Three Layer Defense Architecture
Anthropic has introduced a free security plugin for its Claude Code terminal tool that reviews code edits, model outputs, and commits as developers work. The plugin operates across three distinct checkpoints designed to catch threats at different stages of a coding session. At the file edit level, a fast deterministic pattern match flags dangerous constructs such as eval(), new Function(), os.system(), child_process.exec(), pickle deserialization, and DOM injection vectors like dangerouslySetInnerHTML. This initial layer requires no AI inference and adds zero usage cost.
At the end of each conversational turn, a separate Claude model reviews the full git diff of all changes made during that session. This reviewer starts from a fresh context without prior investment in the original approach, catching logic level vulnerabilities that string matching cannot detect. These include authorization bypass, insecure direct object references, server-side request forgery, and weak cryptography implementations.
Impact and Adoption
When Claude commits or pushes code using its Bash tool, a deeper agentic review reads surrounding callers, sanitizers, and related files to minimize false positives. Internal testing showed the plugin reduced security related comments on pull requests by 30 to 40 percent. The plugin uses Claude Opus 4.7 by default for both end-of-turn and commit reviews, though developers can configure alternative models through environment variables.
The plugin is available to all Claude Code users on every plan and can be installed from the plugin marketplace. Industry leaders including J.P. Morgan’s Shalini Goyal have praised the approach for embedding security guidance directly into the coding session rather than relying on post-development reviews.
Source: Cyber Security News
