- Update 26+ files to import secureFs from @automaker/platform - Create shared type files: github.ts, worktree.ts, claude.ts in libs/types - Create exec-utils.ts for cross-platform shell execution - Delete redundant wrapper files: secure-fs.ts, stream-processor.ts, enhancement-prompts.ts - Update GitHub routes to use createLogError pattern - Add isENOENT helper to routes/common.ts - Fix test imports to use @automaker/prompts and @automaker/platform 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
22 KiB
Auto Mode Service Refactoring Plan
Overview
This document proposes a refactoring of apps/server/src/services/auto-mode-service.ts (2,497 lines) into smaller, focused modules while leveraging shared packages per docs/llm-shared-packages.md.
Current Problems
- Monolithic Service - 2,497 lines with 16+ public methods mixing concerns
- Giant Method -
runAgent()is 658 lines handling planning, approval, task execution, file I/O - Code Duplication - Stream processing repeated 4x, types duplicated in 3 files
- Missing Shared Package Usage - Local types that exist in
@automaker/types, prompts not in@automaker/prompts - No Structured Logging - Uses
console.loginstead ofcreateLogger - Untyped Events - Event strings scattered, no type safety
Proposed Architecture
apps/server/src/services/
├── auto-mode/
│ ├── index.ts # Re-exports AutoModeService (facade)
│ ├── auto-mode-service.ts # Orchestrator (~300 lines)
│ ├── feature-executor.ts # Feature execution logic
│ ├── plan-approval-service.ts # Plan approval flow
│ ├── task-executor.ts # Multi-task execution
│ ├── worktree-manager.ts # Git worktree operations
│ ├── output-writer.ts # Incremental file writing
│ └── types.ts # Internal types (RunningFeature, etc.)
├── feature-loader.ts # Existing - extend with status updates
└── ...
libs/types/src/
├── planning.ts # NEW: ParsedTask, PlanSpec, AutoModeEventType
├── feature.ts # Update: use PlanSpec from planning.ts
└── index.ts # Export planning types
libs/prompts/src/
├── planning.ts # NEW: PLANNING_PROMPTS, task parsing, buildTaskPrompt
├── enhancement.ts # Existing
└── index.ts # Export planning functions
apps/server/src/lib/
├── stream-processor.ts # NEW: Reusable stream processing utility
└── ...
Phase 1: Shared Package Updates (No Breaking Changes)
1.1 Add Planning Types to @automaker/types
Create libs/types/src/planning.ts:
// Task and plan status types
export type TaskStatus = 'pending' | 'in_progress' | 'completed' | 'failed';
export type PlanSpecStatus = 'pending' | 'generating' | 'generated' | 'approved' | 'rejected';
export interface ParsedTask {
id: string; // e.g., "T001"
description: string; // e.g., "Create user model"
filePath?: string; // e.g., "src/models/user.ts"
phase?: string; // e.g., "Phase 1: Foundation"
status: TaskStatus;
}
export interface PlanSpec {
status: PlanSpecStatus;
content?: string;
version: number;
generatedAt?: string;
approvedAt?: string;
reviewedByUser: boolean;
tasksCompleted?: number;
tasksTotal?: number;
currentTaskId?: string;
tasks?: ParsedTask[];
}
// Auto-mode event types for type safety
export type AutoModeEventType =
| 'auto_mode_started'
| 'auto_mode_stopped'
| 'auto_mode_idle'
| 'auto_mode_feature_start'
| 'auto_mode_feature_complete'
| 'auto_mode_progress'
| 'auto_mode_tool'
| 'auto_mode_error'
| 'auto_mode_task_started'
| 'auto_mode_task_complete'
| 'auto_mode_phase_complete'
| 'planning_started'
| 'plan_approval_required'
| 'plan_approved'
| 'plan_rejected'
| 'plan_auto_approved'
| 'plan_revision_requested';
Update libs/types/src/feature.ts to import PlanSpec:
import type { PlanSpec } from './planning.js';
export interface Feature {
// ... existing fields ...
