Skip to content

Agents

@maestria/opencode provides one manager agent (@orchestrator) and 7 specialist subagents. Each agent has a focused domain, clear methodology, and specific permissions.

The @orchestrator decomposes complex tasks into atomic units, delegates to specialists via task(), integrates results, and verifies completion. Specialists handle one domain each — exploration, design, implementation, debugging, planning, review, or documentation.

AgentModeRoleKey Permissions
@orchestratorallManager — delegates, never implementsedit: deny, task: * allow
@adventurersubagentCodebase reconnaissanceedit: deny, bash: * ask
@architectsubagentArchitecture decisionsedit: deny, webfetch: allow
@buildersubagentFocused implementationedit: allow, bash: * ask
@diagnosesubagentSystematic debuggingedit: ask, bash: * ask
@plannersubagentImplementation plansedit: ask, webfetch: allow
@reviewersubagentCode reviewedit: deny
@writersubagentDocumentationedit: allow

A typical multi-step task follows this flow through the agent system:

@orchestrator (manager)
├─ @adventurer → Explore codebase, map dependencies
├─ @architect → Design approach, evaluate options
├─ @planner → Create phased implementation plan
├─ @builder → Implement changes
└─ @reviewer → Review for correctness and quality

Each agent hands off to the next. The orchestrator integrates results and verifies completion.