glossary · Running it

Child tasks

definition

Child tasks are pieces of work split off from a parent task, each with its own workflow, where children that share an order start together and the parent waits until all of them are done.

in one line: Split the work. Same order starts together. The parent waits.

in context

Where do child tasks fit?

The process stays a line. The work fans out. Children with the same order start together. The parent waits for all of them.

Data modelorder 1
Login designorder 1
Wire the floworder 2, after both
Example. Orders 1, 1, 2: two children start together, the third after.
two ways to say it

What are child tasks, in plain words?

Same idea at two depths: the plain version, then what the engine actually does.

A workflow has one current step. When work is parallel, split the work instead of the process.

Each child can run in its own session, with its own agent.

for engineers

wait_for_children and sequential_children gates hold the parent. Same order activates together.

next_actions lists one action per active child. Results are not merged; the parent reads the tasks.

the longer answer

Why do child tasks matter?

Some work is naturally parallel: a feature needs a data model, a screen design and documentation, and none of them depends on the others. Inside a single workflow that is awkward, because a workflow is a line with one current step. The clean answer is to keep the process a line and split the work instead. The parent task carries the plan; each child carries one piece with the workflow that fits it.

Ordering is what makes this usable. Children grouped in the same order can start at the same time. A later order starts only when the earlier one is finished, so "build the data model and the design together, then wire the login flow, then write the docs" is expressed as orders 1, 1, 2 and 3. The parent reaches a waiting step and does not continue until all of its children are complete.

Child tasks are also how different agents and runners share one piece of work. Each child can run in a different session, with a different agent, or on a different executor, while the parent keeps the overall picture.

in convops

How do child tasks work in ConvOps?

In ConvOps, a parent task's workflow can carry a gate that waits for its children, either waiting for all of them or activating them order by order. Children with the same order activate together, and the response to the agent lists one next action per active child, so your orchestrator, or several people, can start them at once. Each child keeps its own workflow, gates and history. Steps inside any one workflow still run one at a time; the parallelism lives at the task level. The engine does not merge child results; the parent continues once every child is complete, reading what they produced from the tasks and the shared memory.

questions

What do people ask about child tasks?

How do child tasks run in parallel?

Children with the same order start together; a later order starts when the earlier one is finished. The parent waits at a step until its children are complete. The parallelism lives at the task level, not inside one workflow.

Can child tasks use different agents?

Yes. Each child keeps its own workflow, gates and history, and can run in a different session, with a different agent, or on a different executor.

Does ConvOps merge the results of child tasks?

No. The parent continues once every child is complete and reads what they produced from the tasks and the shared memory.