A trustworthy specialist invocation contract where specialists are stable role contracts invoked from claimed Mission Control tasks, selected through main-agent delegation plus policy validation, constrained by narrow permission envelopes, and required to write back both operator-readable and machine-readable results. Current direction: Specialists should be stable role contracts first, not loose agent identities. Start with one core specialist class, but design the UI and system so more can be added and managed later. The main agent should propose specialist invocation, while the app/helper layer validates routing against policy. Misroutes should stop for human review first, at least early on to build trust. Specialists should operate under narrow policy envelopes, return partial results with explicit claim decisions, and always write back both a task comment and a structured result envelope. Retries should be policy-driven, escalation should land in the task surface first, and follow-up work should begin as draft proposals rather than autonomous creation.

Define and build the artifact workspace as a first-class product subsystem, covering compact inline previews, modal or file-manager UX, upload plus pasted-link intake, metadata and CRUD flows, provenance and scope rendering, and the path to conversation attachments. Current direction: Draft generated from legacy content. Review and replace with an explicit current direction.

Added after Fatih asked whether the current LAN bind posture is a concern and then explicitly asked to track the cleanup as its own packet instead of losing it in chat.

Opened after Fatih said he would rather get a clear documented operator workflow later than try to reason through secret handling in the middle of setup.

Opened after Fatih liked the persistence whiteboard and asked for this to become a reusable Mission Control decision feature with choice buttons, confirm modal, and optional comment capture. Prototype: http://127.0.0.1:18793/whiteboard/persistence/