AI BriefingClaude Code Prompt Update
Key point
The Claude Code system prompt version 2.1.157 has updated its security monitoring and tool use guidance.
Details
The agent prompt feature has been strengthened, expanding the scope of security monitoring for autonomous agent behavior. High-risk review is performed for persistent configuration changes, external data submission, new destinations, and low-information behaviors where intent is inferred through the agent's narration.
Tool use guidance has been added. In particular, to improve tool-calling performance for the latest Opus model, it is recommended that tool descriptions specify when each tool should be called.
Also, an Opus 4.8 migration guide has been included. It advises including tool trigger instructions not only in the system prompt but also in each tool's own description.
The new EnterWorktree tool allows switching, via a path, from an existing worktree session or a pinned agent to another registered .claude/worktrees/ worktree.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.