliInstall.stderr}`);
+                }
+                // Verify no test files in sandbox
+                await verifyNoTestFiles(sandbox);
+                // Run Gemini CLI with direct API access
+                // Using stream-json format for detailed event transcript (similar to Codex's --json)
+                const geminiResult = await sandbox.runCommand('gemini', [
+                    '--prompt',
+                    options.prompt,
+                    '--model',
+                    options.model,
+                    '--approval-mode',
+                    'yolo',
+                    '--output-format',
+                    'stream-json',
+                ], {
+                    env: {
+                        [GEMINI_DIRECT.apiKeyEnvVar]: options.apiKey,
+                    },
+                });
+                agentOutput = geminiResult.stdout + geminiResult.stderr;
+                transcript = extractTranscriptFromOutput(agentOutput);
+                if (geminiResult.exitCode !== 0) {
+                    // Extract meaningful error from output (last few lines usually contain the error)
+                    const errorLines = agentOutput.trim().split('\n').slice(-5).join('\n');
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        transcript,
+                        error: errorLines || `Gemini CLI exited with code ${geminiResult.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);
+                // Inject transcript context so EVAL.ts tests can assert on agent behavior
+                await injectTranscriptContext(sandbox, transcript, 'gemini', options.model);
+                // Run validation scripts
+                const validationResults = await runValidation(sandbox, options.scripts ?? []);
+                // Capture generated files
+                const { generatedFiles, deletedFiles } = await captureGeneratedFiles(sandbox);
+                return {
+                    success: validationResults.allPassed,
+                    output: agentOutput,
+                    transcript,
+                    duration: Date.now() - startTime,
+                    testResult: validationResults.test,
+                    scriptsResults: validationResults.scripts,
+                    sandboxId: sandbox.sandboxId,
+                    generatedFiles,
+                    deletedFiles,
+                };
+            }
+            catch (error) {
+                // Check if this was an abort
+                if (aborted) {
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        transcript,
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox?.sandboxId,
+                    };
+                }
+                return {
+                    success: false,
+                    output: agentOutput,
+                    transcript,
+                    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=gemini.js.map

package/dist/lib/agents/gemini.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gemini.js","sourceRoot":"","sources":["../../../src/lib/agents/gemini.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,aAAa,EACb,gBAAgB,EAChB,uBAAuB,GACxB,MAAM,aAAa,CAAC;AAKrB;;;GAGG;AACH,SAAS,2BAA2B,CAAC,MAAc;IACjD,IAAI,CAAC,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QAC9B,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,4EAA4E;IAC5E,2DAA2D;IAC3D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;QAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,iBAAiB;IAC/B,OAAO;QACL,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,YAAY;QAEzB,eAAe;YACb,OAAO,aAAa,CAAC,YAAY,CAAC;QACpC,CAAC;QAED,eAAe;YACb,OAAO,sBAAsB,CAAC;QAChC,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,UAA8B,CAAC;YACnC,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,8EAA8E;gBAC9E,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,MAAM,gBAAgB,CAAC,OAAO,CAAC,CAAC;gBAEhC,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,IAAI,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACjE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBAC/D,CAAC;gBACD,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,MAAM,GAAG,CAAC,aAAa,CAAC,MAAM,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACtG,MAAM,IAAI,KAAK,CAAC,iCAAiC,aAAa,CAAC,QAAQ,OAAO,MAAM,EAAE,CAAC,CAAC;gBAC1F,CAAC;gBAED,8BAA8B;gBAC9B,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,oBAAoB;iBACrB,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,8BAA8B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACrE,CAAC;gBAED,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,wCAAwC;gBACxC,qFAAqF;gBACrF,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAC3C,QAAQ,EACR;oBACE,UAAU;oBACV,OAAO,CAAC,MAAM;oBACd,SAAS;oBACT,OAAO,CAAC,KAAK;oBACb,iBAAiB;oBACjB,MAAM;oBACN,iBAAiB;oBACjB,aAAa;iBACd,EACD;oBACE,GAAG,EAAE;wBACH,CAAC,aAAa,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;qBAC7C;iBACF,CACF,CAAC;gBAEF,WAAW,GAAG,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC;gBACxD,UAAU,GAAG,2BAA2B,CAAC,WAAW,CAAC,CAAC;gBAEtD,IAAI,YAAY,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAChC,kFAAkF;oBAClF,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,UAAU;wBACV,KAAK,EAAE,UAAU,IAAI,+BAA+B,YAAY,CAAC,QAAQ,EAAE;wBAC3E,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,0EAA0E;gBAC1E,MAAM,uBAAuB,CAAC,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;gBAE5E,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,EAAE,cAAc,EAAE,YAAY,EAAE,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE9E,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;oBACd,YAAY;iBACb,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,UAAU;wBACV,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,UAAU;oBACV,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/index.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Agent registry with built-in agents.
+ */
+import { registerAgent, getAgent, listAgents, hasAgent } from './registry.js';
+export { registerAgent, getAgent, listAgents, hasAgent };
+export type { Agent, AgentRunOptions, AgentRunResult } from './types.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/lib/agents/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAiB9E,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;AAGzD,YAAY,EAAE,KAAK,EAAE,eAAe,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC"}

