@hyperdrive.bot/bmad-workflow 1.0.10 → 1.0.12

package/dist/commands/decompose.js

@@ -203,7 +203,7 @@ export default class Decompose extends Command {
                     }
                 }
                 this.log(colors.bold('\n⚙️  Phase 3: Executing task graph...\n'));
-                const executor = new DependencyGraphExecutor(taskGraph, createAgentRunner(flags.provider, this.logger), new BatchProcessor(flags['max-parallel'], 0, this.logger), this.fileManager, this.logger, flags.cwd);
+                const executor = new DependencyGraphExecutor(taskGraph, createAgentRunner(flags.provider, this.logger), new BatchProcessor(flags['max-parallel'], 0, this.logger), this.fileManager, this.logger, flags.cwd, flags.model);
                 const executionResult = await executor.execute((layerIndex, totalLayers, layerSize) => {
                     this.log(colors.info(`\n🔄 Starting Layer ${layerIndex + 1}/${totalLayers} (${layerSize} task${layerSize > 1 ? 's' : ''} in parallel)`));
                 }, (taskId, _layerIndex, _taskIndex, _totalTasks) => {

package/dist/commands/epics/create.d.ts

@@ -40,6 +40,9 @@ export default class EpicsCreate extends Command {
         model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
         provider: import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
         task: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
+        timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'max-retries': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'retry-backoff': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
     };
     /**
      * AI agent runner service (Claude or Gemini)

package/dist/commands/epics/create.js

@@ -22,6 +22,7 @@ import * as colors from '../../utils/colors.js';
 import { ValidationError } from '../../utils/errors.js';
 import { createLogger, generateCorrelationId } from '../../utils/logger.js';
 import { createSpinner } from '../../utils/progress.js';
+import { runAgentWithRetry } from '../../utils/retry.js';
 import { agentFlags } from '../../utils/shared-flags.js';
 /**
  * Epics Create Command
@@ -136,10 +137,13 @@ export default class EpicsCreate extends Command {
                 epicDir,
                 epics: toCreate,
                 interval: flags.interval,
+                maxRetries: flags['max-retries'],
                 prdPath: args['prd-path'],
                 prefix: flags.prefix || '',
                 references: flags.reference,
+                retryBackoff: flags['retry-backoff'],
                 task: flags.task,
+                timeout: flags.timeout,
             });
             // Display summary
             this.displaySummary({
@@ -208,7 +212,7 @@ export default class EpicsCreate extends Command {
      * Create epic files using Claude AI agents
      */
     async createEpics(options) {
-        const { agent, epicDir, epics, interval, prdPath, prefix, references, task } = options;
+        const { agent, epicDir, epics, interval, maxRetries, prdPath, prefix, references, retryBackoff, task, timeout } = options;
         const results = [];
         /* eslint-disable no-await-in-loop */
         for (let i = 0; i < epics.length; i++) {
@@ -217,10 +221,13 @@ export default class EpicsCreate extends Command {
                 agent,
                 epic,
                 epicDir,
+                maxRetries,
                 prdPath,
                 prefix,
                 references,
+                retryBackoff,
                 task,
+                timeout,
             });
             results.push(result);
             // Wait interval if not last epic
@@ -235,7 +242,7 @@ export default class EpicsCreate extends Command {
      * Create a single epic file
      */
     async createSingleEpic(options) {
-        const { agent, epic, epicDir, prdPath, prefix, references, task } = options;
+        const { agent, epic, epicDir, maxRetries, prdPath, prefix, references, retryBackoff, task, timeout } = options;
         const fileName = this.generateEpicFileName(epic, prefix);
         const filePath = path.join(epicDir, fileName);
         const spinner = createSpinner(`Creating Epic ${epic.number}: ${epic.title}...`);
@@ -262,10 +269,14 @@ export default class EpicsCreate extends Command {
             });
             // Step 3: Run Claude AI agent to populate epic content sections
             spinner.text = `Populating epic ${epic.number} with AI agent...`;
-            const result = await this.agentRunner.runAgent(prompt, {
+            const result = await runAgentWithRetry(this.agentRunner, prompt, {
                 agentType: 'sm',
                 references,
-                timeout: 1_800_000, // 30 minutes
+                timeout: timeout ?? 2_700_000,
+            }, {
+                backoffMs: retryBackoff,
+                logger: this.logger,
+                maxRetries,
             });
             if (!result.success) {
                 throw new Error(result.errors || 'Claude agent failed to populate epic');

