Follow Discord
Sweep 22 Oct 2026 · 20:42Z Build v2.1.309 528 read Stable v2.1.308 Latest v2.1.308 Next v2.1.309 Feeds RSS JSON llms.txt llms-full.txt Unofficial
Reading a new release v2.1.309 Downloading and diffing · 2/6 0 findings $0.00 so far

Claude Code v2.1.307 ·

Command-init instructions warn that subagents sound sure when they are wrong

In the command-init prompt, the Agent tool's description now warns that subagent mistakes sound as confident as their findings

You'll notice Improvements
JSON All of v2.1.307
You'll noticeTier: how much it should matter to you
1Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
SubagentsArea: what it touches
ImprovementsKind: in v2.1.307,
ImprovementsSection of the release
What

The Agent tool lets Claude hand a task to a subagent, a separate helper that returns a summary. Its description in the command-init prompt, used by the command-line tool and the SDK, has been rewritten.

The old text mainly listed when to delegate. The new text warns that a subagent's mistakes come back sounding just as confident as its findings. It also warns that a subagent given Claude's own guess tends to report that guess as confirmed.

The standard Agent tool description is unchanged.

Why

These are instructions Claude follows, so they may make Claude more careful about trusting a subagent's answer. That effect has not been measured.

See this entry in the whole of v2.1.307 →

Feedback