Claude Security plugin officially open beta! This is not just another scanning tool, but a direct integration of 'cybersecurity researcher level' capabilities into the Claude Code terminal. It will read your architecture like a real person, track cross file data streams, establish threat models, and then use multiple agents to verify each other, identifying complex logic vulnerabilities and authentication bypasses that traditional static scanners often miss. After scanning, you can directly generate patch suggestions that match your programming style, which can be applied after human review. What does this represent? AI is shifting high-level security to the daily workflow of every developer. No need to wait for the security team to schedule, no need to throw the code to the cloud, you can check it yourself before submission. In the era of AI accelerated attacks, this is equivalent to equipping the defense end with the same level of firepower. Truly making 'secure programming' a default, rather than a remedy afterwards. Recommended practical process 1. Enter/code security → select Scan codebase (or Scan changes) 2. Choose the scanning depth (Medium is usually sufficient, High/Max is deeper but more expensive) 3. Wait for the scan to complete (will generate CLAUDE-SECURITY timestamp/folder) 4. Refer to the report (CLAUDE-SECURATY-RESULTS. md) 5. Enter/code security again → select Suggest patches 6. Select the finding to be repaired (e.g. F1, F2) 7. Use git apply to apply the generated. patch file, review it yourself, and then commit antic For major projects, it is recommended to scan only the key directories first, rather than scanning the entire repo at once Scanning is non deterministic, and running the same code twice may result in slightly different results Repairs will never be automatically applied, they must be reviewed by someone This way, you can directly turn "security scanning+patch suggestions" into a part of your daily workflow in the terminal!