Feature Request: Setting to always use subagents for debugging

When Claude Code encounters a failing test or bug, it attempts to diagnose the issue itself using its accumulated conversation context. This is a liability — the context creates assumptions that compound over attempts and prevent Claude from seeing the actual problem.

A subagent launched with only the minimal context needed to examine a failure consistently outperforms Claude's own diagnosis because it has no prior assumptions.

Feature request: Add a user-configurable setting in settings.json that instructs Claude Code to always delegate debugging to a subagent rather than attempting to diagnose failures itself. This is the kind of setting a user would enable once and never turn off — so a command is unnecessary overhead.

Real example: While writing Jest tests for AWS Lambda code, variables were not returning expected values for mocked API calls. Claude made multiple failed attempts, each building on flawed assumptions. When a subagent was launched with only the minimal context needed, it immediately identified the root cause that Claude had repeatedly missed.