Claude Agent SDK adds a verbatim_prompts option to stop untrusted text triggering file reads or commands

Version 0.2.158 of the Claude Agent SDK for Python, released on 23 September 2026, adds a verbatim_prompts option (off by default). When enabled, user messages reach the CLI exactly as written, with no @path file expansion and no slash-command dispatch. This stops untrusted text inlined into prompts from triggering unintended file reads or command execution. It requires CLI 2.1.248 or later.

What it means If your agent puts fetched web pages, emails or user text into prompts, turning this on removes an easy injection route.