@aws/ml-container-creator 0.2.6 → 0.4.0

package/src/app.js

@@ -119,6 +119,23 @@ export async function run(projectName, options) {
     let answers;
     if (configManager.shouldSkipPrompts()) {
         console.log('\n🚀 Skipping prompts - using configuration from other sources');
+
+        // Fail-fast if required parameters are missing
+        const missing = configManager.getMissingRequiredParameters();
+        if (missing.length > 0) {
+            console.error('\n❌ Cannot skip prompts — required parameters are missing:\n');
+            for (const param of missing) {
+                const matrix = configManager._getParameterMatrix()[param];
+                const cliFlag = matrix?.cliOption ? `--${matrix.cliOption}` : '';
+                const envVar = matrix?.envVar || '';
+                const hints = [cliFlag, envVar].filter(Boolean).join(' or ');
+                console.error(`   • ${param}${hints ? ` (${hints})` : ''}`);
+            }
+            console.error('\n   Provide these via CLI flags, environment variables, or a config file.');
+            console.error('   Run "ml-container-creator --help" for available options.\n');
+            process.exit(1);
+        }
+
         answers = configManager.getFinalConfiguration();
 
         // Infer modelSource from model name prefix if not set
@@ -305,6 +322,17 @@ export async function writeProject(templateDir, destDir, answers, registryConfig
         ignorePatterns.push('**/sample_model/**');
     }
 
+    // Exclude do/benchmark when benchmarking is not selected
+    if (!answers.includeBenchmark) {
+        ignorePatterns.push('**/do/benchmark');
+    }
+
+    // Exclude do/test when hosted-model-endpoint is not selected
+    const testTypes = answers.testTypes || [];
+    if (!testTypes.includes('hosted-model-endpoint')) {
+        ignorePatterns.push('**/do/test');
+    }
+
     // Always exclude triton and diffusors source directories
     ignorePatterns.push('**/triton/**');
     ignorePatterns.push('**/diffusors/**');
@@ -400,6 +428,20 @@ export async function writeProject(templateDir, destDir, answers, registryConfig
     _copyFile(path.join(LIB_DIR, 'manifest-cli.js'), path.join(doLibDir, 'manifest-cli.js'));
     _copyFile(path.join(LIB_DIR, 'asset-manager.js'), path.join(doLibDir, 'asset-manager.js'));
     _copyFile(path.join(LIB_DIR, 'bootstrap-config.js'), path.join(doLibDir, 'bootstrap-config.js'));
+
+    // Generate .gitignore with benchmarks/ when benchmarking is enabled
+    if (answers.includeBenchmark) {
+        const gitignorePath = path.join(destDir, '.gitignore');
+        const gitignoreContent = '# Benchmark results (generated by do/benchmark)\nbenchmarks/\n';
+        if (fs.existsSync(gitignorePath)) {
+            const existing = fs.readFileSync(gitignorePath, 'utf8');
+            if (!existing.includes('benchmarks/')) {
+                fs.appendFileSync(gitignorePath, `\n${gitignoreContent}`);
+            }
+        } else {
+            fs.writeFileSync(gitignorePath, gitignoreContent);
+        }
+    }
 }
 
