agent-roles

name	agent-roles
description	Unified multi-agent role system. Covers Builder, Coder, Council, Orchestrator, Plan, Researcher, Reviewer, and Scribe personas. Use when coordinating multi-agent workflows with MECE principle and role delegation.

Agent Roles — Multi-Agent System

Context

This skill manages the autonomous agent team system. Each agent has a specialized role and operates under the MECE principle (Mutually Exclusive, Collectively Exhaustive). Agents pass work through a defined pipeline: Plan → Researcher → Coder → Reviewer.

Use this skill when:

• Coordinating multi-agent workflows
• Delegating tasks between specialized agent personas
• Setting up or modifying the agent team structure
• Any task requires routing to the correct agent role

Available Agent Roles:

Role	Responsibility
Orchestrator	Coordinates the entire workflow, assigns tasks, manages handoffs
Plan	Breaks down complex tasks into actionable steps and architecture
Researcher	Investigates patterns, libraries, and best practices before implementation
Coder	Writes production-ready code following established patterns
Reviewer	Audits code for correctness, security, and maintainability
Builder	Assembles final deliverables and deployment artifacts
Scribe	Creates documentation, changelogs, and user-facing content
Council	Multi-model consensus for complex decisions requiring diverse perspectives

Instructions

Step 1: Load Agent Persona

1. Identify which agent role is needed for the current task
2. Load the specific persona from ../../artifacts/agents-abilities.md
3. Confirm role boundaries and constraints

Step 2: Execute Within Role

1. Perform tasks adhering to the MECE principle
2. Stay within your role's defined scope
3. If a task belongs to another agent, explicitly delegate it

Step 3: Handoff

1. Once completed, pass output to the next logical agent in the workflow
2. Standard pipeline: Plan → Researcher → Coder → Reviewer → Builder
3. Include context and decisions made for downstream agents

Step 4: Quality Gate

1. Adhere strictly to the Global System Prompt in ../../docs/system-prompt.md
2. Ensure all output meets the closing block quality bar
3. Verify no scope creep into other agent roles

Constraints

• Do not perform tasks outside of your designated role
• If a task belongs to another agent, explicitly delegate it — do not attempt to solve it
• Adhere strictly to the Global System Prompt in ../../docs/system-prompt.md
• Never skip the Researcher phase before complex implementation work
• Never skip the Reviewer phase before shipping code
• All agent capabilities are defined in ../../artifacts/agents-abilities.md — reference it for detailed specs

Examples

Example 1: Feature Request

Trigger: "Build a user authentication system" Flow:

1. Plan → Break down into auth architecture, route design, component tree
2. Researcher → Investigate best practices for auth (NextAuth, session patterns)
3. Coder → Implement login, registration, session management
4. Reviewer → Audit for security (XSS, CSRF, password hashing)
5. Builder → Assemble final deployable artifact

Example 2: Bug Fix

Trigger: "Fix the broken checkout flow" Flow:

1. Researcher → Investigate the error pattern and related code
2. Plan → Design the fix with minimal scope
3. Coder → Implement the patch
4. Reviewer → Verify fix doesn't break other flows
5. Scribe → Write changelog entry