Swarm Executor FAQs

Question 1

Can I stop and resume a swarm execution later?

Accepted Answer

Yes. The executor maintains state in a .tasks.json file and leverages Claude Code's native task system to track progress, allowing you to resume interrupted executions without losing track of completed work.

Question 2

Does this skill work with any markdown plan?

Accepted Answer

It is designed to work with structured markdown plans that follow the ### T<id>: format, which allows it to accurately parse dependencies, file scopes, and acceptance criteria for subagent dispatch.

Question 3

What happens if there is a merge conflict between subagents?

Accepted Answer

It employs a tiered resolution strategy: first attempting additive merging for non-conflicting lines, then using a specialized resolver subagent to logic-check the merge, and only escalating to the user if automated tools cannot guarantee a correct resolution.

Question 4

How does Swarm Executor handle parallel code changes safely?

Accepted Answer

The skill uses git worktree isolation for every subagent, ensuring each task is developed in a separate, clean environment. Changes are only merged back into the main branch once the task is complete and validated.

Question 5

How does it decide which tasks to run first?

Accepted Answer

The executor analyzes the 'depends_on' field for each task in your plan to create a dependency graph. It then groups all 'unblocked' tasks into waves, executing them in parallel until all prerequisites for the next wave are met.

Swarm Executor

主要功能

使用场景

Swarm Executor

主要功能

使用场景