 /**
@@ -493,7 +535,9 @@ async function _ensureTemplateVariables(answers, registryConfigManager = null) {
         chatTemplate: null,
         chatTemplateSource: null,
         hfToken: null,
+        hfTokenArn: null,
         ngcApiKey: null,
+        ngcTokenArn: null,
         envVars: {},
         inferenceAmiVersion: null,
         accelerator: null,
@@ -510,7 +554,7 @@ async function _ensureTemplateVariables(answers, registryConfigManager = null) {
         codebuildProjectName: null,
         modelName: null,
         modelFormat: null,
-        includeSampleModel: false,
+        includeSampleModel: true,
         includeTesting: true,
         testTypes: [],
         buildTimestamp: new Date().toISOString(),
@@ -622,6 +666,55 @@ async function _ensureTemplateVariables(answers, registryConfigManager = null) {
             // Silently continue - defaults are already set
         }
     }
+
+    // Populate baseImage from the catalog when still falsy (covers --skip-prompts and
+    // cases where MCP/CLI/config did not provide a base image).
+    // Precedence: MCP > CLI > config > catalog default (this block).
+    if (!answers.baseImage && registryConfigManager?.frameworkRegistry) {
+        const backendKey = answers.backend || answers.modelServer;
+        if (backendKey) {
+            const frameworkVersions = registryConfigManager.frameworkRegistry[backendKey];
+            if (frameworkVersions) {
+                let resolvedConfig = null;
+                if (answers.frameworkVersion && frameworkVersions[answers.frameworkVersion]) {
+                    resolvedConfig = frameworkVersions[answers.frameworkVersion];
+                } else {
+                    // Fall back to latest version
+                    const versions = Object.keys(frameworkVersions).sort((a, b) =>
+                        b.localeCompare(a, undefined, { numeric: true })
+                    );
+                    if (versions.length > 0) {
+                        resolvedConfig = frameworkVersions[versions[0]];
+                    }
+                }
+                if (resolvedConfig?.baseImage) {
+                    answers.baseImage = resolvedConfig.baseImage;
+                }
+            }
+        }
+    }
+
+    // Populate icGpuCount from instance catalog when not explicitly set.
+    // The deploy template uses IC_GPU_COUNT unconditionally for NumberOfAcceleratorDevicesRequired,
+    // so it must always have a value for GPU deployments.
+    if (answers.icGpuCount == null && answers.instanceType) {
+        // Use gpuCount from instance-sizer recommendation if available
+        if (answers.gpuCount) {
+            answers.icGpuCount = answers.gpuCount;
+        } else {
+            // Look up from instances catalog
+            try {
+                const catalogPath = path.resolve(__dirname, '..', 'servers', 'lib', 'catalogs', 'instances.json');
+                const catalogData = JSON.parse(fs.readFileSync(catalogPath, 'utf-8'));
+                const instanceInfo = catalogData?.catalog?.[answers.instanceType];
+                if (instanceInfo?.gpus && instanceInfo.gpus > 0) {
+                    answers.icGpuCount = instanceInfo.gpus;
+                }
+            } catch {
+                // Silently continue — template fallback handles missing value
+            }
+        }
+    }
 }
 
 /**
@@ -958,7 +1051,8 @@ function _setExecutablePermissions(destDir) {
         'do/submit',
         'do/register',
         'do/ci',
-        'do/manifest'
+        'do/manifest',
+        'do/benchmark'
     ];
 
