minimum-viable-agentic

# Minimum Viable Agentic Layer Skill

Guide teams through creating the essential agentic layer components to start agentic coding.

## When to Use

- Starting agentic coding in a new project
- Adding agentic layer to existing codebase
- Understanding bare minimum requirements
- Quick-starting agentic automation

## Core Concept

> "For your minimum viable agentic layer, you really only need these pieces: AI developer workflow directory, prompts, and plans."

## Minimum Viable Structure

```text
project/
├── specs/                    # Plans for agents
│   └── (generated plans go here)
├── .claude/
│   └── commands/            # Agentic prompts
│       ├── chore.md         # Chore planning
│       └── implement.md     # Implementation HOP
└── adws/                    # AI Developer Workflows
    ├── adw_modules/
    │   └── agent.py         # Core execution
    └── adw_chore_implement.py  # Gateway script
```

Total: 4-5 files to bootstrap.

## Implementation Workflow

### Step 1: Create Directory Structure

```bash
mkdir -p specs
mkdir -p .claude/commands
mkdir -p adws/adw_modules
```

### Step 2: Create Chore Template

`.claude/commands/chore.md`:

```markdown
# Chore Planning

Create a detailed plan for this chore task.

## Task
$ARGUMENTS

## Output
Create a spec file at: specs/chore-{adw_id}-{name}.md

Include:
- Task description
- Files to modify
- Step-by-step implementation
- Validation criteria
```

### Step 3: Create Implement HOP

`.claude/commands/implement.md`:

```markdown
# Implementation

Implement the plan provided.

## Plan File
$ARGUMENTS

Read the plan file and implement each step.
Report changes with git diff --stat when complete.
```

### Step 4: Create Agent Module

`adws/adw_modules/agent.py`:

- Claude Code subprocess execution
- Request/response data models
- Output file handling
- Error handling

### Step 5: Create Gateway Script

`adws/adw_chore_implement.py`:

- Accept chore description
- Execute /chore to generate plan
- Execute /impleme