2 min read AI-generated

Claude Code 2.1.172 – Agents That Spawn Agents

Copy article as Markdown

With version 2.1.172, sub-agents in Claude Code can now spawn their own sub-agents – up to five levels deep. Plus a plugin search bar and smarter model and region handling.

Featured image for "Claude Code 2.1.172 – Agents That Spawn Agents"

Claude Code got two releases in the last few days: 2.1.172 and, shortly after, the bugfix release 2.1.173. The most interesting feature hides in the 2.1.172 changelog as a single line – and still has the potential to reshape entire workflows.

Nested sub-agents

Sub-agents can now spawn their own sub-agents – up to five levels deep. Sounds like a small thing, but it’s a real architectural update.

The reason: context management. When an agent breaks a big task into subtasks and delegates each to its own sub-agent, the main context stays clean. Now a sub-agent can do the same thing again – split its subtask further and hand it off to its own helpers. That lets a complex project decompose into a tree structure without any single context overflowing.

This is exactly the kind of feature you underestimate until you need it. Anyone who has tried to push a large refactor through a single agent knows the problem: at some point it loses the thread.

Plugin search and model logic

Alongside that, there are solid day-to-day improvements. The /plugin browser for marketplaces now has a search bar – finally, because plugin lists have grown a lot over the past months.

Amazon Bedrock now reads the AWS region from your ~/.aws config when AWS_REGION isn’t set. And Claude Code retries a turn once on the fallback model when the API returns an unexpected, normally non-retryable error. Small things that remove friction.

2.1.173 right after

The follow-up release 2.1.173 clears two snags: Fable 5 model names with a [1m] suffix are now normalized correctly – Fable 5 ships with 1M context anyway, so the suffix is redundant. And on Windows, a spurious warning about missing sandbox dependencies disappears.

My take: Nested sub-agents are the feature everyone will be talking about in a few months – even though today it’s just one changelog line. Agents orchestrating agents are the logical next step if you actually want to automate large tasks. Five levels deep sounds like a lot, but this is exactly where it gets decided whether a nice coding assistant becomes a real tool for complex projects. I’m definitely going to try this.

Sources: Claude Code Changelog, Releasebot: Claude Code