     shellScripts.forEach(script => {

package/src/lib/architecture-sync.js

@@ -0,0 +1,171 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+/**
+ * Architecture Sync
+ *
+ * Fetches model registry source files from server GitHub repositories
+ * and extracts supported model_type values into the model-servers catalog.
+ */
+
+import { readFileSync, writeFileSync } from 'node:fs';
+
+/**
+ * Parse vLLM's model registry Python source to extract model_type keys.
+ *
+ * vLLM's registry maps architecture class names to (module, impl_class) tuples:
+ *   "LlamaForCausalLM": ("llama", "LlamaForCausalLM"),
+ *   "Qwen2ForCausalLM": ("qwen2", "Qwen2ForCausalLM"),
+ *
+ * The module name (first tuple element) corresponds to the model_type.
+ * Also matches older formats where model_type is used directly as dict key.
+ *
+ * @param {string} source - Python source code content
+ * @returns {string[]} Sorted array of model_type strings
+ */
+export const parseVllmRegistry = (source) => {
+    const modelTypes = new Set();
+    const patterns = [
+        // Tuple value format: ("module_name", "ClassName") — extract module_name
+        /\("([a-z][a-z0-9_]*)"\s*,\s*"[A-Z]/g,
+        // Direct lowercase key format (older registries): "model_type": (
+        /"([a-z][a-z0-9_]*)":\s*\(/g,
+        // Direct lowercase key format: "model_type": ClassName
+        /"([a-z][a-z0-9_]*)":\s*[A-Z]/g,
+        // Direct lowercase key format: "model_type": [
+        /"([a-z][a-z0-9_]*)":\s*\[/g
+    ];
+    for (const pattern of patterns) {
+        let match;
+        while ((match = pattern.exec(source)) !== null) {
+            modelTypes.add(match[1]);
+        }
+    }
+    return [...modelTypes].sort();
+};
+
+/**
+ * Parse SGLang's model_registry.py to extract model_type keys.
+ *
+ * Matches patterns like:
+ *   "llama": ModelClass,
+ *   "qwen2": (ModulePath, ClassName),
+ *
+ * @param {string} source - Python source code content
+ * @returns {string[]} Sorted array of model_type strings
+ */
+export const parseSglangRegistry = (source) => {
+    const modelTypes = new Set();
+    const patterns = [
+        /"([a-z][a-z0-9_]*)":\s*\(/g,
+        /"([a-z][a-z0-9_]*)":\s*[A-Z]/g,
+        /"([a-z][a-z0-9_]*)":\s*\[/g
+    ];
+    for (const pattern of patterns) {
+        let match;
+        while ((match = pattern.exec(source)) !== null) {
+            modelTypes.add(match[1]);
+        }
+    }
+    return [...modelTypes].sort();
+};
+
+/**
+ * Parse TensorRT-LLM's models __init__.py to extract model_type keys.
+ *
+ * Matches patterns from MODEL_MAP or similar dict structures:
+ *   "llama": LlamaForCausalLM,
+ *   "gpt2": GPT2LMHeadModel,
+ *
+ * @param {string} source - Python source code content
+ * @returns {string[]} Sorted array of model_type strings
+ */
+export const parseTensorRTRegistry = (source) => {
+    const modelTypes = new Set();
+    const patterns = [
+        /"([a-z][a-z0-9_]*)":\s*[A-Z]/g,
+        /"([a-z][a-z0-9_]*)":\s*\(/g,
+        /'([a-z][a-z0-9_]*)':\s*[A-Z]/g,
+        /'([a-z][a-z0-9_]*)':\s*\(/g
+    ];
+    for (const pattern of patterns) {
+        let match;
+        while ((match = pattern.exec(source)) !== null) {
+            modelTypes.add(match[1]);
+        }
+    }
+    return [...modelTypes].sort();
+};
+
+/**
+ * Configuration mapping each server to its GitHub repository,
+ * registry file path, tag prefix, and parser function.
+ */
+export const SERVER_REGISTRY_SOURCES = {
+    vllm: {
+        repo: 'vllm-project/vllm',
+        file: 'vllm/model_executor/models/registry.py',
+        tagPrefix: 'v',
+        parser: parseVllmRegistry
+    },
+    sglang: {
+        repo: 'sgl-project/sglang',
+        file: 'python/sglang/srt/models/model_registry.py',
+        tagPrefix: 'v',
+        parser: parseSglangRegistry
+    },
+    'tensorrt-llm': {
+        repo: 'NVIDIA/TensorRT-LLM',
+        file: 'tensorrt_llm/models/__init__.py',
+        tagPrefix: 'v',
+        parser: parseTensorRTRegistry
+    }
+};
+
+/**
+ * Sync supported model architectures from server GitHub repositories
+ * into the model-servers catalog.
+ *
+ * For each server entry in the catalog that has a matching source config,
+ * fetches the model registry file from GitHub at the version tag and
+ * parses it to extract supported model_type values.
+ *
+ * @param {string} catalogPath - Path to model-servers.json
+ * @returns {object} Summary with counts and failures
+ */
+export const syncArchitectures = async (catalogPath) => {
+    const catalog = JSON.parse(readFileSync(catalogPath, 'utf8'));
+    const summary = { servers: [], failures: [] };
+
+    for (const [server, entries] of Object.entries(catalog)) {
+        const source = SERVER_REGISTRY_SOURCES[server];
+        if (!source) continue;
+
+        for (const entry of entries) {
+            const version = entry.labels?.framework_version;
+            if (!version) continue;
+
+            const tag = `${source.tagPrefix}${version}`;
+            const url = `https://raw.githubusercontent.com/${source.repo}/${tag}/${source.file}`;
+
+            try {
+                const response = await fetch(url);
+                if (!response.ok) {
+                    summary.failures.push({ server, version, reason: `HTTP ${response.status}` });
+                    console.log(`   ⚠️  ${server} ${version}: fetch failed (HTTP ${response.status})`);
+                    continue;
+                }
+                const content = await response.text();
+                entry.supportedModelTypes = source.parser(content);
+                summary.servers.push({ server, version, count: entry.supportedModelTypes.length });
+                console.log(`   ✓ ${server} ${version}: ${entry.supportedModelTypes.length} architectures`);
+            } catch (err) {
+                summary.failures.push({ server, version, reason: err.message });
+                console.log(`   ⚠️  ${server} ${version}: fetch failed (${err.message})`);
+            }
+        }
+    }
+
+    writeFileSync(catalogPath, JSON.stringify(catalog, null, 4));
+    return summary;
+};

