by Corepass

Use Remy

Use subagents

Divide work, follow progress, and review the combined results.

Scope: The steps below use a local task in the desktop app.

Subagents handle parts of a task, with the results brought together in the main chat. Remy can call them when needed, or you can explicitly ask for them in your prompt.

1. Choose work that can be divided

Useful examples include reading different modules, reviewing a set of changes from different angles, and gathering evidence from separate sources.

A simple edit can usually stay in the main chat. When dividing work, specify what each subtask should examine, the questions it should answer, and what it should deliver.

2. Ask for subtasks

For example, send this to check whether the frontend and backend handle sign-in errors consistently:

Example prompt
Use two subagents to examine this project's sign-in flow.

Subtask one: Read the frontend sign-in form, request parameters, and error messages. List the relevant files and findings.
Subtask two: Read the backend sign-in endpoint, parameter validation, and error responses. List the relevant files and findings.

Both subtasks must analyze only. Do not edit files, install dependencies, commit, or push code.
When they finish, compare their findings in the main chat. List inconsistencies, the supporting evidence, and the suggested order of fixes.
For anything that lacks sufficient evidence, explain what still needs checking.

After sending, check the execution record to see whether subtasks were actually created. A statement that work will be delegated does not by itself mean a subagent has started.

3. Follow progress and results

  1. Find the subtask entry in the main chat.
  2. Click an available subtask title to open its instructions, execution, and results in the right panel.
  3. Return to the main chat to review the combined findings and remaining work.

Even after the main chat returns text, check whether any subtasks are still running. One finished subtask does not mean the whole task is complete.

4. Change the scope or stop a subtask

To change direction, give explicit instructions in the main chat. For example:

Example prompt
Continue checking sign-in failure messages only. Stop analyzing registration and password recovery.
Adjust the remaining subtasks, and distinguish completed checks from stopped work in the final summary.

If a running subtask offers Stop, hover over its entry and click the control. Wait for its status to update and confirm that execution has ended.

Stopping does not automatically undo file edits already made. Check the current files and existing results before continuing.

5. Review the final work

After the main chat brings the results together, check:

  • Whether each subtask finished, failed, stopped, or lacked necessary material.
  • Whether conclusions point to actual files, source material, or execution results.
  • Whether findings conflict and the main chat explains any differences.
  • Whether code changes were checked with relevant tests and actual behavior.

Agreement between subagents still needs supporting evidence. If you ask them to edit code, assign clear file scopes and review the combined changes. Do not assume parallel edits will resolve conflicts automatically.

6. Keep exploring