package/dist/lib/agents/index.js

@@ -0,0 +1,20 @@
+/**
+ * Agent registry with built-in agents.
+ */
+import { registerAgent, getAgent, listAgents, hasAgent } from './registry.js';
+import { createClaudeCodeAgent } from './claude-code.js';
+import { createCodexAgent } from './codex.js';
+import { createOpenCodeAgent } from './opencode.js';
+import { createGeminiAgent } from './gemini.js';
+import { createCursorAgent } from './cursor.js';
+// Register all agent variants (Vercel AI Gateway + Direct API)
+registerAgent(createClaudeCodeAgent({ useVercelAiGateway: true })); // vercel-ai-gateway/claude-code
+registerAgent(createClaudeCodeAgent({ useVercelAiGateway: false })); // claude-code
+registerAgent(createCodexAgent({ useVercelAiGateway: true })); // vercel-ai-gateway/codex
+registerAgent(createCodexAgent({ useVercelAiGateway: false })); // codex
+registerAgent(createOpenCodeAgent()); // opencode
+registerAgent(createGeminiAgent()); // gemini
+registerAgent(createCursorAgent()); // cursor
+// Re-export registry functions
+export { registerAgent, getAgent, listAgents, hasAgent };
+//# sourceMappingURL=index.js.map

package/dist/lib/agents/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/lib/agents/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAC;AAC9E,OAAO,EAAE,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACzD,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAC9C,OAAO,EAAE,mBAAmB,EAAE,MAAM,eAAe,CAAC;AACpD,OAAO,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AAChD,OAAO,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AAEhD,+DAA+D;AAC/D,aAAa,CAAC,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAG,gCAAgC;AACtG,aAAa,CAAC,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAE,cAAc;AACpF,aAAa,CAAC,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAQ,0BAA0B;AAChG,aAAa,CAAC,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,CAAO,QAAQ;AAC9E,aAAa,CAAC,mBAAmB,EAAE,CAAC,CAAC,CAAiC,WAAW;AACjF,aAAa,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAmC,SAAS;AAC/E,aAAa,CAAC,iBAAiB,EAAE,CAAC,CAAC,CAAmC,SAAS;AAE/E,+BAA+B;AAC/B,OAAO,EAAE,aAAa,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC"}

package/dist/lib/agents/opencode.d.ts

@@ -0,0 +1,11 @@
+/**
+ * OpenCode CLI agent implementation.
+ * Uses host-mounted credentials (GitHub Copilot, etc.) for model access.
+ */
+import type { Agent } from './types.js';
+/**
+ * Create OpenCode agent using host-mounted credentials.
+ * Requires Docker sandbox (bind mounts are not supported on Vercel sandbox).
+ */
+export declare function createOpenCodeAgent(): Agent;
+//# sourceMappingURL=opencode.d.ts.map

package/dist/lib/agents/opencode.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"opencode.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/opencode.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAKH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AAsGzE;;;GAGG;AACH,wBAAgB,mBAAmB,IAAI,KAAK,CAqM3C"}