package/dist/commands/stories/create.d.ts

@@ -38,6 +38,9 @@ export default class StoriesCreateCommand extends Command {
         model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
         provider: import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
         task: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
+        timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'max-retries': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'retry-backoff': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
     };
     private agentRunner;
     private batchProcessor;

package/dist/commands/stories/create.js

@@ -24,6 +24,7 @@ import * as colors from '../../utils/colors.js';
 import { ValidationError } from '../../utils/errors.js';
 import { createLogger, generateCorrelationId } from '../../utils/logger.js';
 import { createSpinner } from '../../utils/progress.js';
+import { runAgentWithRetry } from '../../utils/retry.js';
 import { agentFlags } from '../../utils/shared-flags.js';
 /**
  * Stories Create Command
@@ -220,10 +221,13 @@ export default class StoriesCreateCommand extends Command {
             const processor = async (story) => this.createStory({
                 agent: flags.agent,
                 epicPath,
+                maxRetries: flags['max-retries'],
                 prefix: flags.prefix,
                 references: flags.reference,
+                retryBackoff: flags['retry-backoff'],
                 story,
                 task: flags.task,
+                timeout: flags.timeout,
             });
             // Progress callback for batch updates
             const onProgress = (info) => {
@@ -266,7 +270,7 @@ export default class StoriesCreateCommand extends Command {
      * Create a single story file
      */
     async createStory(options) {
-        const { agent, epicPath, prefix, references, story, task } = options;
+        const { agent, epicPath, maxRetries, prefix, references, retryBackoff, story, task, timeout } = options;
         const storyDir = this.pathResolver.getStoryDir();
         const filename = this.generateStoryFilename(story, prefix);
         const filePath = path.join(storyDir, filename);
@@ -290,10 +294,14 @@ export default class StoriesCreateCommand extends Command {
                 storyFilePath: absolutePath,
                 task,
             });
-            // Step 3: Run Claude agent
-            const result = await this.agentRunner.runAgent(prompt, {
+            // Step 3: Run Claude agent with retry on timeout/killed
+            const result = await runAgentWithRetry(this.agentRunner, prompt, {
                 agentType: 'sm',
-                timeout: 1_800_000, // 30 minutes
+                timeout: timeout ?? 2_700_000,
+            }, {
+                backoffMs: retryBackoff,
+                logger: this.logger,
+                maxRetries,
             });
             // Step 4: Verify file was updated
             if (result.success) {

package/dist/commands/stories/develop.d.ts

@@ -40,6 +40,9 @@ export default class StoriesDevelopCommand extends Command {
         model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
         provider: import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
         task: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
+        timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'max-retries': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'retry-backoff': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
     };
     private agentRunner;
     private fileManager;

package/dist/commands/stories/develop.js

@@ -25,12 +25,8 @@ import { StoryParserFactory } from '../../services/parsers/story-parser-factory.
 import * as colors from '../../utils/colors.js';
 import { createLogger, generateCorrelationId } from '../../utils/logger.js';
 import { createSpinner } from '../../utils/progress.js';
+import { runAgentWithRetry } from '../../utils/retry.js';
 import { agentFlags } from '../../utils/shared-flags.js';
-/**
- * Agent timeout in milliseconds (30 minutes)
- * Dev agents need sufficient time for comprehensive implementation
- */
-const DEV_AGENT_TIMEOUT_MS = 1_800_000;
 /**
  * Stories Develop Command
  *
@@ -251,10 +247,13 @@ export default class StoriesDevelopCommand extends Command {
             const result = await this.developStory({
                 agent: flags.agent,
                 cwd: flags.cwd,
+                maxRetries: flags['max-retries'],
                 references: flags.reference,
+                retryBackoff: flags['retry-backoff'],
                 storyMetadata,
                 storyPath,
                 task: flags.task,
+                timeout: flags.timeout,
             });
             results.push(result);
             if (result.success) {
@@ -275,7 +274,7 @@ export default class StoriesDevelopCommand extends Command {
      * Develop a single story
      */
     async developStory(options) {
-        const { agent, cwd, references, storyMetadata, storyPath, task } = options;
+        const { agent, cwd, maxRetries, references, retryBackoff, storyMetadata, storyPath, task, timeout } = options;
         const storyNumber = isEpicStory(storyMetadata) ? storyMetadata.number : storyMetadata.id;
         this.logger.info({ storyNumber, storyPath }, 'Starting story development');
         try {
@@ -305,11 +304,15 @@ export default class StoriesDevelopCommand extends Command {
                 storyPath,
                 task,
             });
-            // Step 4: Run Claude dev agent
+            // Step 4: Run Claude dev agent with retry on timeout/killed
             this.logger.info({ storyNumber }, 'Running Claude dev agent');
-            const result = await this.agentRunner.runAgent(prompt, {
+            const result = await runAgentWithRetry(this.agentRunner, prompt, {
                 agentType: 'dev',
-                timeout: DEV_AGENT_TIMEOUT_MS,
+                timeout: timeout ?? 2_700_000,
+            }, {
+                backoffMs: retryBackoff,
+                logger: this.logger,
+                maxRetries,
             });
             if (!result.success) {
                 throw new Error(`Dev agent failed: ${result.errors}`);

