action-boundary

Agent boundary enforcement requires declaring explicit mandates in agent definitions so autonomous agents stay in their lane. Agents flag out-of-scope issues in a Notes section instead of modifying files outside their defined mandate.

Read-only behavior for reviewer agents means planners and reviewers never modify files during multi-agent planning. They strictly observe the plan, flag boundary violations, and report uncertainties without taking implementation action.

Enforcing role boundaries to stop cross-concern leakage requires agents to follow the plan exactly rather than instincts. Agents note uncertainties and avoid adding features outside the declared plan, ensuring actions stay within defined mandates.

No, agents should not report issues outside their mandate. The boundary enforcement protocol dictates that if you notice an out-of-scope issue, you stay in your lane because a separate agent with that specific mandate will catch it.

The best way to prevent scope creep is to declare explicit boundaries in the agent definition stating exactly what the agent does and does not do. Agents must honour this declaration exactly and follow the plan without adding features.