ABOV3 logo
Home / Documentation / Agents

AI Agents

Configure specialized AI agents with specific tools, permissions, and behaviors in ABOV3.

Built-in Agents

🔨 Build Agent

Type: Primary Agent (default)

Purpose: Code generation, implementation, and technical tasks

Best for: Writing code, fixing bugs, implementing features, refactoring

✅ bash ✅ edit ✅ write ✅ read ✅ grep ✅ glob

📋 Plan Agent

Type: Primary Agent

Purpose: Architecture, design, and planning tasks

Best for: System design, planning implementations, code reviews

❌ bash ❌ edit ❌ write ✅ read ✅ grep ✅ glob

💬 General Agent

Type: Subagent

Purpose: General questions, explanations, and non-coding tasks

Best for: Explanations, documentation, general Q&A

❌ bash ❌ edit ❌ write ✅ read ✅ webfetch

Agent Configuration

Configuration Methods

  1. JSON Configuration - In ~/.config/abov3/abov3.json
  2. Markdown Files - In ~/.config/abov3/agent/*.md
  3. Project-specific - In .abov3/agent/*.md

Configuration Options

Option Type Description Example
description string Brief explanation of agent's purpose "Handles security audits"
mode string Agent type: primary, subagent, or all "subagent"
model string Override default model "anthropic/claude-3-opus"
temperature number Response randomness (0.0-1.0) 0.7
prompt string Custom system prompt file "security-prompt.md"
tools object Enable/disable specific tools {"bash": false}
permission object Tool permission levels {"edit": "ask"}
disable boolean Disable the agent false

Creating Custom Agents

Method 1: Markdown Configuration

Create a file in ~/.config/abov3/agent/security.md:

---
description: Security audit specialist
mode: subagent
temperature: 0.3
tools:
  bash: false
  write: false
  edit: false
permission:
  read: allow
  webfetch: ask
---

You are a security expert specializing in code audits and vulnerability detection.

Focus on:
- Identifying security vulnerabilities
- Checking for exposed credentials
- Finding SQL injection risks
- Detecting XSS vulnerabilities
- Reviewing authentication flows

Always provide CVE references when applicable.

Method 2: JSON Configuration

Add to ~/.config/abov3/abov3.json:

{
  "agent": {
    "debugger": {
      "description": "Specialized debugging assistant",
      "mode": "primary",
      "model": "anthropic/claude-3-sonnet",
      "temperature": 0.5,
      "tools": {
        "bash": true,
        "read": true,
        "grep": true,
        "edit": false
      },
      "permission": {
        "bash": "ask"
      }
    }
  }
}

Tools and Permissions

Available Tools

File Operations

  • read - Read file contents
  • write - Create new files
  • edit - Modify existing files
  • patch - Apply patches
  • glob - Find files by pattern
  • grep - Search file contents
  • list - List directory contents

Other Tools

  • bash - Execute shell commands
  • webfetch - Fetch web content
  • todowrite - Manage task lists
  • todoread - Read task lists

Permission Levels

  • allow - No restrictions, tool can be used freely
  • ask - Requires user confirmation before each use
  • deny - Tool is completely disabled

💡 Security Best Practice

Set bash: "ask" for agents that don't need frequent shell access. This prevents unintended command execution.

Example Agents

📝 Documentation Writer

---
description: Technical documentation specialist
mode: subagent
tools:
  bash: false
  edit: true
  write: true
---

You are a technical writer. Create clear, comprehensive documentation
following industry best practices. Use proper markdown formatting.

🧪 Test Generator

---
description: Unit test and integration test creator
mode: subagent
model: anthropic/claude-3-opus
tools:
  write: true
  read: true
  bash: true
permission:
  bash: ask
---

You specialize in creating comprehensive test suites.
Always aim for high code coverage and edge case handling.

🔍 Code Reviewer

---
description: Code review and best practices enforcer
mode: subagent
temperature: 0.3
tools:
  read: true
  grep: true
  edit: false
  write: false
---

You are a senior developer conducting code reviews.
Focus on code quality, performance, security, and maintainability.

Using Agents

In TUI

  • Switch agents with Tab / Shift+Tab
  • View available agents with /agents
  • Mention subagents with @agent-name

In CLI

# Use specific agent
./abov3-linux-x64 run --agent debugger "Find the bug in this code"

# List available agents
./abov3-linux-x64 agent list

Interactive Creation

# Create new agent interactively
./abov3-linux-x64 agent create

🎯 Agent Selection Strategy

  • Build Agent - When you need to write or modify code
  • Plan Agent - For architecture decisions and code reviews
  • General Agent - For explanations and documentation
  • Custom Agents - For specialized tasks in your workflow