Agent Configuration

Agent Board loads AI agent configurations from multiple sources and merges them into a unified agent list.

Agent Sources

Agents are loaded in this priority order (later sources can override earlier ones):

1. Skills Repository — SKILL.md files in skills/*/ directories
2. Agent Definitions — *.agent.md files in agents/ directories
3. Custom Config — JSON file specified in agents.configPath
4. Built-in Defaults — Fallback agents (planner, developer, reviewer)

Skills Repository Structure

Point agents.repoPaths to a directory containing skills:

ai-agents/
├── skills/
│   ├── architecture/
│   │   └── SKILL.md
│   ├── frontend-patterns/
│   │   └── SKILL.md
│   ├── testing/
│   │   └── SKILL.md
│   └── golang-patterns/
│       └── SKILL.md
└── agents/
    ├── code-reviewer.agent.md
    ├── lead-architect.agent.md
    └── python-expert.agent.md

SKILL.md Format

Each skill folder contains a SKILL.md with optional frontmatter:

---
name: Architecture Reviewer
description: Reviews architecture decisions and patterns
---

# Architecture Review Skill

You are an expert architecture reviewer...

The entire file content becomes the agent's systemPrompt.

Agent.md Format

Agent definition files use Markdown with a heading:

# Lead Architect

You are a lead architect responsible for...

Role Mapping

Agents are automatically assigned roles based on their folder/file name:

Names	Role
architecture, system-design, lead-architect	architect
testing, test-strategist	reviewer
project-management, task-orchestrator	planner
devops-agent, gcp-patterns	devops
code-reviewer, anti-patterns	reviewer
Everything else	developer

Custom Agent Config (JSON)

Create a JSON file for custom agent definitions:

[
  {
    "id": "agent-custom-reviewer",
    "name": "Custom Reviewer",
    "role": "reviewer",
    "avatar": "🔍",
    "color": "#e74c3c",
    "model": "copilot-claude-sonnet-4",
    "temperature": 0.3,
    "maxContextTokens": 128000,
    "systemPrompt": "You are a code reviewer specializing in..."
  }
]

Then reference it in board.yaml:

agents:
  configPath: ~/configs/my-agents.json

Per-Agent Backend

Each agent can use a different AI backend:

{
  "id": "agent-developer",
  "backend": "claude-cli",
  "...": "..."
}

If no backend is specified, the agent uses the board's default backend.

AgentConfig Type

interface AgentConfig {
  id: string           // Unique identifier (e.g., "agent-architecture")
  name: string         // Display name
  role: string         // planner | developer | reviewer | architect | devops
  avatar: string       // Emoji avatar
  color: string        // Hex color for UI
  model: string        // LLM model identifier
  temperature: number  // 0.0 - 1.0
  maxContextTokens: number
  systemPrompt: string // Full system prompt text
  backend?: BackendId  // Optional backend override
}