package/src/lib/arn-detection.js

@@ -0,0 +1,22 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+/**
+ * ARN Detection Utility
+ *
+ * Provides a pure function for distinguishing AWS Secrets Manager ARNs
+ * from plaintext values. Used by the prompt flow and CLI to determine
+ * whether user input should be treated as a secret reference or a
+ * literal token value.
+ */
+
+const SECRETS_MANAGER_ARN_PREFIX = 'arn:aws:secretsmanager:';
+
+/**
+ * Determines if a value is a Secrets Manager ARN.
+ * @param {*} value - The input value to check
+ * @returns {boolean} True if the value is a Secrets Manager ARN
+ */
+export function isSecretsManagerArn(value) {
+    return typeof value === 'string' && value.startsWith(SECRETS_MANAGER_ARN_PREFIX);
+}

package/src/lib/bootstrap-command-handler.js

@@ -24,6 +24,8 @@ import { fileURLToPath } from 'node:url';
 import BootstrapConfig from './bootstrap-config.js';
 import AwsProfileParser from './aws-profile-parser.js';
 import AssetManager from './asset-manager.js';
+import McpCommandHandler from './mcp-command-handler.js';
+import RegistryCommandHandler from './registry-command-handler.js';
 import { runPrompts } from '../prompt-adapter.js';
 
 const __filename = fileURLToPath(import.meta.url);
@@ -114,6 +116,11 @@ export default class BootstrapCommandHandler {
 
         console.log('\n🚀 Bootstrap — Shared AWS Infrastructure Setup\n');
 
+        // Verify AWS CLI v2 is installed
+        if (!this._verifyCliV2()) {
+            return;
+        }
+
         // Determine bootstrap profile name
         let profileName;
         if (nonInteractive) {
@@ -192,6 +199,9 @@ export default class BootstrapCommandHandler {
             if (stackOutputs.BatchS3BucketName) {
                 profileData.batchS3Bucket = stackOutputs.BatchS3BucketName;
             }
+            if (stackOutputs.BenchmarkS3BucketName) {
+                profileData.benchmarkS3Bucket = stackOutputs.BenchmarkS3BucketName;
+            }
 
             console.log('  ✅ Bootstrap stack deployed successfully');
         } catch (error) {
@@ -311,6 +321,9 @@ export default class BootstrapCommandHandler {
 
         // Display summary
         this._displaySummary(profileName, profileData);
+
+        // Step 6: Post-setup chain (mcp init → sync-architectures → sync-schemas)
+        await this._runPostSetupChain(options);
     }
 
