You and Claude spend twenty minutes narrowing a failure to a race condition in one module. You then ask Claude to delegate the fix to a subagent, and the subagent comes back asking which failure it is meant to be fixing. Why?

Mediumclaude-codeworkflowssubagentscontext

Last reviewed 21 Aug 2026

It will pick the diagnosis up on a second call, once the first has primed it
It can see the discussion but has not been told which part of it matters
It never had the discussion, so its prompt has to carry the diagnosisCorrect answer
It has no working directory of its own, so the module path cannot be resolved

Show answer & explanation

Correct answer: C. It never had the discussion, so its prompt has to carry the diagnosis

Explanation

The principle — the most common failure in delegation is not the worker doing the job badly. It is the brief assuming knowledge the worker was never given.

Why the key is correct — a subagent starts with a fresh, isolated context window and does not see your conversation history, the skills already invoked, or the files Claude has already read. Twenty minutes of joint diagnosis lives entirely in the window you are steering, so a delegated worker has no access to any of it. That makes the prompt the only channel, and the diagnosis has to travel through it.

Why the others are wrong — the worker is not failing to pick the relevant part out of the discussion, because the discussion is not there to pick from. A second invocation does not help: each invocation creates a new instance rather than continuing an earlier one, so it begins exactly as blind as the first, and repeating an under-specified request returns the same question. And the working directory is not the problem; a subagent starts in the main conversation's current working directory, so paths resolve as they normally would.

Remember this — write the brief for a competent colleague who has just walked in, because that is literally the situation.

Sources — Anthropic's Claude Code subagent documentation and glossary.

Sources

Primary sourceCreate custom subagents — Anthropic · accessed 17 Aug 2026
“Each subagent starts with a fresh, isolated context window. It doesn't see your conversation history, the skills you've already invoked, or the files Claude has already read.”
Primary sourceGlossary — Anthropic · accessed 17 Aug 2026
“A specialized AI assistant that runs in its own context window with a custom system prompt, specific tool access, and independent permissions. It works on a delegated task and returns a summary to the main conversation.”
Primary sourceCreate custom subagents — Anthropic · accessed 19 Aug 2026
“Each subagent invocation creates a new instance rather than continuing an earlier one.”
Primary sourceCreate custom subagents — Anthropic · accessed 17 Aug 2026
“A subagent starts in the main conversation's current working directory.”

Practise 10 questions on this topic

Take Claude Code Workflows — Timed Test 1 (10 questions) — scored instantly, explanation for every question, no login.

Start the test

You and Claude spend twenty minutes narrowing a failure to a race condition in one module. You then ask Claude to delegate the fix to a subagent, and the subagent comes back asking which failure it is meant to be fixing. Why?

Mediumclaude-codeworkflowssubagentscontext

Last reviewed 21 Aug 2026

It will pick the diagnosis up on a second call, once the first has primed it
It can see the discussion but has not been told which part of it matters
It never had the discussion, so its prompt has to carry the diagnosisCorrect answer
It has no working directory of its own, so the module path cannot be resolved

Show answer & explanation

Correct answer: C. It never had the discussion, so its prompt has to carry the diagnosis

Explanation

The principle — the most common failure in delegation is not the worker doing the job badly. It is the brief assuming knowledge the worker was never given.

Remember this — write the brief for a competent colleague who has just walked in, because that is literally the situation.

Sources — Anthropic's Claude Code subagent documentation and glossary.

Sources

Primary sourceCreate custom subagents — Anthropic · accessed 17 Aug 2026
“Each subagent starts with a fresh, isolated context window. It doesn't see your conversation history, the skills you've already invoked, or the files Claude has already read.”
Primary sourceGlossary — Anthropic · accessed 17 Aug 2026
“A specialized AI assistant that runs in its own context window with a custom system prompt, specific tool access, and independent permissions. It works on a delegated task and returns a summary to the main conversation.”
Primary sourceCreate custom subagents — Anthropic · accessed 19 Aug 2026
“Each subagent invocation creates a new instance rather than continuing an earlier one.”
Primary sourceCreate custom subagents — Anthropic · accessed 17 Aug 2026
“A subagent starts in the main conversation's current working directory.”

Practise 10 questions on this topic

Take Claude Code Workflows — Timed Test 1 (10 questions) — scored instantly, explanation for every question, no login.

Start the test

You and Claude spend twenty minutes narrowing a failure to a race condition in one module. You then ask Claude to delegate the fix to a subagent, and the subagent comes back asking which failure it is meant to be fixing. Why?

Related questions

You and Claude spend twenty minutes narrowing a failure to a race condition in one module. You then ask Claude to delegate the fix to a subagent, and the subagent comes back asking which failure it is meant to be fixing. Why?

Related questions