Task Threads & Follow-Ups

Task threads let you keep working with a task after the first execution. Instead of starting over with a new prompt, you can send a follow-up that continues from the task's existing context, output, status, worktree, selected model, assigned agent, and prior conversation.

Use follow-ups when the task is still the right unit of work but needs clarification, correction, a second pass, or review feedback applied.

What Users See

Surface	What It Means
Task thread	The conversation around one task, including the original prompt, model output, and follow-up messages.
Follow-up input	A way to continue the same task without creating another board card.
Queued follow-up	The message has been accepted but is waiting for execution capacity.
Streaming response	The follow-up is actively running and appending new output to the thread.
Changes update	New follow-up changes can refresh diffs and review state.
Alerts	Failed follow-ups or tasks needing more input can create attention notices.

When A Follow-Up Triggers

A follow-up is useful after a task has produced enough context to continue from. Common triggers include asking the agent to address review comments, refine an implementation, explain a failure, add tests, resolve a small issue discovered during review, or continue a task that explicitly asked for more direction.

Channel-created tasks can also receive follow-up context early. If a Slack, Telegram, Discord, or Email reply arrives before the first execution exists, OpenVibely keeps it with the task so the instruction is available when the task starts.

Follow-ups are different from Chat orchestration. Chat is the central place to coordinate many tasks. A task follow-up is scoped to one existing task thread.

Runtime Boundary

Task follow-ups are execution work, so they respect worker capacity. If project or model capacity is full, OpenVibely accepts the follow-up and queues it instead of dropping the message.

Interactive project Chat intentionally stays responsive even when task workers are full. That is why Chat can keep orchestrating while task follow-ups wait for available execution slots.

Task-thread streams reconnect from their last received output offset and replay durable output before finalizing a turn. A brief browser or gateway disconnect should not require resending the follow-up; wait for catch-up first.

Context Rules

Rule	Why It Matters
Continue the existing task	The follow-up uses task thread context instead of acting like a brand-new task.
Preserve execution profile	The selected model, effective agent context, and task mode remain available.
Re-run supporting hooks	Lifecycle behavior such as memory recall and skill routing can still support the new turn.
Keep worktree orientation	Repository-backed follow-ups continue from the task's worktree when available.
Update review state only when needed	A follow-up that creates new changes can make review pending again; a read-only follow-up should not disturb merged state.

Examples

User Intent	Better As
"Fix the test failure in this task's changes."	Task follow-up
"Apply my review comment to the implementation you just made."	Task follow-up
"Create three separate approaches and run them in parallel."	Chat Orchestrate
"This completed task should now update the docs too."	Task follow-up or a chained child task, depending on whether it is the same unit of work
"Split this feature into backend, frontend, and tests."	Chat Orchestrate or task chaining

Steering From A Task Thread

When a follow-up is queued and the task thread shows a pending input row, a Steer button appears on that row. Steering from a task thread works the same way as Chat steering: it redirects the active turn rather than queuing behind it.

Use task-thread steering when you have already sent a follow-up but want to correct or narrow it before the worker picks it up. You can also cancel a queued follow-up before it is applied. If the turn has already started, the steer may be rejected and you should send a new follow-up instead.

Stop An Active Follow-Up

While a follow-up response is running, the send control becomes Stop response. Stopping cancels the active model run and records a cancellation state in the thread. It does not cancel other queued follow-ups or the entire task; use the queued-row cancel action or task cancel control for those separate intents.

What It Does Not Do

It does not bypass worker capacity for task execution.
It does not create a new task card unless you ask Chat or task actions to create one.
It does not make every attachment or temporary prompt detail durable memory.
It does not automatically mean generated changes are safe to merge.

Page	Why It Matters
Tasks	The task detail screen contains the thread and follow-up controls.
Prompt Queue & Steering	Explains queued prompts and active-turn steering in Chat.
Worker Capacity & Dispatch	Follow-ups wait for task execution capacity.
Task Diffs & Review	Follow-up changes can update the Changes tab.
Git Worktrees & Merge Safety	Repository-backed follow-ups continue in isolated task worktrees.