     /**
@@ -377,6 +390,9 @@ export default class BootstrapCommandHandler {
                 if (outputs.BatchS3BucketName) {
                     console.log(`  ✅ S3 bucket (batch): ${outputs.BatchS3BucketName}`);
                 }
+                if (outputs.BenchmarkS3BucketName) {
+                    console.log(`  ✅ S3 bucket (benchmark): ${outputs.BenchmarkS3BucketName}`);
+                }
                 if (outputs.StackVersion) {
                     console.log(`  📋 Stack version: ${outputs.StackVersion}`);
                 }
@@ -447,6 +463,20 @@ export default class BootstrapCommandHandler {
                     console.log(`  ⚠️  S3 bucket: ${profile.config.batchS3Bucket} — could not validate`);
                 }
             }
+
+            if (profile.config.benchmarkS3Bucket) {
+                try {
+                    const benchmarkExists = this._resourceExists(
+                        `s3api head-bucket --bucket ${profile.config.benchmarkS3Bucket}`,
+                        profile.config.awsProfile
+                    );
+                    console.log(benchmarkExists
+                        ? `  ✅ S3 bucket (benchmark): ${profile.config.benchmarkS3Bucket}`
+                        : `  ⚠️  S3 bucket (benchmark): ${profile.config.benchmarkS3Bucket} — missing`);
+                } catch {
+                    console.log(`  ⚠️  S3 bucket (benchmark): ${profile.config.benchmarkS3Bucket} — could not validate`);
+                }
+            }
         }
 
         // Display deployed resources from manifest
@@ -1000,6 +1030,7 @@ export default class BootstrapCommandHandler {
             if (stackOutputs.EcrRepositoryName) profileConfig.ecrRepositoryName = stackOutputs.EcrRepositoryName;
             if (stackOutputs.AsyncS3BucketName) profileConfig.asyncS3Bucket = stackOutputs.AsyncS3BucketName;
             if (stackOutputs.BatchS3BucketName) profileConfig.batchS3Bucket = stackOutputs.BatchS3BucketName;
+            if (stackOutputs.BenchmarkS3BucketName) profileConfig.benchmarkS3Bucket = stackOutputs.BenchmarkS3BucketName;
             profileConfig.stackName = stackName;
 
             console.log('  ✅ Bootstrap stack updated');
@@ -1054,6 +1085,74 @@ export default class BootstrapCommandHandler {
         // Save updated profile
         this.config.setProfile(name, profileConfig);
         console.log(`\n✅ Update complete for profile "${name}"`);
+
+        // Re-run post-setup chain after updating AWS resources
+        await this._runPostSetupChain(options);
+    }
+
+    /**
+     * Run the post-setup chain: mcp init → registry sync-architectures → sync-schemas.
+     * Each step is independent — failures are collected and reported at the end.
+     *
+     * @param {object} options - Parsed CLI options (checks skipPostSetup)
+     */
+    async _runPostSetupChain(options = {}) {
+        if (options['skip-post-setup']) {
+            console.log('\n⏭️  Skipping post-setup chain (--skip-post-setup)');
+            return;
+        }
+
+        console.log('\n🔗 Running post-setup configuration...\n');
+
+        const failures = [];
+
+        // 1. MCP init — register bundled MCP servers
+        console.log('📡 Registering MCP servers...');
+        try {
+            const generatorAdapter = {
+                destinationPath(...segments) {
+                    return path.resolve(process.cwd(), ...segments);
+                }
+            };
+            const mcpHandler = new McpCommandHandler(generatorAdapter);
+            await mcpHandler.handle(['init'], {});
+        } catch (error) {
+            failures.push({ step: 'mcp init', error: error.message });
+            console.log(`  ⚠️  mcp init failed: ${error.message}`);
+        }
+
+        // 2. Registry sync-architectures — populate supportedModelTypes
+        console.log('\n📋 Syncing model architecture registry...');
+        try {
+            const registryHandler = new RegistryCommandHandler();
+            await registryHandler.handle(['sync-architectures'], {});
+        } catch (error) {
+            failures.push({ step: 'registry sync-architectures', error: error.message });
+            console.log(`  ⚠️  registry sync-architectures failed: ${error.message}`);
+        }
+
+        // 3. Schema sync — download AWS service models
+        console.log('\n📐 Syncing service schemas...');
+        try {
+            await this._handleSyncSchemas();
+        } catch (error) {
+            failures.push({ step: 'sync-schemas', error: error.message });
+            console.log(`  ⚠️  sync-schemas failed: ${error.message}`);
+        }
+
+        // Report results
+        if (failures.length === 0) {
+            console.log('\n✅ Bootstrap complete — all systems operational');
+        } else {
+            console.log(`\n⚠️  Bootstrap complete with ${failures.length} warning${failures.length === 1 ? '' : 's'}:`);
+            for (const { step, error } of failures) {
+                console.log(`  • ${step}: ${error}`);
+            }
+            console.log('\n  These steps can be re-run individually:');
+            console.log('    ml-container-creator mcp init');
+            console.log('    ml-container-creator registry sync-architectures');
+            console.log('    ml-container-creator bootstrap sync-schemas');
+        }
     }
 
