The problem
How it works
Anthropic released two ways to have Claude Code look for vulnerabilities. In the terminal, /security-review checks pending changes for problems such as injection, cross-site scripting, broken authentication and insecure data handling, and Claude can then be asked to fix what it found. In CI, the open-source claude-code-security-review GitHub Action runs on each pull request, analyses only the changed files, drops findings in categories it treats as noisy (such as denial of service or rate limiting) and posts inline comments with severity and suggested fixes. Teams can add their own scanning and false-positive instructions.
Results and limitations
Anthropic says it uses the action on its own code. Its examples: the action flagged a remote code execution flaw reachable through DNS rebinding in a local HTTP server for an internal tool, and an SSRF risk in an internal credentials proxy, both fixed before merging. These are vendor claims, not independent results.
Before you try it
The README says the action is not hardened against prompt injection and should review only trusted pull requests; require maintainer approval before workflows run on outside contributors’ changes. It needs an Anthropic API key enabled for Claude Code, stored as a repository secret. Treat it as an extra reviewer, not a replacement for a security audit, and expect some false positives and misses. Results depend on your setup and were not independently reproduced by Top 200 Repos.
Tools and repositories
- Tools
- Claude Code · GitHub Actions
- Prerequisites
- A GitHub repository and an Anthropic API key enabled for Claude Code; the /security-review command needs Claude Code installed.
Try it yourself: Ask for a focused security review before you commit
Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].
Review my uncommitted changes for security problems only. This is a [Type of application] that handles [Sensitive data it handles]. Pay special attention to [Areas to focus on]. For each finding give the file and line, how an attacker could exploit it, a severity and a minimal fix. Skip style issues and theoretical problems you cannot tie to a concrete exploit path. Do not change any code until I choose which findings to fix.Run it on a branch, and have a person confirm each finding before the fix is merged.
Related videos
About anthropics/claude-code
The SECRET Claude Code Upgrade That Has Everyone Going Nuts (Full Guide) (opens in a new tab)
The title names the project, the description links the project website.
Watch on YouTube (opens in a new tab)About anthropics/claude-code
The only Claude Code guide you'll ever need (Opus 4.5) (opens in a new tab)
The title names the project, the description links the project website.
Watch on YouTube (opens in a new tab)Independent creators. Listing a video implies no endorsement, sponsorship or partnership.
Sources
- Automate security reviews with Claude Code (opens in a new tab)
Vendor post · claude.com · Published 6 Aug 2025 · Source checked 12 Nov 2026, 10:07 UTC
What this source supports
- Introduced the /security-review command and a GitHub Action for automated security reviews
- The command checks for SQL injection, XSS, authentication flaws, insecure data handling and dependency issues, and Claude can implement fixes
- The action reviews new pull requests, filters false positives and posts inline comments
- Anthropic says the action caught an RCE via DNS rebinding and an SSRF vulnerability in its own code before merge
- anthropics/claude-code-security-review (opens in a new tab)
Project documentation · github.com · Undated · Source checked 12 Nov 2026, 09:55 UTC
What this source supports
- The action is diff-aware, comments on PRs and applies false-positive filtering
- It excludes categories such as denial of service, rate limiting and open redirects by default
- Not hardened against prompt injection; use only on trusted PRs with approval required for external contributors
- Needs an Anthropic API key enabled for Claude API and Claude Code; MIT licence