Problem
Coordinating several task agents requires repeatedly checking their conversations, interpreting status changes, and deciding what should happen next. A persistent workspace conversation could provide that coordination while preserving the workspace's existing Kanban workflow, execution profiles, review steps, and human approvals.
Proposed solution
Add a workspace coordinator with a reusable global role and a persistent conversation. Each workspace assignment chooses its execution profile, execution environment, and local context. The coordinator can create or follow ordinary Kanban tasks, receive task-completion callbacks, and report progress in the same conversation.
An optional automation destination could deliver a scheduled prompt to that conversation, with an explicit distinction between prompt delivery and completed work.
There is a local prototype. Before preparing a contribution, I would like maintainer guidance on whether this should extend Office or remain a separate experimental feature sharing core task, session, and runtime infrastructure. A small sequence of focused PRs would be preferable to submitting the entire prototype at once.
Affected area
Workflow/queue, Web UI, Backend API, Agent lifecycle.
Who needs this?
Individual developers coordinating multiple task agents.
Target workflow
- Define a role such as “Project coordinator” with generic instructions.
- Assign it to a workspace using an existing execution profile and executor.
- Open its persistent conversation and ask, “Summarize the open tasks and suggest the next step.”
- Follow the coordinator's linked Kanban tasks and review its progress reports.
- Optionally route a daily project-summary automation to the same conversation.
Alternatives considered
- A regular task with task-management MCP tools: uses existing primitives, but lacks a dedicated workspace assignment and conversation entry point.
- Office: provides related scheduling and agent-management capabilities. Maintainer guidance is requested on reusing or extending its UX and ownership model.
Acceptance criteria
- Existing Kanban workflows, profile choices, review gates, and approval requirements remain authoritative.
- Workspace assignments have persistent conversations accessible on desktop and mobile.
- Role edits apply to assignments without copying or moving conversation history.
- Scope and ownership checks protect conversations and task access.
- Completion callbacks and scheduled delivery do not duplicate work or replay uncertain external actions after restart.
- Disabling the feature stops new execution and retains stored configuration.
- Migration and regression checks pass against current main before release.
Risks and constraints
This is an architectural proposal, not a production-ready patch. The prototype needs integration with current main, a focused contribution boundary, and separation from unfinished assistant work. No claim of enforced read-only agent execution is made.
The accompanying media uses a fresh disposable database, fictional project records, generic prompts, and a scripted local demo provider. It contains no personal conversation history or real provider credentials.
Synthetic prototype demo
These captures use fictional Garden Notes project data, generic example prompts, and scripted replies from a local demo provider. They demonstrate the prototype's interface; the example automation is not scheduled.
Mobile conversation
Workspace assignment: desktop and mobile
Execution profile and workspace context: desktop and mobile
Shared role settings: desktop and mobile
Automation destination: desktop and mobile
References
Before submitting
Problem
Coordinating several task agents requires repeatedly checking their conversations, interpreting status changes, and deciding what should happen next. A persistent workspace conversation could provide that coordination while preserving the workspace's existing Kanban workflow, execution profiles, review steps, and human approvals.
Proposed solution
Add a workspace coordinator with a reusable global role and a persistent conversation. Each workspace assignment chooses its execution profile, execution environment, and local context. The coordinator can create or follow ordinary Kanban tasks, receive task-completion callbacks, and report progress in the same conversation.
An optional automation destination could deliver a scheduled prompt to that conversation, with an explicit distinction between prompt delivery and completed work.
There is a local prototype. Before preparing a contribution, I would like maintainer guidance on whether this should extend Office or remain a separate experimental feature sharing core task, session, and runtime infrastructure. A small sequence of focused PRs would be preferable to submitting the entire prototype at once.
Affected area
Workflow/queue, Web UI, Backend API, Agent lifecycle.
Who needs this?
Individual developers coordinating multiple task agents.
Target workflow
Alternatives considered
Acceptance criteria
Risks and constraints
This is an architectural proposal, not a production-ready patch. The prototype needs integration with current main, a focused contribution boundary, and separation from unfinished assistant work. No claim of enforced read-only agent execution is made.
The accompanying media uses a fresh disposable database, fictional project records, generic prompts, and a scripted local demo provider. It contains no personal conversation history or real provider credentials.
Synthetic prototype demo
These captures use fictional Garden Notes project data, generic example prompts, and scripted replies from a local demo provider. They demonstrate the prototype's interface; the example automation is not scheduled.
Mobile conversation
Workspace assignment: desktop and mobile
Execution profile and workspace context: desktop and mobile
Shared role settings: desktop and mobile
Automation destination: desktop and mobile
References
Before submitting