     /**
@@ -1205,11 +1304,28 @@ export default class BootstrapCommandHandler {
                         'sagemaker:DescribeEndpointConfig',
                         'sagemaker:DescribeModel',
                         'sagemaker:DescribeInferenceComponent',
+                        'sagemaker:ListInferenceComponents',
                         'sagemaker:InvokeEndpoint',
                         'sagemaker:InvokeEndpointAsync'
                     ],
                     Resource: '*'
                 },
+                {
+                    Sid: 'SageMakerBenchmarking',
+                    Effect: 'Allow',
+                    Action: [
+                        'sagemaker:CreateAIBenchmarkJob',
+                        'sagemaker:DescribeAIBenchmarkJob',
+                        'sagemaker:ListAIBenchmarkJobs',
+                        'sagemaker:StopAIBenchmarkJob',
+                        'sagemaker:DeleteAIBenchmarkJob',
+                        'sagemaker:CreateAIWorkloadConfig',
+                        'sagemaker:DescribeAIWorkloadConfig',
+                        'sagemaker:ListAIWorkloadConfigs',
+                        'sagemaker:DeleteAIWorkloadConfig'
+                    ],
+                    Resource: '*'
+                },
                 {
                     Sid: 'ECRPull',
                     Effect: 'Allow',
@@ -1242,12 +1358,43 @@ export default class BootstrapCommandHandler {
                     Effect: 'Allow',
                     Action: [
                         's3:GetObject',
+                        's3:PutObject',
+                        's3:AbortMultipartUpload',
                         's3:ListBucket'
                     ],
                     Resource: [
                         'arn:aws:s3:::ml-container-creator-*',
                         'arn:aws:s3:::ml-container-creator-*/*'
                     ]
+                },
+                {
+                    Sid: 'SNSPublish',
+                    Effect: 'Allow',
+                    Action: 'sns:Publish',
+                    Resource: 'arn:aws:sns:*:*:ml-container-creator-*'
+                },
+                {
+                    Sid: 'SecretsManagerBenchmark',
+                    Effect: 'Allow',
+                    Action: [
+                        'secretsmanager:CreateSecret',
+                        'secretsmanager:PutSecretValue',
+                        'secretsmanager:GetSecretValue',
+                        'secretsmanager:DescribeSecret'
+                    ],
+                    Resource: 'arn:aws:secretsmanager:*:*:secret:ml-container-creator/*'
+                },
+                {
+                    Sid: 'QuotaAndAvailability',
+                    Effect: 'Allow',
+                    Action: [
+                        'service-quotas:GetServiceQuota',
+                        'service-quotas:ListServiceQuotas',
+                        'sagemaker:ListTrainingPlans',
+                        'sagemaker:DescribeTrainingPlan',
+                        'sagemaker:ListEndpoints'
+                    ],
+                    Resource: '*'
                 }
             ]
         };
@@ -1397,9 +1544,15 @@ export default class BootstrapCommandHandler {
 
