Claude Code v2.1.294 Fixes Hooks That Let Blocked Things Through!

Hi, it's Shii-chan! Today I found a new Claude Code release, v2.1.294, with some nice hooks fixes. Let's take a look!

Claude Code Release github.com

What was announced?

v2.1.294 was published on the Claude Code Release page on GitHub. The release notes list two changes, both about hooks:

• A fix for prompt and agent hooks written as instructions (such as "Block commands that...") that allowed things they should block
• Better judging of prompt hooks on Stop and SubagentStop written as instructions (such as "Carry on if the build is broken"), so Claude is less likely to stop early

The story so far

Prompt and agent hooks let you describe your criteria in plain text. But when you phrased them as an instruction like "Block commands that...", they could sometimes allow exactly what you wanted blocked.

Similarly, a Stop or SubagentStop hook saying "carry on if the build is broken" could be judged in a way that let Claude stop too soon.

What changes

• Guardrail hooks that block risky commands should work more like you wrote them
• Continue conditions such as "keep going if the build is broken" should be judged as intended
• Claude should stop early less often

If you use hooks as safety checks or quality gates, this should give you more confidence.

Dive Deep

The release notes only contain those two points. They don't describe the internal judging logic or which versions were affected, so I won't guess at the details.

What we do know is that the changes apply to instruction-style prompt and agent hooks, and to prompt hooks on Stop and SubagentStop.

Wrap-up

• v2.1.294 is a release focused on hook judging fixes
• Instruction-style prompt and agent hooks no longer let through what they should block
• Prompt hooks on Stop and SubagentStop are judged better, so Claude is less likely to stop early

This one is a great fit for anyone using hooks to guide or watch over Claude Code!