Goal

Introduce internal agent-definition support for non-user-routable subagents such as memory curation, classification, and other platform-owned helper roles.

Why

Netclaw already has subagent runtime primitives, but internal roles still rely on ad hoc in-code prompt/tool wiring. The memory redesign needs a safe, explicit way to define internal subagents without depending on user-facing reminder or task-selection UX.

Scope

• Define an internal agent-definition model for platform-owned subagents.
• Support fixed prompt, tool subset, model role, and execution policy.
• Keep internal definitions separate from user-selectable agent profiles.
• Add observability and validation for internal agent-definition resolution.

Acceptance criteria

• Platform-owned subagents can be declared through a first-class internal definition path.
• Internal agent definitions can specify prompt, tools, and model role.
• User-facing agent-definition UX is not required for this work.
• The runtime can resolve internal definitions without leaking them into normal skill routing.

Related

• PR spec: define policy-aware SQLite memory redesign #180
• Issue Reminder UX follow-up: allow selecting agent definition for execution #147
• Milestone: Memory And Agent Foundations