     /**
      * Optionally create S3 buckets for async/batch deployments.
+     * Always creates the benchmark S3 bucket (unconditional).
      * @returns {Promise<object|null>} Bucket names or null if skipped
      */
     async _setupS3Buckets() {
+        // Always create benchmark bucket (unconditional — avoids re-bootstrap when benchmarking is enabled later)
+        const benchmarkBucketName = `ml-container-creator-benchmark-${this._currentRegion}-${this._currentAccountId}`;
+        const tags = this._buildResourceTags();
+        const benchmarkS3Bucket = await this._createS3Bucket(benchmarkBucketName, tags);
+
         const { useS3 } = await this._promptFn([{
             type: 'confirm',
             name: 'useS3',
@@ -1408,17 +1561,16 @@ export default class BootstrapCommandHandler {
         }]);
 
         if (!useS3) {
-            return null;
+            return { benchmarkS3Bucket };
         }
 
         const asyncBucketName = `ml-container-creator-async-${this._currentRegion}-${this._currentAccountId}`;
         const batchBucketName = `ml-container-creator-batch-${this._currentRegion}-${this._currentAccountId}`;
 
-        const tags = this._buildResourceTags();
         const asyncS3Bucket = await this._createS3Bucket(asyncBucketName, tags);
         const batchS3Bucket = await this._createS3Bucket(batchBucketName, tags);
 
-        return { asyncS3Bucket, batchS3Bucket };
+        return { asyncS3Bucket, batchS3Bucket, benchmarkS3Bucket };
     }
 
     /**
@@ -1475,6 +1627,28 @@ export default class BootstrapCommandHandler {
 
     // ── AWS CLI helpers ─────────────────────────────────────────────
 
+    /**
+     * Verify AWS CLI v2 is installed. Returns true if v2 is detected, false otherwise.
+     * Extracted as a method so tests can override it.
+     * @returns {boolean}
+     */
+    _verifyCliV2() {
+        try {
+            const versionOutput = execSync('aws --version', { encoding: 'utf8', stdio: ['pipe', 'pipe', 'pipe'] }).trim();
+            if (!versionOutput.includes('aws-cli/2')) {
+                console.log(`  ❌ AWS CLI v2 is required. Detected: ${versionOutput.split(' ')[0]}`);
+                console.log('  Install: https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html');
+                console.log('  Some features (benchmarking, newer SageMaker APIs) require CLI v2.\n');
+                return false;
+            }
+            return true;
+        } catch {
+            console.log('  ❌ AWS CLI not found.');
+            console.log('  Install: https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html\n');
+            return false;
+        }
+    }
+
     /**
      * Execute an AWS CLI command and return parsed JSON output.
      * @param {string} command - AWS CLI command (without 'aws' prefix)
@@ -1649,6 +1823,7 @@ SETUP OPTIONS:
   --skip-s3                           Skip S3 bucket creation
   --ci                                Provision CI testing infrastructure
   --skip-ci                           Skip CI infrastructure provisioning
+  --skip-post-setup                   Skip post-setup chain (mcp init, sync-architectures, sync-schemas)
 
 STATUS OPTIONS:
   --verify                            Check each active resource against AWS APIs for drift detection

package/src/lib/cli-handler.js

@@ -375,7 +375,7 @@ For more information, visit: https://github.com/awslabs/ml-container-creator
                 type: 'confirm',
                 name: 'includeSampleModel',
                 message: 'Include sample model:',
-                default: false,
+                default: true,
                 when: answers => answers.framework !== 'transformers'
             },
             {
@@ -515,7 +515,7 @@ ml-container-creator \\
             'framework': 'sklearn',
             'modelServer': 'flask',
             'modelFormat': 'pkl',
-            'includeSampleModel': false,
+            'includeSampleModel': true,
             'includeTesting': true,
             'testTypes': ['local-model-cli', 'local-model-server', 'hosted-model-endpoint'],
             'buildTarget': 'codebuild',