planSpec?: PlanSpec; // Now references shared type
}
1.2 Add Planning Prompts to @automaker/prompts
Create libs/prompts/src/planning.ts:
import type { PlanningMode, ParsedTask } from '@automaker/types';
// Planning mode prompts (moved from auto-mode-service.ts lines 57-211)
export const PLANNING_PROMPTS = {
lite: `## Planning Phase (Lite Mode)...`,
lite_with_approval: `## Planning Phase (Lite Mode)...`,
spec: `## Specification Phase (Spec Mode)...`,
full: `## Full Specification Phase (Full SDD Mode)...`,
};
/**
* Get planning prompt for a mode
*/
export function getPlanningPrompt(mode: PlanningMode, requireApproval?: boolean): string {
if (mode === 'skip') return '';
if (mode === 'lite' && requireApproval) return PLANNING_PROMPTS.lite_with_approval;
return PLANNING_PROMPTS[mode] || '';
}
/**
* Parse tasks from generated spec content
* Looks for ```tasks code block and extracts task lines
*/
export function parseTasksFromSpec(specContent: string): ParsedTask[] {
// ... moved from auto-mode-service.ts lines 218-265
}
/**
* Parse a single task line
* Format: - [ ] T###: Description | File: path/to/file
*/
export function parseTaskLine(line: string, currentPhase?: string): ParsedTask | null {
// ... moved from auto-mode-service.ts lines 271-295
}
/**
* Build a focused prompt for executing a single task
*/
export function buildTaskPrompt(
task: ParsedTask,
allTasks: ParsedTask[],
taskIndex: number,
planContent: string,
userFeedback?: string
): string {
// ... moved from auto-mode-service.ts lines 2389-2458
}
Phase 2: Extract Utility Classes
2.1 Create Stream Processor
Create apps/server/src/lib/stream-processor.ts:
import type { ProviderMessage } from '../providers/types.js';
export interface StreamHandlers {
onText?: (text: string) => void;
onToolUse?: (name: string, input: unknown) => void;
onError?: (error: string) => void;
onComplete?: (result: string) => void;
}
/**
* Process provider message stream with unified handling
* Eliminates the 4x duplicated stream processing pattern
*/
export async function* processStream(
stream: AsyncGenerator<ProviderMessage>,
handlers: StreamHandlers
): AsyncGenerator<{ text: string; isComplete: boolean }> {
for await (const msg of stream) {
if (msg.type === 'assistant' && msg.message?.content) {
for (const block of msg.message.content) {
if (block.type === 'text') {
handlers.onText?.(block.text || '');
yield { text: block.text || '', isComplete: false };
} else if (block.type === 'tool_use') {
handlers.onToolUse?.(block.name, block.input);
}
}
} else if (msg.type === 'error') {
handlers.onError?.(msg.error || 'Unknown error');
throw new Error(msg.error || 'Unknown error');
} else if (msg.type === 'result' && msg.subtype === 'success') {
handlers.onComplete?.(msg.result || '');
yield { text: msg.result || '', isComplete: true };
}
}
}
2.2 Create Output Writer
Create apps/server/src/services/auto-mode/output-writer.ts:
import * as secureFs from '../../lib/secure-fs.js';
import path from 'path';
/**
* Handles incremental, debounced file writing for agent output
*/
export class OutputWriter {
private content = '';
private writeTimeout: ReturnType<typeof setTimeout> | null = null;
private readonly debounceMs: number;
constructor(
private readonly outputPath: string,
debounceMs = 500
) {
this.debounceMs = debounceMs;
}
append(text: string): void {
this.content += text;
this.scheduleWrite();
}
getContent(): string {
return this.content;
}
private scheduleWrite(): void {
if (this.writeTimeout) {
clearTimeout(this.writeTimeout);
}
this.writeTimeout = setTimeout(() => this.flush(), this.debounceMs);
}
async flush(): Promise<void> {
if (this.writeTimeout) {
clearTimeout(this.writeTimeout);
this.writeTimeout = null;
}
try {
