Alfred

---
description: "Agentic AI automation - From SPEC to code with autonomous loop"
argument-hint: '"task description" [--loop] [--max N] [--seq] | resume SPEC-XXX'
type: utility
allowed-tools: Task, AskUserQuestion, TodoWrite, Bash, Read, Write, Edit, Glob, Grep
model: inherit
---

## Pre-execution Context

!git status --porcelain
!git branch --show-current

## Essential Files

@.moai/config/sections/ralph.yaml
@.moai/config/sections/git-strategy.yaml
@.moai/config/sections/quality.yaml
@.moai/config/sections/llm.yaml

---

# /moai:alfred - Agentic AI Autonomous Automation

## Core Principle: Fully Autonomous Automation

User provides a goal, AI autonomously plans, executes, and completes.

USER: "Add authentication feature" ↓ AI: Explore → Plan → Implement → Verify → Repeat ↓ AI: All issues resolved ↓ AI: DONE

## Command Purpose

Autonomously execute the full MoAI workflow:

1. **Parallel Exploration** (Explore + Research simultaneously)
2. **SPEC Generation** (after user approval)
3. **DDD Implementation** (auto iterative fixing)
4. **Documentation Sync**
5. **Completion Marker Detection** (`<moai>DONE</moai>`)

Feature Description: $ARGUMENTS

## Quick Start

```bash
# Default autonomous execution (parallel exploration)
/moai:alfred "Add JWT authentication"

# Enable auto loop (max 50 iterations)
/moai:alfred "JWT auth" --loop --max 50

# Sequential exploration + auto loop
/moai:alfred "JWT auth" --loop --sequential

# Resume previous work
/moai:alfred resume SPEC-AUTH-001

Command Options

Option	Alias	Description	Default
--loop	-	Enable auto iterative fixing	ralph.yaml
--max N	--max-iterations	Maximum iteration count	100
--sequential	--seq	Sequential exploration (for debugging)	Parallel
--branch	-	Auto-create feature branch	git-strategy
--pr	-	Auto-create PR	git-strategy
--resume SPEC	-	Resume previous work	-

Completion Promise

AI must add a marker when work is complete:

## Complete

All implementation complete, tests passing, docs updated. <moai>DONE</moai>

Marker Types:

• <moai>DONE</moai> - Task complete
• <moai>COMPLETE</moai> - Full completion
• <moai:done /> - XML format

Agentic Autonomous Flow

START: /moai:alfred "task description"

PHASE 0: Parallel Exploration (autonomous)

[SOFT] Apply --ultrathink keyword for autonomous workflow orchestration
WHY: Alfred must analyze task complexity, determine routing strategy, and select appropriate agents
IMPACT: Sequential thinking ensures optimal agent delegation and execution strategy

  ┌── Explore Agent: Codebase analysis
  ├── Research Agent: Documentation/issue search
  └── Quality Agent: Current state diagnosis
  ↓
Integration → Execution plan generation

PHASE 1: SPEC Generation
  └── Written in EARS format
  ↓
User Approval
  ↓
PHASE 2: DDD Implementation (autonomous loop)
  │
  └── WHILE (issues_exist AND iteration < max):
       ├── Diagnostics (LSP + Tests + Coverage)
       ├── TODO generation
       ├── Fix execution
       ├── Verification
       └── Completion marker detected? → BREAK
  ↓
PHASE 3: Documentation Sync
  └── Add marker: <moai>DONE</moai>
  ↓
COMPLETE

Phase 0: Parallel Exploration Implementation

By default, execute three exploration agents simultaneously for optimal performance:

Step 1 - Launch Parallel Agent Tasks:

In a single response, invoke all three Task tools simultaneously (Claude Code executes them in parallel automatically):

1. Explore Agent:

   • subagent_type: Explore
   • description: Codebase analysis for task context
   • prompt: Include user task description, request relevant files, architecture patterns, existing implementations

2. Research Agent:

   • subagent_type: Explore (with WebSearch/WebFetch focus)
   • description: External documentation and best practices research
   • prompt: Include user task description, request API docs, library documentation, similar implementations

3. Quality Agent:

   • subagent_type: manager-quality
   • description: Current project quality assessment
   • prompt: Request test coverage status, lint status, technical debt assessment

All three Task calls in the same response triggers automatic parallel execution (up to 10 concurrent).

Step 2 - Collect and Integrate Results:

After all three agents complete:

1. Collect outputs from each agent response

2. Extract key findings:

   • From Explore: Relevant files, architecture patterns, existing code to reference
   • From Research: External knowledge, best practices, API patterns
   • From Quality: Current test coverage, code quality baseline, known issues

3. Synthesize into unified exploration report

Step 3 - Generate Execution Plan:

Based on integrated findings:

1. Identify implementation approach
2. List files to create/modify
3. Define test strategy
4. Estimate scope and complexity

Step 4 - User Approval Checkpoint:

Present integrated findings and proposed plan to user via AskUserQuestion:

• Options: Proceed to SPEC creation, Modify approach, Cancel

Error Handling:

If any agent fails:

• Continue with results from successful agents
• Note missing information in plan
• Offer to retry failed agent or proceed with partial information

WHY: Parallel exploration reduces Phase 0 time from 45-90 seconds to 15-30 seconds (2-3x speedup) IMPACT: Faster initial context gathering without sacrificing comprehensiveness

TODO-Obsessive Rule

[HARD] TodoWrite Tool Mandatory Usage:

1. Immediate Creation: When issues are discovered, call TodoWrite tool to add items with pending status
2. Immediat