This guide shows the minimal steps required to install, configure, and begin using the multi-agent system.
No explanations. No theory. Just the fastest way to get running.
npm install -g opencodeOr update:
opencode upgradegit clone <your-repo-url>
cd <your-repo>Make sure these exist:
AGENTS.md
.opencode/
agents/
system/
agenttasks-templates/
agenttasks/
memory/
skills/
If the structure is missing, copy the template from your starter repository.
If using MCP servers:
cat opencode.jsoncEnable or disable MCP integrations as needed.
opencode .This loads:
- all agents
- agent definitions
- Skills
- Memory
- workflows
- AgentTask templates
Inside OpenCode:
@Project-Manager helpIf you see the help text, everything works.
Example:
@Project-Manager Lets build something cool!This will:
- analyze complexity
- create the task file under
.opencode/agenttasks/ - embed relevant Skills
- reference Memory if available
- assign it to the correct agent
Once an AgentTask is created:
@Developer implement .opencode/agenttasks/<task-file>.mdDeveloper will:
- read the task
- search Memory
- load required Skills
- implement deliverables
- run validations
- produce a completion summary
Create a new folder:
.opencode/skills/<skill-name>/
Add:
metadata.yamloverview.mdpatterns/(optional)workflows/(optional)
The system picks it up automatically.
Memory entries are created automatically when AgentTasks require it.
To add manually:
.opencode/memory/<category>/<entry>.md
Follow the schema in MEMORY.md.