await secureFs.mkdir(path.dirname(this.outputPath), { recursive: true });
await secureFs.writeFile(this.outputPath, this.content);
} catch (error) {
console.error(`[OutputWriter] Failed to write: ${error}`);
}
}
}
Phase 3: Extract Service Classes
3.1 Plan Approval Service
Create apps/server/src/services/auto-mode/plan-approval-service.ts:
import type { EventEmitter } from '../../lib/events.js';
import type { PlanSpec, AutoModeEventType } from '@automaker/types';
import { createLogger } from '@automaker/utils';
interface PendingApproval {
resolve: (result: ApprovalResult) => void;
reject: (error: Error) => void;
featureId: string;
projectPath: string;
}
interface ApprovalResult {
approved: boolean;
editedPlan?: string;
feedback?: string;
}
const logger = createLogger('PlanApprovalService');
export class PlanApprovalService {
private pendingApprovals = new Map<string, PendingApproval>();
constructor(private events: EventEmitter) {}
waitForApproval(featureId: string, projectPath: string): Promise<ApprovalResult> {
logger.debug(`Registering pending approval for feature ${featureId}`);
return new Promise((resolve, reject) => {
this.pendingApprovals.set(featureId, { resolve, reject, featureId, projectPath });
});
}
async resolve(
featureId: string,
approved: boolean,
editedPlan?: string,
feedback?: string
): Promise<{ success: boolean; error?: string }> {
const pending = this.pendingApprovals.get(featureId);
if (!pending) {
return { success: false, error: `No pending approval for ${featureId}` };
}
pending.resolve({ approved, editedPlan, feedback });
this.pendingApprovals.delete(featureId);
return { success: true };
}
cancel(featureId: string): void {
const pending = this.pendingApprovals.get(featureId);
if (pending) {
pending.reject(new Error('Plan approval cancelled'));
this.pendingApprovals.delete(featureId);
}
}
hasPending(featureId: string): boolean {
return this.pendingApprovals.has(featureId);
}
}
3.2 Task Executor
Create apps/server/src/services/auto-mode/task-executor.ts:
import type { ExecuteOptions, ParsedTask } from '@automaker/types';
import type { EventEmitter } from '../../lib/events.js';
import type { BaseProvider } from '../../providers/base-provider.js';
import { buildTaskPrompt } from '@automaker/prompts';
import { createLogger } from '@automaker/utils';
const logger = createLogger('TaskExecutor');
interface TaskExecutionContext {
workDir: string;
featureId: string;
projectPath: string;
provider: BaseProvider;
model: string;
maxTurns: number;
allowedTools?: string[];
abortController: AbortController;
planContent: string;
userFeedback?: string;
}
interface TaskProgress {
taskId: string;
taskIndex: number;
tasksTotal: number;
status: 'started' | 'completed' | 'failed';
output?: string;
phaseComplete?: number;
}
export class TaskExecutor {
constructor(private events: EventEmitter) {}
async *executeAll(
tasks: ParsedTask[],
context: TaskExecutionContext
): AsyncGenerator<TaskProgress> {
for (let taskIndex = 0; taskIndex < tasks.length; taskIndex++) {
const task = tasks[taskIndex];
if (context.abortController.signal.aborted) {
throw new Error('Feature execution aborted');
}
logger.info(`Starting task ${task.id}: ${task.description}`);
yield {
taskId: task.id,
taskIndex,
tasksTotal: tasks.length,
status: 'started',
};
const taskPrompt = buildTaskPrompt(
task,
tasks,
taskIndex,
context.planContent,
context.userFeedback
);
const taskStream = context.provider.executeQuery({
prompt: taskPrompt,
model: context.model,
maxTurns: Math.min(context.maxTurns, 50),
cwd: context.workDir,
allowedTools: context.allowedTools,
abortController: context.abortController,
});
let taskOutput = '';
for await (const msg of taskStream) {
// Process stream messages...
if (msg.type === 'assistant' && msg.message?.content) {
for (const block of msg.message.content) {
if (block.type === 'text') {
taskOutput += block.text || '';
this.events.emit('auto-mode:event', {
type: 'auto_mode_progress',
featureId: context.featureId,
content: block.text,
});
}
}
}
}
logger.info(`Task ${task.id} completed`);
yield {
taskId: task.id,
taskIndex,
tasksTotal: tasks.length,
status: 'completed',
output: taskOutput,
phaseComplete: this.checkPhaseComplete(task, tasks, taskIndex),
};
}
}
private checkPhaseComplete(
task: ParsedTask,
allTasks: ParsedTask[],
taskIndex: number
): number | undefined {
if (!task.phase) return undefined;
const nextTask = allTasks[taskIndex + 1];
if (!nextTask || nextTask.phase !== task.phase) {
const phaseMatch = task.phase.match(/Phase\s*(\d+)/i);
return phaseMatch ? parseInt(phaseMatch[1], 10) : undefined;
}
return undefined;
}
}
3.3 Worktree Manager
Create apps/server/src/services/auto-mode/worktree-manager.ts:
import { exec } from 'child_process';
import { promisify } from 'util';
import path from 'path';
import { createLogger } from '@automaker/utils';
const execAsync = promisify(exec);
const logger = createLogger('WorktreeManager');
export class WorktreeManager {
/**
* Find existing worktree path for a branch
*/
async findWorktreeForBranch(projectPath: string, branchName: string): Promise<string | null> {
try {
const { stdout } = await execAsync('git worktree list --porcelain', {
cwd: projectPath,
});
const lines = stdout.split('\n');
let currentPath: string | null = null;
let currentBranch: string | null = null;
for (const line of lines) {
if (line.startsWith('worktree ')) {
currentPath = line.slice(9);
} else if (line.startsWith('branch ')) {
currentBranch = line.slice(7).replace('refs/heads/', '');
} else if (line === '' && currentPath && currentBranch) {
if (currentBranch === branchName) {
return path.isAbsolute(currentPath)
? path.resolve(currentPath)
: path.resolve(projectPath, currentPath);
}
currentPath = null;
currentBranch = null;
}
}
// Check last entry
if (currentPath && currentBranch === branchName) {
return path.isAbsolute(currentPath)
? path.resolve(currentPath)
: path.resolve(projectPath, currentPath);
}
return null;
} catch (error) {
logger.warn(`Failed to find worktree for branch ${branchName}: ${error}`);
return null;
}
}
/**
* Resolve working directory for feature execution
*/
async resolveWorkDir(
projectPath: string,
branchName: string | undefined,
useWorktrees: boolean
): Promise<{ workDir: string; worktreePath: string | null }> {
let worktreePath: string | null = null;
if (useWorktrees && branchName) {
worktreePath = await this.findWorktreeForBranch(projectPath, branchName);
if (worktreePath) {
logger.info(`Using worktree for branch "${branchName}": ${worktreePath}`);
} else {
logger.warn(`Worktree for branch "${branchName}" not found, using project path`);
}
}
const workDir = worktreePath ? path.resolve(worktreePath) : path.resolve(projectPath);
return { workDir, worktreePath };
}
}
Phase 4: Refactor AutoModeService
4.1 Simplified AutoModeService
The refactored auto-mode-service.ts (~300-400 lines) becomes an orchestrator:
import type { Feature, ExecuteOptions, PlanSpec, ParsedTask } from '@automaker/types';
import { createLogger, classifyError, buildPromptWithImages, loadContextFiles } from '@automaker/utils';
import { resolveModelString, DEFAULT_MODELS } from '@automaker/model-resolver';
import { getPlanningPrompt, parseTasksFromSpec } from '@automaker/prompts';
import { getFeatureDir } from '@automaker/platform';
import { ProviderFactory } from '../providers/provider-factory.js';
import { validateWorkingDirectory } from '../lib/sdk-options.js';
import type { EventEmitter } from '../lib/events.js';
import { PlanApprovalService } from './auto-mode/plan-approval-service.js';
import { TaskExecutor } from './auto-mode/task-executor.js';
import { WorktreeManager } from './auto-mode/worktree-manager.js';
import { OutputWriter } from './auto-mode/output-writer.js';
import { FeatureLoader } from './feature-loader.js';
const logger = createLogger('AutoModeService');
export class AutoModeService {
private events: EventEmitter;
private runningFeatures = new Map<string, RunningFeature>();
private featureLoader = new FeatureLoader();
private planApproval: PlanApprovalService;
private taskExecutor: TaskExecutor;
private worktreeManager: WorktreeManager;
constructor(events: EventEmitter) {
this.events = events;
this.planApproval = new PlanApprovalService(events);
this.taskExecutor = new TaskExecutor(events);
this.worktreeManager = new WorktreeManager();
}
// Public methods remain the same API, but delegate to sub-services
async executeFeature(...): Promise<void> {
// Validation
validateWorkingDirectory(projectPath);
// Resolve work directory via WorktreeManager
const { workDir, worktreePath } = await this.worktreeManager.resolveWorkDir(
projectPath, feature.branchName, useWorktrees
);
// Build prompt (simplified)
const prompt = getPlanningPrompt(feature.planningMode, feature.requirePlanApproval)
+ this.buildFeaturePrompt(feature);
// Execute agent (delegated to runAgent which uses TaskExecutor)
await this.runAgent(workDir, featureId, prompt, ...);
}
// Plan approval delegated to PlanApprovalService
waitForPlanApproval = this.planApproval.waitForApproval.bind(this.planApproval);
resolvePlanApproval = this.planApproval.resolve.bind(this.planApproval);
cancelPlanApproval = this.planApproval.cancel.bind(this.planApproval);
hasPendingApproval = this.planApproval.hasPending.bind(this.planApproval);
}
File Changes Summary
New Files to Create
| File | Purpose | Est. Lines |
|---|---|---|
libs/types/src/planning.ts |
Planning types | ~50 |
libs/prompts/src/planning.ts |
Planning prompts & parsing | ~200 |
apps/server/src/lib/stream-processor.ts |
Stream utility | ~50 |
apps/server/src/services/auto-mode/index.ts |
Re-exports | ~10 |
apps/server/src/services/auto-mode/plan-approval-service.ts |
Approval logic | ~100 |
apps/server/src/services/auto-mode/task-executor.ts |
Task execution | ~150 |
apps/server/src/services/auto-mode/worktree-manager.ts |
Git worktrees | ~80 |
apps/server/src/services/auto-mode/output-writer.ts |
File I/O | ~60 |
apps/server/src/services/auto-mode/types.ts |
Internal types | ~40 |
Files to Modify
| File | Changes |
|---|---|
libs/types/src/index.ts |
Export planning types |
libs/types/src/feature.ts |
Import PlanSpec |
libs/prompts/src/index.ts |
Export planning functions |
apps/server/src/services/auto-mode-service.ts |
Refactor to orchestrator |
apps/ui/src/store/app-store.ts |
Import types from @automaker/types |
apps/ui/src/components/.../planning-mode-selector.tsx |
Import PlanningMode |
apps/server/tests/.../auto-mode-task-parsing.test.ts |
Import from @automaker/prompts |
Files to Delete (after refactoring)
None - old file becomes the slim orchestrator.
Verification Checklist
After refactoring, verify:
npm run build:packagessucceedsnpm run lintpassesnpm run test:packagespassesnpm run test:serverpassesnpm run test(E2E) passes- Feature execution works in UI
- Plan approval flow works (spec/full modes)
- Task progress events appear correctly
- Resume feature works
- Follow-up feature works
Migration Strategy
- Phase 1: Add shared package updates (non-breaking)
- Phase 2: Extract utilities (stream-processor, output-writer)
- Phase 3: Extract services one at a time, keeping old code as fallback
- Phase 4: Wire up orchestrator, remove old code
- Phase 5: Update tests to use new imports
Each phase should be a separate PR for easier review.
Benefits
| Metric | Before | After |
|---|---|---|
| Main file lines | 2,497 | ~400 |
| Largest method | 658 lines | ~100 lines |
| Code duplication | 4x stream processing | 1 utility |
| Type safety | None for events | Full |
| Testability | Hard (monolith) | Easy (focused units) |
| Logging | console.log | createLogger |
Open Questions
- Should
startAutoLoop/stopAutoLoopremain in AutoModeService or become a separateAutoLoopService? - Should we add a
FeatureRepositoryclass to consolidate all feature file operations? - Is the recovery mechanism in
resolvePlanApprovalstill needed with the refactored architecture?