Unclear It is not clear which sessions include these instructions, possibly only remote or cloud ones.
What
Claude Code has new instructions for the model about commands it writes to run on your machine. They say:
- Your machine checks each command line before running it, and may ask you about, or refuse, a line it cannot fully read.
- Paths, names and ids should be written out literally rather than built from shell variables,
forloops or$(…). - Chaining commands with
&∨is fine. - The
$(cat <<'EOF'form for commit messages is fine.
Why
This should change how Claude writes shell commands, so fewer of them trigger a question or a refusal from your machine.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear which sessions include these instructions, possibly only remote or cloud ones.