SYSTEMX Agent 0 and Subagents

How Agent 0 Works With Subagents

Agent 0 is a coordination pattern, not a privileged process in the current template. One accountable lead—human or a supervised primary agent—breaks work into bounded lanes, assigns clear ownership, collects evidence, and decides what may be merged or deployed. Subagents are assistants with restricted scopes; they do not inherit release authority.

Operating rules

• Give every subagent a goal, permitted files/tools, time or token budget, and explicit completion evidence.
• Keep lanes disjoint. Do not let multiple agents edit the same surface without a coordinator and merge plan.
• Do not provide production secrets, broad credentials, or unnecessary customer data to agents or their tools.
• Treat all output as a proposed change. Independently review code, commands, documentation, and security implications.
• Require each lane to report scope, files changed, checks run, results, unresolved risks, and next action.
• Keep the human operator responsible for access, cost, compliance, Git history, and release approval.

Subagents multiply token use and tool calls. They can also multiply mistakes if the coordinator lacks a clear plan. Use them where parallel research or isolated implementation provides real value; do not use them as a substitute for an engineering owner. The project’s intake, status, and runbook files should be the durable record—not ephemeral agent conversation.