Anthropic has launched a new plugin for its Claude Code platform called Security Guidance. The new tool automatically checks the code while the AI agent is running, detecting errors and potential threats.
According to the company, the plugin works actively directly during the session, without installing additional tools or special “skills”. While editing files, the system analyzes the code in the background, identifies typical errors inherent in AI models, and provides suggestions for their correction.
When a commit is created, the plugin tries to detect deep security vulnerabilities by examining the broader context of the project.
Users can change the inspection levels and also add their own analysis rules. It is also possible to disable one of the scanning stages if necessary.
Anthropic said that after the implementation of this plugin within the company, the number of pull requests (PR) related to security vulnerabilities has decreased by 30-40%.
The Security Guidance plugin is now available in the “marketplace” section of the Claude Code platform. Installation is carried out with the following command:
