agent-eval-opencode 0.10.0

package/dist/index.d.ts

@@ -0,0 +1,30 @@
+/**
+ * agent-eval
+ *
+ * Framework for testing AI coding agents in isolated sandboxes.
+ */
+export type { AgentType, ModelTier, EvalFilter, Sandbox, SetupFunction, ExperimentConfig, ResolvedExperimentConfig, EvalFixture, EvalRunResult, EvalRunData, EvalSummary, ProgressEvent, ExperimentResults, FailureType, Classification, } from './lib/types.js';
+export { REQUIRED_EVAL_FILES, EXCLUDED_FILES } from './lib/types.js';
+export { CONFIG_DEFAULTS, validateConfig, resolveConfig, loadConfig, resolveEvalNames, } from './lib/config.js';
+export { FixtureValidationError, discoverFixtures, validateFixtureFiles, validatePackageJson, loadFixture, loadAllFixtures, getFixtureFiles, readFixtureFiles, } from './lib/fixture.js';
+export type { SandboxOptions, CommandResult, SandboxFile, SandboxBackend, SandboxBackendInfo, } from './lib/sandbox.js';
+export { SandboxManager, DEFAULT_SANDBOX_TIMEOUT, IGNORED_PATTERNS, TEST_FILE_PATTERNS, collectLocalFiles, splitTestFiles, verifyNoTestFiles, createSandbox, resolveBackend, getSandboxBackendInfo, } from './lib/sandbox.js';
+export type { DockerSandboxOptions } from './lib/docker-sandbox.js';
+export { DockerSandboxManager } from './lib/docker-sandbox.js';
+export type { AgentRunOptions, AgentRunResult } from './lib/agents/types.js';
+export { TRANSCRIPT_CONTEXT_DIR, TRANSCRIPT_CONTEXT_PATH } from './lib/agents/shared.js';
+export type { Agent, ScriptResult } from './lib/agents/types.js';
+export { getAgent, listAgents, registerAgent } from './lib/agents/index.js';
+export type { SaveResultsOptions, ReusableResult } from './lib/results.js';
+export { agentResultToEvalRunData, createEvalSummary, createExperimentResults, saveResults, formatResultsTable, formatRunResult, createProgressDisplay, scanReusableResults, } from './lib/results.js';
+export { computeFingerprint } from './lib/fingerprint.js';
+export { isClassifierEnabled, classifyFailure, classifyWithAI, isNonModelFailure, } from './lib/classifier.js';
+export { housekeep } from './lib/housekeeping.js';
+export type { RunExperimentOptions } from './lib/runner.js';
+export { runExperiment, runSingleEval, StartRateLimiter } from './lib/runner.js';
+export { Dashboard, createConsoleProgressHandler } from './lib/dashboard.js';
+export type { InitOptions } from './lib/init.js';
+export { initProject, getPostInitInstructions } from './lib/init.js';
+export type { ToolName, TranscriptEvent, WebFetchInfo, FileOperationInfo, ShellCommandInfo, TranscriptSummary, Transcript, ParseableAgent, } from './lib/o11y/index.js';
+export { parseTranscript, parseTranscriptSummary, loadTranscript, SUPPORTED_AGENTS, parseClaudeCodeTranscript, parseCodexTranscript, parseOpenCodeTranscript, } from './lib/o11y/index.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,YAAY,EACV,SAAS,EACT,SAAS,EACT,UAAU,EACV,OAAO,EACP,aAAa,EACb,gBAAgB,EAChB,wBAAwB,EACxB,WAAW,EACX,aAAa,EACb,WAAW,EACX,WAAW,EACX,aAAa,EACb,iBAAiB,EACjB,WAAW,EACX,cAAc,GACf,MAAM,gBAAgB,CAAC;AAGxB,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGrE,OAAO,EACL,eAAe,EACf,cAAc,EACd,aAAa,EACb,UAAU,EACV,gBAAgB,GACjB,MAAM,iBAAiB,CAAC;AAGzB,OAAO,EACL,sBAAsB,EACtB,gBAAgB,EAChB,oBAAoB,EACpB,mBAAmB,EACnB,WAAW,EACX,eAAe,EACf,eAAe,EACf,gBAAgB,GACjB,MAAM,kBAAkB,CAAC;AAG1B,YAAY,EACV,cAAc,EACd,aAAa,EACb,WAAW,EACX,cAAc,EACd,kBAAkB,GACnB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,cAAc,EACd,uBAAuB,EACvB,gBAAgB,EAChB,kBAAkB,EAClB,iBAAiB,EACjB,cAAc,EACd,iBAAiB,EACjB,aAAa,EACb,cAAc,EACd,qBAAqB,GACtB,MAAM,kBAAkB,CAAC;AAG1B,YAAY,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AACpE,OAAO,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAG/D,YAAY,EAAE,eAAe,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAG7E,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAGzF,YAAY,EAAE,KAAK,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACjE,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAG5E,YAAY,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AAC3E,OAAO,EACL,wBAAwB,EACxB,iBAAiB,EACjB,uBAAuB,EACvB,WAAW,EACX,kBAAkB,EAClB,eAAe,EACf,qBAAqB,EACrB,mBAAmB,GACpB,MAAM,kBAAkB,CAAC;AAG1B,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAG1D,OAAO,EACL,mBAAmB,EACnB,eAAe,EACf,cAAc,EACd,iBAAiB,GAClB,MAAM,qBAAqB,CAAC;AAG7B,OAAO,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAGlD,YAAY,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AAC5D,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAGjF,OAAO,EAAE,SAAS,EAAE,4BAA4B,EAAE,MAAM,oBAAoB,CAAC;AAG7E,YAAY,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,uBAAuB,EAAE,MAAM,eAAe,CAAC;AAGrE,YAAY,EACV,QAAQ,EACR,eAAe,EACf,YAAY,EACZ,iBAAiB,EACjB,gBAAgB,EAChB,iBAAiB,EACjB,UAAU,EACV,cAAc,GACf,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,eAAe,EACf,sBAAsB,EACtB,cAAc,EACd,gBAAgB,EAChB,yBAAyB,EACzB,oBAAoB,EACpB,uBAAuB,GACxB,MAAM,qBAAqB,CAAC"}

package/dist/index.js

@@ -0,0 +1,29 @@
+/**
+ * agent-eval
+ *
+ * Framework for testing AI coding agents in isolated sandboxes.
+ */
+// Re-export constants
+export { REQUIRED_EVAL_FILES, EXCLUDED_FILES } from './lib/types.js';
+// Re-export config utilities
+export { CONFIG_DEFAULTS, validateConfig, resolveConfig, loadConfig, resolveEvalNames, } from './lib/config.js';
+// Re-export fixture utilities
+export { FixtureValidationError, discoverFixtures, validateFixtureFiles, validatePackageJson, loadFixture, loadAllFixtures, getFixtureFiles, readFixtureFiles, } from './lib/fixture.js';
+export { SandboxManager, DEFAULT_SANDBOX_TIMEOUT, IGNORED_PATTERNS, TEST_FILE_PATTERNS, collectLocalFiles, splitTestFiles, verifyNoTestFiles, createSandbox, resolveBackend, getSandboxBackendInfo, } from './lib/sandbox.js';
+export { DockerSandboxManager } from './lib/docker-sandbox.js';
+// Re-export transcript context constants
+export { TRANSCRIPT_CONTEXT_DIR, TRANSCRIPT_CONTEXT_PATH } from './lib/agents/shared.js';
+export { getAgent, listAgents, registerAgent } from './lib/agents/index.js';
+export { agentResultToEvalRunData, createEvalSummary, createExperimentResults, saveResults, formatResultsTable, formatRunResult, createProgressDisplay, scanReusableResults, } from './lib/results.js';
+// Re-export fingerprinting
+export { computeFingerprint } from './lib/fingerprint.js';
+// Re-export classifier
+export { isClassifierEnabled, classifyFailure, classifyWithAI, isNonModelFailure, } from './lib/classifier.js';
+// Re-export housekeeping
+export { housekeep } from './lib/housekeeping.js';
+export { runExperiment, runSingleEval, StartRateLimiter } from './lib/runner.js';
+// Re-export dashboard utilities
+export { Dashboard, createConsoleProgressHandler } from './lib/dashboard.js';
+export { initProject, getPostInitInstructions } from './lib/init.js';
+export { parseTranscript, parseTranscriptSummary, loadTranscript, SUPPORTED_AGENTS, parseClaudeCodeTranscript, parseCodexTranscript, parseOpenCodeTranscript, } from './lib/o11y/index.js';
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAqBH,sBAAsB;AACtB,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAErE,6BAA6B;AAC7B,OAAO,EACL,eAAe,EACf,cAAc,EACd,aAAa,EACb,UAAU,EACV,gBAAgB,GACjB,MAAM,iBAAiB,CAAC;AAEzB,8BAA8B;AAC9B,OAAO,EACL,sBAAsB,EACtB,gBAAgB,EAChB,oBAAoB,EACpB,mBAAmB,EACnB,WAAW,EACX,eAAe,EACf,eAAe,EACf,gBAAgB,GACjB,MAAM,kBAAkB,CAAC;AAU1B,OAAO,EACL,cAAc,EACd,uBAAuB,EACvB,gBAAgB,EAChB,kBAAkB,EAClB,iBAAiB,EACjB,cAAc,EACd,iBAAiB,EACjB,aAAa,EACb,cAAc,EACd,qBAAqB,GACtB,MAAM,kBAAkB,CAAC;AAI1B,OAAO,EAAE,oBAAoB,EAAE,MAAM,yBAAyB,CAAC;AAK/D,yCAAyC;AACzC,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAIzF,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAI5E,OAAO,EACL,wBAAwB,EACxB,iBAAiB,EACjB,uBAAuB,EACvB,WAAW,EACX,kBAAkB,EAClB,eAAe,EACf,qBAAqB,EACrB,mBAAmB,GACpB,MAAM,kBAAkB,CAAC;AAE1B,2BAA2B;AAC3B,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAE1D,uBAAuB;AACvB,OAAO,EACL,mBAAmB,EACnB,eAAe,EACf,cAAc,EACd,iBAAiB,GAClB,MAAM,qBAAqB,CAAC;AAE7B,yBAAyB;AACzB,OAAO,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AAIlD,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAEjF,gCAAgC;AAChC,OAAO,EAAE,SAAS,EAAE,4BAA4B,EAAE,MAAM,oBAAoB,CAAC;AAI7E,OAAO,EAAE,WAAW,EAAE,uBAAuB,EAAE,MAAM,eAAe,CAAC;AAarE,OAAO,EACL,eAAe,EACf,sBAAsB,EACtB,cAAc,EACd,gBAAgB,EAChB,yBAAyB,EACzB,oBAAoB,EACpB,uBAAuB,GACxB,MAAM,qBAAqB,CAAC"}

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

@@ -0,0 +1,12 @@
+/**
+ * Claude Code agent implementation.
+ * Uses Vercel AI Gateway for model access.
+ */
+import type { Agent } from './types.js';
+/**
+ * Create Claude Code agent with specified authentication method.
+ */
+export declare function createClaudeCodeAgent({ useVercelAiGateway }: {
+    useVercelAiGateway: boolean;
+}): Agent;
+//# sourceMappingURL=claude-code.d.ts.map

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

@@ -0,0 +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;AAqDzE;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,KAAK,CA8NpG"}

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

@@ -0,0 +1,231 @@
+/**
+ * Claude Code agent implementation.
+ * Uses Vercel AI Gateway for model access.
+ */
+import { createSandbox, collectLocalFiles, splitTestFiles, verifyNoTestFiles, } from '../sandbox.js';
+import { runValidation, captureGeneratedFiles, createVitestConfig, AI_GATEWAY, ANTHROPIC_DIRECT, initGitAndCommit, injectTranscriptContext, } from './shared.js';
+/**
+ * Capture the Claude Code transcript from the sandbox.
+ * Claude Code stores transcripts at ~/.claude/projects/-{workdir}/{session-id}.jsonl
+ */
+async function captureTranscript(sandbox) {
+    try {
+        // Get the working directory to construct the transcript path
+        const workdir = sandbox.getWorkingDirectory();
+        // Claude Code uses the path with slashes replaced by dashes
+        const projectPath = workdir.replace(/\//g, '-');
+        const claudeProjectDir = `~/.claude/projects/${projectPath}`;
+        // Find the most recent .jsonl file (the transcript)
+        const findResult = await sandbox.runShell(`ls -t ${claudeProjectDir}/*.jsonl 2>/dev/null | head -1`);
+        if (findResult.exitCode !== 0 || !findResult.stdout.trim()) {
+            return undefined;
+        }
+        const transcriptPath = findResult.stdout.trim();
+        const content = await sandbox.readFile(transcriptPath);
+        return content || undefined;
+    }
+    catch {
+        // Transcript capture is best-effort
+        return undefined;
+    }
+}
+/**
+ * Create Claude Code agent with specified authentication method.
+ */
+export function createClaudeCodeAgent({ useVercelAiGateway }) {
+    return {
+        name: useVercelAiGateway ? 'vercel-ai-gateway/claude-code' : 'claude-code',
+        displayName: useVercelAiGateway ? 'Claude Code (Vercel AI Gateway)' : 'Claude Code',
+        getApiKeyEnvVar() {
+            return useVercelAiGateway ? AI_GATEWAY.apiKeyEnvVar : ANTHROPIC_DIRECT.apiKeyEnvVar;
+        },
+        getDefaultModel() {
+            return 'opus';
+        },
+        async run(fixturePath, options) {
+            const startTime = Date.now();
+            let sandbox = null;
+            let agentOutput = '';
+            let transcript;
+            let aborted = false;
+            let sandboxStopped = false;
+            let hasReturned = false;
+            const captureTranscriptBestEffort = async () => {
+                if (!sandbox || sandboxStopped || transcript)
+                    return;
+                transcript = await captureTranscript(sandbox);
+            };
+            // Handle abort signal
+            const abortHandler = () => {
+                aborted = true;
+                if (sandbox && !sandboxStopped) {
+                    sandboxStopped = true;
+                    sandbox.stop().catch(() => { });
+                }
+            };
+            if (options.signal) {
+                if (options.signal.aborted) {
+                    hasReturned = true;
+                    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) {
+                    hasReturned = true;
+                    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) {
+                    hasReturned = true;
+                    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 Claude Code CLI globally
+                const cliInstall = await sandbox.runCommand('npm', [
+                    'install',
+                    '-g',
+                    '@anthropic-ai/claude-code',
+                ]);
+                if (cliInstall.exitCode !== 0) {
+                    throw new Error(`Claude Code install failed: ${cliInstall.stderr}`);
+                }
+                // Verify no test files in sandbox
+                await verifyNoTestFiles(sandbox);
+                // Run Claude Code with appropriate authentication
+                const claudeResult = await sandbox.runCommand('claude', ['--print', '--model', options.model, '--dangerously-skip-permissions', options.prompt], {
+                    env: useVercelAiGateway
+                        ? {
+                            // AI Gateway configuration for Claude Code
+                            ANTHROPIC_BASE_URL: AI_GATEWAY.baseUrl,
+                            ANTHROPIC_AUTH_TOKEN: options.apiKey,
+                            ANTHROPIC_API_KEY: '',
+                        }
+                        : {
+                            // Direct Anthropic API
+                            ANTHROPIC_API_KEY: options.apiKey,
+                        },
+                });
+                agentOutput = claudeResult.stdout + claudeResult.stderr;
+                if (claudeResult.exitCode !== 0) {
+                    await captureTranscriptBestEffort();
+                    // Extract meaningful error from output (last few lines usually contain the error)
+                    const errorLines = agentOutput.trim().split('\n').slice(-5).join('\n');
+                    hasReturned = true;
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        transcript,
+                        error: errorLines || `Claude Code exited with code ${claudeResult.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);
+                // Capture transcript before validation when available
+                await captureTranscriptBestEffort();
+                // Inject transcript context so EVAL.ts tests can assert on agent behavior
+                await injectTranscriptContext(sandbox, transcript, 'claude-code', options.model);
+                // Run validation scripts
+                const validationResults = await runValidation(sandbox, options.scripts ?? []);
+                // Capture generated files
+                const { generatedFiles, deletedFiles } = await captureGeneratedFiles(sandbox);
+                hasReturned = true;
+                return {
+                    success: validationResults.allPassed,
+                    output: agentOutput,
+                    transcript,
+                    duration: Date.now() - startTime,
+                    testResult: validationResults.test,
+                    scriptsResults: validationResults.scripts,
+                    sandboxId: sandbox.sandboxId,
+                    generatedFiles,
+                    deletedFiles,
+                };
+            }
+            catch (error) {
+                await captureTranscriptBestEffort();
+                // Check if this was an abort
+                if (aborted) {
+                    hasReturned = true;
+                    return {
+                        success: false,
+                        output: agentOutput,
+                        transcript,
+                        error: 'Aborted',
+                        duration: Date.now() - startTime,
+                        sandboxId: sandbox?.sandboxId,
+                    };
+                }
+                hasReturned = true;
+                return {
+                    success: false,
+                    output: agentOutput,
+                    transcript,
+                    error: error instanceof Error ? error.message : String(error),
+                    duration: Date.now() - startTime,
+                    sandboxId: sandbox?.sandboxId,
+                };
+            }
+            finally {
+                // If we're about to return and sandbox is still up, try one final transcript capture.
+                if (hasReturned) {
+                    await captureTranscriptBestEffort();
+                }
+                // Clean up abort listener
+                if (options.signal) {
+                    options.signal.removeEventListener('abort', abortHandler);
+                }
+                if (sandbox && !sandboxStopped) {
+                    sandboxStopped = true;
+                    await sandbox.stop();
+                }
+            }
+        },
+    };
+}
+//# sourceMappingURL=claude-code.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"claude-code.js","sourceRoot":"","sources":["../../../src/lib/agents/claude-code.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,EACV,gBAAgB,EAChB,gBAAgB,EAChB,uBAAuB,GACxB,MAAM,aAAa,CAAC;AAKrB;;;GAGG;AACH,KAAK,UAAU,iBAAiB,CAAC,OAAmB;IAClD,IAAI,CAAC;QACH,6DAA6D;QAC7D,MAAM,OAAO,GAAG,OAAO,CAAC,mBAAmB,EAAE,CAAC;QAC9C,4DAA4D;QAC5D,MAAM,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;QAChD,MAAM,gBAAgB,GAAG,sBAAsB,WAAW,EAAE,CAAC;QAE7D,oDAAoD;QACpD,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CACvC,SAAS,gBAAgB,gCAAgC,CAC1D,CAAC;QAEF,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;YAC3D,OAAO,SAAS,CAAC;QACnB,CAAC;QAED,MAAM,cAAc,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;QACvD,OAAO,OAAO,IAAI,SAAS,CAAC;IAC9B,CAAC;IAAC,MAAM,CAAC;QACP,oCAAoC;QACpC,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,EAAE,kBAAkB,EAAmC;IAC3F,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,+BAA+B,CAAC,CAAC,CAAC,aAAa;QAC1E,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,iCAAiC,CAAC,CAAC,CAAC,aAAa;QAEnF,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,gBAAgB,CAAC,YAAY,CAAC;QACtF,CAAC;QAED,eAAe;YACb,OAAO,MAAM,CAAC;QAChB,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACvD,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;YAC3B,IAAI,WAAW,GAAG,KAAK,CAAC;YAExB,MAAM,2BAA2B,GAAG,KAAK,IAAI,EAAE;gBAC7C,IAAI,CAAC,OAAO,IAAI,cAAc,IAAI,UAAU;oBAAE,OAAO;gBACrD,UAAU,GAAG,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;YAChD,CAAC,CAAC;YAEF,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;oBACzB,WAAW,GAAG,IAAI,CAAC;oBACrB,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,WAAW,GAAG,IAAI,CAAC;oBACnB,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,WAAW,GAAG,IAAI,CAAC;oBACnB,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;gBAE7C,MAAM,gBAAgB,CAAC,OAAO,CAAC,CAAC;gBAE7B,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,mCAAmC;gBACnC,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,2BAA2B;iBAC5B,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,+BAA+B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACtE,CAAC;gBAED,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,kDAAkD;gBAClD,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAC3C,QAAQ,EACR,CAAC,SAAS,EAAE,SAAS,EAAE,OAAO,CAAC,KAAK,EAAE,gCAAgC,EAAE,OAAO,CAAC,MAAM,CAAC,EACvF;oBACE,GAAG,EAAE,kBAAkB;wBACrB,CAAC,CAAC;4BACE,2CAA2C;4BAC3C,kBAAkB,EAAE,UAAU,CAAC,OAAO;4BACtC,oBAAoB,EAAE,OAAO,CAAC,MAAM;4BACpC,iBAAiB,EAAE,EAAE;yBACtB;wBACH,CAAC,CAAC;4BACE,uBAAuB;4BACvB,iBAAiB,EAAE,OAAO,CAAC,MAAM;yBAClC;iBACN,CACF,CAAC;gBAEF,WAAW,GAAG,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC;gBAExD,IAAI,YAAY,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAChC,MAAM,2BAA2B,EAAE,CAAC;oBACpC,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,WAAW,GAAG,IAAI,CAAC;oBACnB,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,UAAU;wBACV,KAAK,EAAE,UAAU,IAAI,gCAAgC,YAAY,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,sDAAsD;gBACtD,MAAM,2BAA2B,EAAE,CAAC;gBAEpC,0EAA0E;gBAC1E,MAAM,uBAAuB,CAAC,OAAO,EAAE,UAAU,EAAE,aAAa,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;gBAEjF,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,WAAW,GAAG,IAAI,CAAC;gBACnB,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,MAAM,2BAA2B,EAAE,CAAC;gBACpC,6BAA6B;gBAC7B,IAAI,OAAO,EAAE,CAAC;oBACZ,WAAW,GAAG,IAAI,CAAC;oBACnB,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,WAAW,GAAG,IAAI,CAAC;gBACnB,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,sFAAsF;gBACtF,IAAI,WAAW,EAAE,CAAC;oBAChB,MAAM,2BAA2B,EAAE,CAAC;gBACtC,CAAC;gBACD,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;AACF,CAAC"}

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

@@ -0,0 +1,12 @@
+/**
+ * OpenAI Codex CLI agent implementation.
+ * Uses Vercel AI Gateway for model access.
+ */
+import type { Agent } from './types.js';
+/**
+ * Create Codex agent with specified authentication method.
+ */
+export declare function createCodexAgent({ useVercelAiGateway }: {
+    useVercelAiGateway: boolean;
+}): Agent;
+//# sourceMappingURL=codex.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"codex.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/codex.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AA8GzE;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,EAAE,kBAAkB,EAAE,EAAE;IAAE,kBAAkB,EAAE,OAAO,CAAA;CAAE,GAAG,KAAK,CA0M/F"}

package/dist/lib/agents/codex.js

@@ -0,0 +1,267 @@
+/**
+ * OpenAI Codex CLI agent implementation.
+ * Uses Vercel AI Gateway for model access.
+ */
+import { createSandbox, collectLocalFiles, splitTestFiles, verifyNoTestFiles, } from '../sandbox.js';
+import { runValidation, captureGeneratedFiles, createVitestConfig, AI_GATEWAY, OPENAI_DIRECT, initGitAndCommit, injectTranscriptContext, } from './shared.js';
+/**
+ * Parse model string with optional query parameters.
+ * e.g. "gpt-5.2-codex?reasoningEffort=high" → { model: "gpt-5.2-codex", reasoningEffort: "high" }
+ */
+function parseModelString(model) {
+    const qIndex = model.indexOf('?');
+    if (qIndex === -1)
+        return { model };
+    const base = model.slice(0, qIndex);
+    const query = model.slice(qIndex + 1);
+    let reasoningEffort;
+    for (const pair of query.split('&')) {
+        const eqIndex = pair.indexOf('=');
+        if (eqIndex === -1)
+            continue;
+        const key = pair.slice(0, eqIndex);
+        const value = decodeURIComponent(pair.slice(eqIndex + 1));
+        if (key === 'reasoningEffort') {
+            reasoningEffort = value;
+        }
+    }
+    return { model: base, reasoningEffort };
+}
+/**
+ * Extract transcript from Codex JSON output.
+ * When run with --json, Codex outputs JSONL to stdout with the full transcript.
+ */
+function extractTranscriptFromOutput(output) {
+    if (!output || !output.trim()) {
+        return undefined;
+    }
+    // The --json output is already the transcript in JSONL format
+    // 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');
+}
+/**
+ * Generate Codex config.toml content.
+ */
+function generateCodexConfig(model, useVercelAiGateway) {
+    if (useVercelAiGateway) {
+        // AI Gateway uses prefixed model names like "openai/gpt-5.2-codex"
+        const fullModel = model.includes('/') ? model : `openai/${model}`;
+        return `# Codex configuration for Vercel AI Gateway
+profile = "default"
+
+[model_providers.vercel]
+name = "Vercel AI Gateway"
+base_url = "${AI_GATEWAY.openAiBaseUrl}"
+env_key = "${AI_GATEWAY.apiKeyEnvVar}"
+wire_api = "responses"
+
+[profiles.default]
+model_provider = "vercel"
+model = "${fullModel}"
+`;
+    }
+    else {
+        // Direct OpenAI API uses unprefixed model names like "gpt-5.2-codex"
+        const directModel = model.includes('/') ? model.split('/').pop() : model;
+        return `# Direct OpenAI API configuration
+profile = "default"
+
+[model_providers.openai]
+name = "OpenAI"
+base_url = "${OPENAI_DIRECT.baseUrl}"
+env_key = "${OPENAI_DIRECT.apiKeyEnvVar}"
+wire_api = "responses"
+
+[profiles.default]
+model_provider = "openai"
+model = "${directModel}"
+`;
+    }
+}
+/**
+ * Create Codex agent with specified authentication method.
+ */
+export function createCodexAgent({ useVercelAiGateway }) {
+    return {
+        name: useVercelAiGateway ? 'vercel-ai-gateway/codex' : 'codex',
+        displayName: useVercelAiGateway ? 'OpenAI Codex (Vercel AI Gateway)' : 'OpenAI Codex',
+        getApiKeyEnvVar() {
+            return useVercelAiGateway ? AI_GATEWAY.apiKeyEnvVar : OPENAI_DIRECT.apiKeyEnvVar;
+        },
+        getDefaultModel() {
+            return 'openai/gpt-5.2-codex';
+        },
+        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 Codex CLI globally
+                const cliInstall = await sandbox.runCommand('npm', [
+                    'install',
+                    '-g',
+                    '@openai/codex',
+                ]);
+                if (cliInstall.exitCode !== 0) {
+                    throw new Error(`Codex CLI install failed: ${cliInstall.stderr}`);
+                }
+                // Parse model string for query parameters (e.g. "gpt-5.2-codex?reasoningEffort=high")
+                const { model: baseModel, reasoningEffort } = parseModelString(options.model);
+                // Create Codex config directory and config file
+                await sandbox.runShell('mkdir -p ~/.codex');
+                const configContent = generateCodexConfig(baseModel, useVercelAiGateway);
+                await sandbox.runShell(`cat > ~/.codex/config.toml << 'EOF'
+${configContent}
+EOF`);
+                // Verify no test files in sandbox
+                await verifyNoTestFiles(sandbox);
+                // Build Codex CLI command
+                // codex login sets up bearer auth for the CLI; env var provides the key for the model provider
+                const envVarToSet = useVercelAiGateway ? AI_GATEWAY.apiKeyEnvVar : OPENAI_DIRECT.apiKeyEnvVar;
+                const escapedPrompt = options.prompt.replace(/'/g, "'\\''");
+                const reasoningFlag = reasoningEffort ? ` -c model_reasoning_effort="${reasoningEffort}"` : '';
+                const codexResult = await sandbox.runShell(`echo '${options.apiKey}' | codex login --with-api-key && codex exec --model ${baseModel} --dangerously-bypass-approvals-and-sandbox --json --skip-git-repo-check${reasoningFlag} '${escapedPrompt}'`, { [envVarToSet]: options.apiKey });
+                agentOutput = codexResult.stdout + codexResult.stderr;
+                transcript = extractTranscriptFromOutput(agentOutput);
+                if (codexResult.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 || `Codex CLI exited with code ${codexResult.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, 'codex', 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=codex.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"codex.js","sourceRoot":"","sources":["../../../src/lib/agents/codex.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,EACV,aAAa,EACb,gBAAgB,EAChB,uBAAuB,GACxB,MAAM,aAAa,CAAC;AAKrB;;;GAGG;AACH,SAAS,gBAAgB,CAAC,KAAa;IACrC,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IAClC,IAAI,MAAM,KAAK,CAAC,CAAC;QAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAEpC,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;IACpC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACtC,IAAI,eAAmC,CAAC;IAExC,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;QACpC,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,OAAO,KAAK,CAAC,CAAC;YAAE,SAAS;QAC7B,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;QACnC,MAAM,KAAK,GAAG,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;QAC1D,IAAI,GAAG,KAAK,iBAAiB,EAAE,CAAC;YAC9B,eAAe,GAAG,KAAK,CAAC;QAC1B,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC;AAC1C,CAAC;AAED;;;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,8DAA8D;IAC9D,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,SAAS,mBAAmB,CAAC,KAAa,EAAE,kBAA2B;IACrE,IAAI,kBAAkB,EAAE,CAAC;QACvB,mEAAmE;QACnE,MAAM,SAAS,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,EAAE,CAAC;QAClE,OAAO;;;;;cAKG,UAAU,CAAC,aAAa;aACzB,UAAU,CAAC,YAAY;;;;;WAKzB,SAAS;CACnB,CAAC;IACA,CAAC;SAAM,CAAC;QACN,qEAAqE;QACrE,MAAM,WAAW,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAG,CAAC,CAAC,CAAC,KAAK,CAAC;QAC1E,OAAO;;;;;cAKG,aAAa,CAAC,OAAO;aACtB,aAAa,CAAC,YAAY;;;;;WAK5B,WAAW;CACrB,CAAC;IACA,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAE,kBAAkB,EAAmC;IACtF,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,OAAO;QAC9D,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,kCAAkC,CAAC,CAAC,CAAC,cAAc;QAErF,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC;QACnF,CAAC;QAED,eAAe;YACb,OAAO,sBAAsB,CAAC;QAChC,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACvD,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;gBAE7C,MAAM,gBAAgB,CAAC,OAAO,CAAC,CAAC;gBAE7B,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,6BAA6B;gBAC7B,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,eAAe;iBAChB,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACpE,CAAC;gBAED,sFAAsF;gBACtF,MAAM,EAAE,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,GAAG,gBAAgB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;gBAE9E,gDAAgD;gBAChD,MAAM,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;gBAC5C,MAAM,aAAa,GAAG,mBAAmB,CAAC,SAAS,EAAE,kBAAkB,CAAC,CAAC;gBACzE,MAAM,OAAO,CAAC,QAAQ,CAAC;EAC3B,aAAa;IACX,CAAC,CAAC;gBAEA,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,0BAA0B;gBAC1B,+FAA+F;gBAC/F,MAAM,WAAW,GAAG,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC;gBAC9F,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;gBAC5D,MAAM,aAAa,GAAG,eAAe,CAAC,CAAC,CAAC,+BAA+B,eAAe,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC/F,MAAM,WAAW,GAAG,MAAM,OAAO,CAAC,QAAQ,CACxC,SAAS,OAAO,CAAC,MAAM,wDAAwD,SAAS,2EAA2E,aAAa,KAAK,aAAa,GAAG,EACrM,EAAE,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,MAAM,EAAE,CAClC,CAAC;gBAEF,WAAW,GAAG,WAAW,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC;gBACtD,UAAU,GAAG,2BAA2B,CAAC,WAAW,CAAC,CAAC;gBAEtD,IAAI,WAAW,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC/B,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,8BAA8B,WAAW,CAAC,QAAQ,EAAE;wBACzE,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,OAAO,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;gBAE3E,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;AACF,CAAC"}

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

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

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

@@ -0,0 +1 @@
+{"version":3,"file":"cursor.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/cursor.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAmC,MAAM,YAAY,CAAC;AA6CzE;;GAEG;AACH,wBAAgB,iBAAiB,IAAI,KAAK,CA0MzC"}