[BUG] Dynamic context injection does not work as documented.

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

According to the docs (https://code.claude.com/docs/en/skills#inject-dynamic-context):

"Each !<command> executes immediately (before Claude sees anything)"

However this interaction with Claude Code 2.1.141 (Haiku) suggests otherwise:

❯ Print the latest user prompt and nothing else. # !`ls CLA*.md`

⏺ Print the latest user prompt and nothing else. # !ls CLA*.md

What Should Happen?

❯ Print the latest user prompt and nothing else. # !ls CLA*.md

⏺ Print the latest user prompt and nothing else. # CLAUDE.md

Error Messages/Logs

Steps to Reproduce

❯ Print the latest user prompt and nothing else. # !ls CLA*.md

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.141

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

iTerm2

Additional Information

Inline code injection is a much anticipated feature.