Skip to main content
OpenCode Pro includes specialized agents designed for complex multi-phase workflows. Each agent has specific tools, skills, and model configurations optimized for their task.

​
Available Agents

Planner

Break down complex tasks into implementation plans before writing code

Reviewer

Code review specialist for logic, security, and quality checks

Scout

Confidence-gated exploration with GO/HOLD verdicts

Orchestrator

Multi-phase development with Research > Plan > Implement workflow

Debugger

Systematic bug investigation and root cause analysis

​
When to Use Agents

Use Orchestrator when building features that touch >5 files or require architecture decisions. It runs through Research, Plan, and Implement phases with validation gates.
Use Planner when you need to break down complex tasks before implementation. It provides read-only exploration and structured plans.
Use Scout before starting unfamiliar tasks. It scores confidence 0-100 and gives a GO/HOLD verdict, running in the background so you can continue working.
Use Reviewer before committing, for PR reviews, or after major changes. It checks logic, security, performance, and test coverage.
Use Debugger for systematic investigation of test failures or runtime errors. It hypothesizes, investigates, and proposes fixes methodically.

​
Agent Capabilities

​
Tools Available

Each agent has access to specific tools optimized for their role: 
Planner:
  - Read
  - Glob
  - Grep

Scout:
  - Read
  - Glob  
  - Grep
  - Bash (read-only)

​
Special Capabilities

Background Execution

Scout runs in the background, allowing you to continue working while it explores the codebase.

Isolated Worktree

Scout runs in an isolated worktree to avoid interfering with your main session.

Project Memory

Orchestrator and Debugger use project memory to recall patterns from previous work.

Advanced Model

Orchestrator and Debugger use the Opus model for complex reasoning tasks.

​
Agent Workflow Patterns

​
Phase-Gated Development

Agents like Orchestrator use multi-phase workflows with validation gates:
Agents never skip phases. Each phase must pass validation before proceeding to the next.

​
Read-Only Exploration

Agents like Planner and Scout perform read-only exploration:
  1. Understand the goal
  2. Explore relevant code
  3. Score confidence or create plan
  4. Present findings for approval
Read-only agents never make changes. They provide analysis and recommendations only.

​
Best Practices

1

Choose the Right Agent

Match the agent to your task. Use Orchestrator for features, Debugger for bugs, Reviewer for quality checks.
2

Provide Clear Context

Give agents clear task descriptions. The better the input, the better the output.
3

Trust the Process

Let agents complete their workflows. Don’t skip validation gates or approval steps.
4

Review Outputs

Always review agent plans and proposals before approving implementation.

​
Configuration

Each agent is configured with frontmatter in their definition file: 
---
name: orchestrator
description: Multi-phase development agent
tools: ["Read", "Glob", "Grep", "Bash", "Edit", "Write"]
skills: ["pro-workflow"]
model: opus
memory: project
background: false
isolation: none
---
​
name
string
required
Unique identifier for the agent
​
description
string
required
Brief description of when to use the agent
​
tools
array
required
List of tools the agent can use
​
skills
array
Optional skills the agent can load
​
model
string
Model preference (e.g., “opus” for advanced reasoning)
​
memory
string
Memory type (e.g., “project” to recall previous work)
​
background
boolean
Whether the agent runs in the background
​
isolation
string
Isolation mode (e.g., “worktree” for isolated execution)

​
Next Steps

Orchestrator Deep Dive

Learn about multi-phase feature development

Scout Exploration

Understand confidence-gated exploration

Build docs developers (and LLMs) love

Get started for freeTalk to us