package/dist/commands/stories/qa.d.ts

@@ -36,6 +36,9 @@ export default class StoriesQaCommand extends Command {
         provider: import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
         'qa-prompt': import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
         reference: import("@oclif/core/interfaces").OptionFlag<string[] | undefined, import("@oclif/core/interfaces").CustomOptions>;
+        timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'agent-retries': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'retry-backoff': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
     };
     private agentRunner;
     private fileManager;

package/dist/commands/stories/qa.js

@@ -23,11 +23,7 @@ import { StoryParserFactory } from '../../services/parsers/story-parser-factory.
 import * as colors from '../../utils/colors.js';
 import { createLogger, generateCorrelationId } from '../../utils/logger.js';
 import { createSpinner } from '../../utils/progress.js';
-/**
- * Agent timeout in milliseconds (30 minutes)
- * QA and Dev agents need longer timeouts for comprehensive analysis
- */
-const AGENT_TIMEOUT_MS = 1_800_000;
+import { runAgentWithRetry } from '../../utils/retry.js';
 /**
  * Stories QA Command
  *
@@ -89,6 +85,21 @@ export default class StoriesQaCommand extends Command {
             description: 'Additional context files for agents',
             multiple: true,
         }),
+        timeout: Flags.integer({
+            default: 2_700_000,
+            description: 'Agent execution timeout in milliseconds (default: 2700000 = 45 minutes)',
+            helpGroup: 'Resilience',
+        }),
+        'agent-retries': Flags.integer({
+            default: 0,
+            description: 'Max retries for timeout/killed agent failures (0 = no retry)',
+            helpGroup: 'Resilience',
+        }),
+        'retry-backoff': Flags.integer({
+            default: 5000,
+            description: 'Backoff delay between retries in milliseconds',
+            helpGroup: 'Resilience',
+        }),
     };
     // Service instances
     agentRunner;
@@ -446,10 +457,17 @@ ${result.finalGate === 'PASS' ? '3. [x] Final QA Validation - PASSED' : result.f
         try {
             // Phase 1: Initial QA Deep Dive
             this.log(colors.info('  Phase 1: QA Deep Dive Review...'));
+            const agentTimeout = flags.timeout ?? 2_700_000;
+            const agentRetries = flags['agent-retries'];
+            const retryBackoff = flags['retry-backoff'];
             const qaPrompt = this.buildQaPrompt(storyPath, flags['qa-prompt'], flags.reference);
-            const qaResult = await this.agentRunner.runAgent(qaPrompt, {
+            const qaResult = await runAgentWithRetry(this.agentRunner, qaPrompt, {
                 agentType: 'tea',
-                timeout: AGENT_TIMEOUT_MS,
+                timeout: agentTimeout,
+            }, {
+                backoffMs: retryBackoff,
+                logger: this.logger,
+                maxRetries: agentRetries,
             });
             if (!qaResult.success) {
                 throw new Error(`QA agent failed: ${qaResult.errors}`);
@@ -465,9 +483,13 @@ ${result.finalGate === 'PASS' ? '3. [x] Final QA Validation - PASSED' : result.f
                 // Run Dev agent to fix issues (sequential retry loop by design)
                 const devPrompt = this.buildDevFixPrompt(storyPath, flags['dev-prompt'], flags.reference);
                 // eslint-disable-next-line no-await-in-loop
-                const devResult = await this.agentRunner.runAgent(devPrompt, {
+                const devResult = await runAgentWithRetry(this.agentRunner, devPrompt, {
                     agentType: 'dev',
-                    timeout: AGENT_TIMEOUT_MS,
+                    timeout: agentTimeout,
+                }, {
+                    backoffMs: retryBackoff,
+                    logger: this.logger,
+                    maxRetries: agentRetries,
                 });
                 if (!devResult.success) {
                     this.logger.warn({ errors: devResult.errors, retriesUsed }, 'Dev fix-forward failed, continuing...');
@@ -476,9 +498,13 @@ ${result.finalGate === 'PASS' ? '3. [x] Final QA Validation - PASSED' : result.f
                 // Phase 3: Re-run QA to validate fixes
                 this.log(colors.info(`  Phase 3: QA Re-validation (Retry ${retriesUsed})...`));
                 // eslint-disable-next-line no-await-in-loop
-                const reQaResult = await this.agentRunner.runAgent(qaPrompt, {
+                const reQaResult = await runAgentWithRetry(this.agentRunner, qaPrompt, {
                     agentType: 'tea',
-                    timeout: AGENT_TIMEOUT_MS,
+                    timeout: agentTimeout,
+                }, {
+                    backoffMs: retryBackoff,
+                    logger: this.logger,
+                    maxRetries: agentRetries,
                 });
                 if (!reQaResult.success) {
                     this.logger.warn({ errors: reQaResult.errors, retriesUsed }, 'QA re-validation failed, continuing...');

package/dist/commands/workflow.d.ts

@@ -42,6 +42,9 @@ export default class Workflow extends Command {
         'skip-epics': import("@oclif/core/interfaces").BooleanFlag<boolean>;
         'skip-stories': import("@oclif/core/interfaces").BooleanFlag<boolean>;
         'story-interval': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'max-retries': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+        'retry-backoff': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
         verbose: import("@oclif/core/interfaces").BooleanFlag<boolean>;
     };
     private cancelled;

package/dist/commands/workflow.js

@@ -118,6 +118,18 @@ export default class Workflow extends Command {
             default: 60,
             description: 'Seconds between story development',
         }),
+        timeout: Flags.integer({
+            default: 2_700_000,
+            description: 'Agent execution timeout in milliseconds (default: 2700000 = 45 minutes)',
+        }),
+        'max-retries': Flags.integer({
+            default: 0,
+            description: 'Max retries for timeout/killed agent failures (0 = no retry)',
+        }),
+        'retry-backoff': Flags.integer({
+            default: 5000,
+            description: 'Backoff delay between retries in milliseconds',
+        }),
         verbose: Flags.boolean({
             char: 'v',
             default: false,
@@ -154,6 +166,7 @@ export default class Workflow extends Command {
                 dryRun: flags['dry-run'],
                 epicInterval: flags['epic-interval'],
                 input: args.input,
+                maxRetries: flags['max-retries'],
                 model: flags.model,
                 parallel: flags.parallel,
                 pipeline: flags.pipeline,
@@ -164,10 +177,12 @@ export default class Workflow extends Command {
                 qaPrompt: flags['qa-prompt'],
                 qaRetries: flags['qa-retries'],
                 references: flags.reference || [],
+                retryBackoffMs: flags['retry-backoff'],
                 skipDev: flags['skip-dev'],
                 skipEpics: flags['skip-epics'],
                 skipStories: flags['skip-stories'],
                 storyInterval: flags['story-interval'],
+                timeout: flags.timeout,
                 verbose: flags.verbose,
             };
             // Log configuration if verbose

package/dist/models/workflow-config.d.ts

@@ -147,6 +147,21 @@ export interface WorkflowConfig {
      * @default 60
      */
     storyInterval: number;
+    /**
+     * Agent execution timeout in milliseconds
+     * @default 2_700_000 (45 minutes)
+     */
+    timeout?: number;
+    /**
+     * Maximum retries for timeout/killed agent failures
+     * @default 0 (no retries, backward compatible)
+     */
+    maxRetries?: number;
+    /**
+     * Backoff delay between retries in milliseconds
+     * @default 5000 (5 seconds)
+     */
+    retryBackoffMs?: number;
     /**
      * Detailed output mode
      *

package/dist/services/orchestration/dependency-graph-executor.d.ts

@@ -27,8 +27,9 @@ export declare class DependencyGraphExecutor {
     private readonly cwd?;
     private readonly fileManager;
     private readonly logger;
+    private readonly model?;
     private readonly taskGraph;
-    constructor(taskGraph: TaskGraph, agentRunner: AIProviderRunner, batchProcessor: BatchProcessor, fileManager: FileManager, logger: pino.Logger, cwd?: string);
+    constructor(taskGraph: TaskGraph, agentRunner: AIProviderRunner, batchProcessor: BatchProcessor, fileManager: FileManager, logger: pino.Logger, cwd?: string, model?: string);
     /**
      * Execute the entire task graph
      *

package/dist/services/orchestration/dependency-graph-executor.js

@@ -14,15 +14,17 @@ export class DependencyGraphExecutor {
     cwd;
     fileManager;
     logger;
+    model;
     taskGraph;
     // eslint-disable-next-line max-params -- Constructor dependencies will be refactored to config object in future
-    constructor(taskGraph, agentRunner, batchProcessor, fileManager, logger, cwd) {
+    constructor(taskGraph, agentRunner, batchProcessor, fileManager, logger, cwd, model) {
         this.taskGraph = taskGraph;
         this.agentRunner = agentRunner;
         this.batchProcessor = batchProcessor;
         this.fileManager = fileManager;
         this.logger = logger;
         this.cwd = cwd;
+        this.model = model;
     }
     /**
      * Execute the entire task graph
@@ -306,6 +308,7 @@ Use the file at the path above to document:
             // Execute agent
             const result = await this.agentRunner.runAgent(fullPrompt, {
                 agentType,
+                model: this.model,
                 references: task.targetFiles, // Pass target files as references
                 timeout: task.estimatedMinutes * 60 * 1000 * 1.5, // 1.5x estimated time as buffer
             });

package/dist/services/orchestration/task-decomposition-service.js

@@ -507,6 +507,7 @@ YOUR RESPONSE MUST START WITH "masterPrompt: |" - NO OTHER TEXT ALLOWED!`;
             this.logger.info('Asking Claude to fix YAML errors');
             const fixResult = await this.agentRunner.runAgent(fixPrompt, {
                 agentType: 'architect',
+                model: options.model,
                 timeout: 60_000, // 1 minute for fix
             });
             if (!fixResult.success) {

package/dist/services/orchestration/workflow-orchestrator.js

@@ -45,6 +45,7 @@
  */
 import { isEpicStory } from '../../models/story.js';
 import { ParserError, ValidationError } from '../../utils/errors.js';
+import { runAgentWithRetry } from '../../utils/retry.js';
 import { PrdFixer } from '../parsers/prd-fixer.js';
 import { FileScaffolder } from '../scaffolding/file-scaffolder.js';
 import { BatchProcessor } from './batch-processor.js';
@@ -491,11 +492,15 @@ Write output to: ${outputPath}`;
                         prompt += '\n';
                     }
                     prompt += '*yolo mode*\n';
-                    // Execute dev agent
-                    const result = await this.agentRunner.runAgent(prompt, {
+                    // Execute dev agent with retry on timeout/killed
+                    const result = await runAgentWithRetry(this.agentRunner, prompt, {
                         agentType: 'dev',
                         references: config.references,
-                        timeout: 1_800_000, // 30 minutes
+                        timeout: config.timeout ?? 2_700_000,
+                    }, {
+                        backoffMs: config.retryBackoffMs,
+                        logger: this.logger,
+                        maxRetries: config.maxRetries,
                     });
                     if (result.success) {
                         // Update story status to Done
@@ -688,10 +693,14 @@ Write output to: ${outputPath}`;
                     prompt += '\n';
                 }
                 prompt += '*yolo mode*\n';
-                const result = await this.agentRunner.runAgent(prompt, {
+                const result = await runAgentWithRetry(this.agentRunner, prompt, {
                     agentType: 'dev',
                     references: config.references,
-                    timeout: 1_800_000, // 30 minutes
+                    timeout: config.timeout ?? 2_700_000,
+                }, {
+                    backoffMs: config.retryBackoffMs,
+                    logger: this.logger,
+                    maxRetries: config.maxRetries,
                 });
                 if (result.success) {
                     // Update story status to Done
@@ -898,10 +907,14 @@ Write output to: ${outputPath}`;
                     }, 'Claude Prompt (Epic)');
                 }
                 // Step 3: Run Claude agent to populate content sections
-                const result = await this.agentRunner.runAgent(prompt, {
+                const result = await runAgentWithRetry(this.agentRunner, prompt, {
                     agentType: 'architect',
                     references: config.references,
-                    timeout: 1_800_000, // 30 minutes
+                    timeout: config.timeout ?? 2_700_000,
+                }, {
+                    backoffMs: config.retryBackoffMs,
+                    logger: this.logger,
+                    maxRetries: config.maxRetries,
                 });
                 // Log output if verbose
                 if (config.verbose) {
@@ -1457,10 +1470,14 @@ Write output to: ${outputPath}`;
                     }, 'Claude Prompt (Story)');
                 }
                 // Step 4: Run Claude agent to populate content sections
-                const result = await this.agentRunner.runAgent(prompt, {
+                const result = await runAgentWithRetry(this.agentRunner, prompt, {
                     agentType: 'sm',
                     references: config.references,
-                    timeout: 1_800_000, // 30 minutes
+                    timeout: config.timeout ?? 2_700_000,
+                }, {
+                    backoffMs: config.retryBackoffMs,
+                    logger: this.logger,
+                    maxRetries: config.maxRetries,
                 });
                 // Log output if verbose
                 if (config.verbose) {
@@ -1684,10 +1701,14 @@ Write output to: ${outputPath}`;
                     }, 'Claude Prompt (Story)');
                 }
                 // Step 4: Run Claude agent to populate content sections
-                const result = await this.agentRunner.runAgent(prompt, {
+                const result = await runAgentWithRetry(this.agentRunner, prompt, {
                     agentType: 'sm',
                     references: config.references,
-                    timeout: 1_800_000, // 30 minutes
+                    timeout: config.timeout ?? 2_700_000,
+                }, {
+                    backoffMs: config.retryBackoffMs,
+                    logger: this.logger,
+                    maxRetries: config.maxRetries,
                 });
                 // Log output if verbose
                 if (config.verbose) {

package/dist/utils/retry.d.ts

@@ -5,6 +5,8 @@
  * transient failures gracefully.
  */
 import type pino from 'pino';
+import type { AgentOptions, AgentResult } from '../models/index.js';
+import type { AIProviderRunner } from '../services/agents/agent-runner.js';
 /**
  * Options for configuring retry behavior
  */
@@ -112,3 +114,30 @@ export declare class RetryStrategy {
  * })
  */
 export declare function isClaudeCliRetryable(error: Error): boolean;
+/**
+ * Options for runAgentWithRetry
+ */
+export interface RunAgentWithRetryOptions {
+    /** Backoff delay between retries in ms. Default: 5000 (short because process already waited for timeout) */
+    backoffMs?: number;
+    /** Logger for retry messages */
+    logger?: pino.Logger;
+    /** Max retry attempts. 0 = no retry (preserves existing behavior). Default: 0 */
+    maxRetries?: number;
+}
+/**
+ * Execute an AI agent with automatic retry on timeout/killed failures
+ *
+ * Bridges the contract gap: runAgent() returns AgentResult (never throws),
+ * but RetryStrategy expects operations that throw. This adapter:
+ * 1. Wraps runAgent in a throwing closure for RetryStrategy
+ * 2. Catches exhausted retries and returns the last AgentResult (never throws)
+ * 3. Short-circuits to plain runAgent when maxRetries <= 0 (zero overhead)
+ *
+ * @param agentRunner - The AI provider runner to execute
+ * @param prompt - The prompt to send
+ * @param options - Agent execution options (timeout, agentType, etc.)
+ * @param retryOptions - Retry configuration
+ * @returns AgentResult (same contract as runAgent — never throws)
+ */
+export declare function runAgentWithRetry(agentRunner: AIProviderRunner, prompt: string, options: Omit<AgentOptions, 'prompt'>, retryOptions?: RunAgentWithRetryOptions): Promise<AgentResult>;

package/dist/utils/retry.js

@@ -4,6 +4,7 @@
  * Provides configurable retry logic with exponential backoff for handling
  * transient failures gracefully.
  */
+import { AgentError } from './errors.js';
 import { createLogger } from './logger.js';
 /**
  * Retry strategy implementation with configurable exponential backoff
@@ -158,3 +159,50 @@ export function isClaudeCliRetryable(error) {
     // Default to not retryable if we can't determine exit code
     return false;
 }
+/**
+ * Execute an AI agent with automatic retry on timeout/killed failures
+ *
+ * Bridges the contract gap: runAgent() returns AgentResult (never throws),
+ * but RetryStrategy expects operations that throw. This adapter:
+ * 1. Wraps runAgent in a throwing closure for RetryStrategy
+ * 2. Catches exhausted retries and returns the last AgentResult (never throws)
+ * 3. Short-circuits to plain runAgent when maxRetries <= 0 (zero overhead)
+ *
+ * @param agentRunner - The AI provider runner to execute
+ * @param prompt - The prompt to send
+ * @param options - Agent execution options (timeout, agentType, etc.)
+ * @param retryOptions - Retry configuration
+ * @returns AgentResult (same contract as runAgent — never throws)
+ */
+export async function runAgentWithRetry(agentRunner, prompt, options, retryOptions = {}) {
+    const maxRetries = retryOptions.maxRetries ?? 0;
+    // Short-circuit: no retries configured — preserve exact existing behavior
+    if (maxRetries <= 0) {
+        return agentRunner.runAgent(prompt, options);
+    }
+    let lastResult;
+    const strategy = new RetryStrategy({
+        backoffMs: retryOptions.backoffMs ?? 5_000,
+        backoffMultiplier: 1.5,
+        isRetryable: isClaudeCliRetryable,
+        logger: retryOptions.logger,
+        maxRetries,
+    });
+    try {
+        return await strategy.execute(async () => {
+            lastResult = await agentRunner.runAgent(prompt, options);
+            if (!lastResult.success) {
+                throw new AgentError(`Agent ${lastResult.agentType} failed with exit code ${lastResult.exitCode}`, {
+                    agentType: lastResult.agentType,
+                    exitCode: lastResult.exitCode,
+                });
+            }
+            return lastResult;
+        });
+    }
+    catch {
+        // All retries exhausted or non-retryable error — return the last AgentResult
+        // This preserves the contract: runAgentWithRetry never throws, just like runAgent
+        return lastResult;
+    }
+}

package/dist/utils/shared-flags.d.ts

@@ -26,4 +26,7 @@ export declare const agentFlags: {
     model: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
     provider: import("@oclif/core/interfaces").OptionFlag<string, import("@oclif/core/interfaces").CustomOptions>;
     task: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
+    timeout: import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+    'max-retries': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
+    'retry-backoff': import("@oclif/core/interfaces").OptionFlag<number, import("@oclif/core/interfaces").CustomOptions>;
 };

package/dist/utils/shared-flags.js

@@ -44,4 +44,19 @@ export const agentFlags = {
         description: 'Override which task command to execute (e.g., develop-story, draft, review-implementation). Defaults to command-appropriate task.',
         helpGroup: 'Agent Customization',
     }),
+    timeout: Flags.integer({
+        default: 2_700_000,
+        description: 'Agent execution timeout in milliseconds (default: 2700000 = 45 minutes)',
+        helpGroup: 'Resilience',
+    }),
+    'max-retries': Flags.integer({
+        default: 0,
+        description: 'Max retries for timeout/killed agent failures (0 = no retry)',
+        helpGroup: 'Resilience',
+    }),
+    'retry-backoff': Flags.integer({
+        default: 5000,
+        description: 'Backoff delay between retries in milliseconds',
+        helpGroup: 'Resilience',
+    }),
 };

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@hyperdrive.bot/bmad-workflow",
   "description": "AI-driven development workflow orchestration CLI for BMAD projects",
-  "version": "1.0.10",
+  "version": "1.0.12",
   "author": {
     "name": "DevSquad",
     "email": "marcelo@devsquad.email",