Devin Can Now Check Your Code for Java4Shell and NGINX2Root Flaws in One Click
This week saw the disclosure of Java4Shell (November 17, a flaw in the Java runtime that allows memory disclosure and remote code execution) and NGINX2Root (November 18, five bugs in nginx). Cognition has added a Threats section to Devin Security: you pick a threat, select repositories, and Devin finds where your code uses the vulnerable component. You can hand the findings to Devin, and it will open pull requests with fixes.
AI-processed from Windsurf Blog; edited by Hamidun News
Picture this: a high-profile vulnerability drops in the morning, and you need to know right away whether your code is affected. Java4Shell, disclosed on November 17, is a flaw in the Java runtime environment that lets an attacker read memory contents and remotely run arbitrary code on a server. NGINX2Root, disclosed on November 18, is a set of five memory-handling bugs in the nginx web server that can also lead to remote code execution.
Engineers usually have to hunt by hand through hundreds of repositories and containers to find where Java or nginx lives, and then work out how dangerous the flaw is for their particular setup. Cognition, the company behind the Devin agent, wants to automate that step. It has launched Devin Security Threats, a managed index of new vulnerabilities, where each entry lets you start a code check and a fix in one click.
If you are responsible for security or maintain many services, this is a way to answer the question "does this affect us?" faster.
How a "threat" works in Devin
In Devin, a threat is a short description of what can go wrong: what is affected, how an attacker would exploit it, and what the fix addresses. Cognition publishes these descriptions for major disclosures such as the two above. Teams can add their own.
Next comes the Scan with Devin button. It opens a new scan with the chosen threat already selected in the Focus on threats field. All that is left is to mark the repositories Devin should check. The scan runs on the same Agentic MapReduce process as a regular Devin Security scan. Findings are ranked against your threat model: severity depends on how your system is built, not only on the vulnerability's general rating. You can then assign the findings to Devin, and it will open pull requests for humans to review.
There is also a mode that needs no manual trigger. If you turn on Auto-triage, you choose the repositories and a minimum severity. After that, Devin scans them on its own for every newly published threat. You can also enable automatic closing of a threat when the scan finds nothing. That removes one more step for the team: no one has to mark by hand that the problem does not affect you.
Custom threats and ready-made skills
The list is not limited to what Cognition publishes. Suppose a bug bounty report, a vendor notice, or an incident postmortem raised a question about your own code. You can create a threat with a title and a description of what could happen and how, then scan repositories against it just as you would for a public one. In effect, it turns a one-off question like "could this happen to us?" into a repeatable check.
For those who prefer to fix things in a session rather than through a scan, the Vulnerability Fixes plugin has gained two skills. fix-nginx-vuln targets the five NGINX2Root bugs, and fix-java-vuln targets Java4Shell. You can install the plugin from the Devin Marketplace or from the Devin CLI with the command devin plugins install CognitionAI/devin-marketplace#plugins/devin-vuln-fixes. After that, just start a session with the prompt "Fix nginx vulnerabilities" or "Fix Java vulnerabilities". The skills are plain open-source Markdown files. You can read them as a checklist or hand them to another coding agent.
What this means in practice, and the caveats
For teams already using Devin, the benefit is clear: a high-profile vulnerability no longer requires building a separate process. You just pick the threat and the repositories. The open skills are also useful to people who do not use Devin: they are a ready-made list of steps that you can adapt to your own tool.
But keep in mind that this is an announcement from the vendor itself. The material gives no data on how accurately Devin finds vulnerable spots, how many false positives it produces, or how often its pull requests pass review without changes. The source limits its technical detail on Java4Shell and NGINX2Root to brief descriptions, so for specifics you should go to the projects' official advisories. The post does not state access terms or pricing.
The practical takeaway is simple. If you are building vulnerability response around an AI agent, start with Auto-triage on a small set of critical repositories and compare the results with a regular dependency scanner. The decision to merge fixes should stay with people: Devin only opens pull requests, and you are the one who has to review them.
- Nov 20, 2026Pedestal AI Emerges From Stealth With $4.7M: Its Agents Cut Failed Deliveries by 20%
- Nov 20, 2026GitLab connects Claude Code to Duo agents with a single skill: 4 workflows from the terminal
- Nov 20, 2026ResolveAI lifts a 27B model from 6.9% to 38.8% of production incidents solved, beating top LLMs
- Nov 20, 2026GitLab makes Duo Agent Platform credits last up to 2x longer
Key Questions Answered
How a "threat" works in Devin?
In Devin, a threat is a short description of what can go wrong: what is affected, how an attacker would exploit it, and what the fix addresses. Cognition publishes these descriptions for major disclosures such as the two above. Teams can add their own. Next comes the Scan with Devin button. It opens a new scan with the chosen threat already selected in the Focus on threats field. All that is left is to mark the repositories Devin should check.
What are the key details on Custom threats and ready-made skills?
The list is not limited to what Cognition publishes. Suppose a bug bounty report, a vendor notice, or an incident postmortem raised a question about your own code. You can create a threat with a title and a description of what could happen and how, then scan repositories against it just as you would for a public one. In effect, it turns a one-off question like "could this happen to us?" into a repeatable check.
What this means in practice, and the caveats?
For teams already using Devin, the benefit is clear: a high-profile vulnerability no longer requires building a separate process. You just pick the threat and the repositories. The open skills are also useful to people who do not use Devin: they are a ready-made list of steps that you can adapt to your own tool. But keep in mind that this is an announcement from the vendor itself.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.