Task plans should disclose subagent delegation and synchronization #59

Problem

Task decomposition can remain too high-level when work is delegated to subagents. In particular, the published plan may say only that findings will be validated and fixed without describing which validations are delegated, why subagents are being spawned, whether their results are blocking or advisory, or when the parent will wait for them.

This makes execution harder to follow and can hide important coordination choices. For example, the parent may begin implementing changes before validator subagents return, causing duplicated investigation or changes based only on the parent's initial analysis.

Reproduction

1. Ask Diffex to handle a multi-part task, such as validating and resolving several independent review findings.
2. Observe a high-level plan such as “Validate findings and implement valid fixes.”
3. The parent spawns one or more validator subagents, but the plan does not expose the delegation or synchronization strategy.
4. The parent may start editing before those validators return, with no plan update explaining that their work is intentionally advisory and concurrent.

Expected behavior

When subagents are used, task decomposition should describe the material delegation and coordination boundaries. The plan should make clear:

• which task or finding each subagent owns;
• why delegation is useful;
• whether the parent will wait before acting or proceed independently in parallel;
• where subagent results will be incorporated or reconciled;
• any change to the plan when execution differs from the original synchronization strategy.

This does not require listing implementation-level details for every tool call. It should expose delegation whenever spawning subagents materially affects task execution, ordering, or validation.