Skip to content
Book a callSkip the hiring grind

Day 19 of 52 · about 20 min · Subagents

Delegate research and review to subagents

Every file Claude reads to answer a side question stays in your conversation and crowds out the task. Send the digging to subagents, keep their short answers, and let a fresh one check the work before you call it done.

You’ll finish with

  • Research answered by Explore, with only its summary in your conversation
  • Two questions answered in parallel by separate subagents
  • A review step that checks a change against what you asked for

You’ll need

  • The Day 18 setup: Claude Code in a git project with the diff-reviewer subagent
  • Subagents spend from the same plan usage as your session, so ask for short answers
Start from the practice project as of Day 18 (.zip, 27 KB)

Before you start

Tell us about your setup and the steps change to fit it. Your answers stay in this browser, and you can change them at any time.

Which system are you on?

Commands differ between systems. Pick the one you’ll run Claude Code on.

The steps

One step at a time. When a step’s check matches, tick it and the next one opens.

0 of 3 steps done0%

  1. Step 1 of 3: Send a research question to Explore

    Let the built-in read-only subagent do the searching and hand back a summary.

    Explore can read and search but not edit. It skips your CLAUDE.md to stay fast, so put any rule it needs in the question.

    In Claude Code

    You type: Use the Explore subagent to find where the discount is applied and which tests cover it. Answer with file paths and line numbers only.

    What happens: Explore reads the files in its own context. You get the short answer.

    Ask for the shape of answer you want back.

    Check it worked

    A short answer naming src/quote.js and test/quote.test.js in the practice project, with no file contents pasted into your conversation.

  2. Step 2 of 3: Ask two questions at once

    Give two independent questions to two subagents and get one summary from each.
    Paste into Claude Code
    Use two subagents in parallel. One lists every place VAT is calculated. The other lists every error message the quote command can print. Give me a short summary from each.

    In order:

    1. You

      You ask two questions

    2. Claude

      Claude starts two subagents

      Each in its own context

    3. Tool

      They search side by side

    4. Claude

      Two summaries come back

      Claude combines them for you

    Parallel works when neither question depends on the other.

    Check it worked

    While they run, two rows sit in the panel below the prompt and /tasks lists them. Then you get both answers.

  3. Step 3 of 3: Check a change in a fresh context

    Before you accept a change, have a subagent compare the diff with what you asked for.

    Use the uncommitted change from Day 18, or ask Claude for a new small one. Then name the task the reviewer checks against.

    Paste into Claude Code
    Use the diff-reviewer subagent to check my uncommitted changes against this task: "add a --rush option that adds 25% to the subtotal before the discount". Report only gaps that affect correctness or the task.

    Same session checks

    • Has the reasoning that produced the change
    • Can be biased toward code it just wrote

    Fresh subagent checks

    • Sees only the diff and your criteria
    • Findings come back to the session that can fix them
    Same review, different starting point.

    Check it worked

    Findings tied to the task, or "No blocking issues". Ask Claude to fix the real ones, then run the review again.

    Note: A reviewer asked for gaps usually finds some. Fix what breaks the task and treat the rest as optional. For a plain bug hunt, /code-review runs a fresh review in one command.

Finish · 3 of 3 steps to go

Your session keeps the task, subagents do the digging

Use Explore for lookups, parallel subagents for unrelated questions, and a fresh reviewer before you accept work.

What you’ll have checked

  • Explore answered without pasting files into your conversation
  • Two subagents answered in parallel
  • A reviewer checked the diff against the task

Lesson updates by email

Get lesson updates by email

Leave your email and we’ll tell you when we add or update a lesson.

We’ll email you when we add or update a lesson. We use your email only for this. Privacy policy

Claude Code for your team

Want Claude Code set up across your team?

HorizonLux sets up and repairs Claude Code for teams: CLAUDE.md rules, hooks that enforce them and reviewed MCP servers. Book a free call to talk it through.

Book a free call Claude Code agency

Related

Free tool

AI Agent Cost Estimator

Free tool

Token Counter

How we build

AI software factory

Questions

Questions about Day 19

Can I ask Explore to continue where it stopped?

No. Explore is one-shot and can’t be resumed. Use a custom subagent when you want to continue the work later.

Sources

Checked against the official Claude Code docs on .

Independent resource, not affiliated with Anthropic. HorizonLux wrote this lesson; Claude Code is made by Anthropic.