Opening a project in a code editor is supposed to be routine. In this case, it is enough to trigger a full malware infection.
Security researchers have linked an ongoing campaign associated with North Korean actors, tracked as Contagious Interview or WaterPlum, to a malware family known as StoatWaffle. Instead of relying on software vulnerabilities, the group is embedding malicious logic directly into Microsoft Visual Studio Code (VS Code) projects, turning a trusted development tool into the starting point of an attack.
The entire mechanism is hidden inside a file developers rarely question: tasks.json. This file is typically used to automate workflows. In these attacks, it has been configured with a setting that forces execution the moment a project folder is opened. No manual action is required beyond opening the workspace.
Research from NTT Security shows that the embedded task connects to an external web application, previously hosted on Vercel, to retrieve additional data. The same task operates consistently regardless of the operating system, meaning the behavior does not change between environments even though most observed cases involve Windows systems.
Once triggered, the malware checks whether Node.js is installed. If it is not present, it downloads and installs it from official sources. This ensures the system can execute the rest of the attack chain without interruption.
What follows is a staged infection process. A downloader repeatedly contacts a remote server to fetch additional payloads. Each stage behaves in the same way, reaching out to new endpoints and executing the returned code as Node.js scripts. This creates a recursive chain where one payload continuously pulls in the next.
StoatWaffle is built as a modular framework. One component is designed for data theft, extracting saved credentials and browser extension data from Chromium-based browsers and Mozilla Firefox. On macOS systems, it also targets the iCloud Keychain database. The collected information is then sent to a command-and-control server.
A second module functions as a remote access trojan, allowing attackers to operate the infected system. It supports commands to navigate directories, list and search files, execute scripts, upload data, run shell commands, and terminate itself when required.
Researchers note that the malware is not static. The operators are actively refining it, introducing new variants and updating existing functionality.
The VS Code-based delivery method is only one part of a broader campaign aimed at developers and the open-source ecosystem. In one instance, attackers distributed malicious npm packages carrying a Python-based backdoor called PylangGhost, marking its first known propagation through npm.
Another campaign, known as PolinRider, involved injecting obfuscated JavaScript into hundreds of public GitHub repositories. That code ultimately led to the deployment of an updated version of BeaverTail, a malware strain already linked to the same threat activity.
A more targeted compromise affected four repositories within the Neutralinojs GitHub organization. Attackers gained access by hijacking a contributor account with elevated permissions and force-pushed malicious code. This code retrieved encrypted payloads hidden within blockchain transactions across networks such as Tron, Aptos, and Binance Smart Chain, which were then used to download and execute BeaverTail. Victims are believed to have been exposed through malicious VS Code extensions or compromised npm packages.
According to analysis from Microsoft, the initial compromise often begins with social engineering rather than technical exploitation. Attackers stage convincing recruitment processes that closely resemble legitimate technical interviews. Targets are instructed to run code hosted on platforms such as GitHub, GitLab, or Bitbucket, unknowingly executing malicious components as part of the assessment.
The individuals targeted are typically experienced professionals, including founders, CTOs, and senior engineers in cryptocurrency and Web3 sectors. Their level of access to infrastructure and digital assets makes them especially valuable. In one recent case, attackers unsuccessfully attempted to compromise the founder of AllSecure.io using this approach.
Multiple malware families are used across these attack chains, including OtterCookie, InvisibleFerret, and FlexibleFerret. InvisibleFerret is commonly delivered through BeaverTail, although recent intrusions show it being deployed after initial access is established through OtterCookie. FlexibleFerret, also known as WeaselStore, exists in both Go and Python variants, referred to as GolangGhost and PylangGhost.
The attackers continue to adjust their techniques. Newer versions of the malicious VS Code projects have moved away from earlier infrastructure and now rely on scripts hosted on GitHub Gist to retrieve additional payloads. These ultimately lead to the deployment of FlexibleFerret. The infected projects themselves are distributed through GitHub repositories.
Security analysts warn that placing malware inside tools developers already trust significantly lowers suspicion. When the code is presented as part of a hiring task or technical assessment, it is more likely to be executed, especially under time pressure.
Microsoft has responded to the misuse of VS Code tasks with security updates. In the January 2026 release (version 1.109), a new setting disables automatic task execution by default, preventing tasks defined in tasks.json from running without user awareness. This setting cannot be overridden at the workspace level, limiting the ability of malicious repositories to bypass protections.
Additional safeguards were introduced in February 2026 (version 1.110), including a second prompt that alerts users when an auto-run task is detected after workspace trust is granted.
Beyond development environments, North Korean-linked operations have expanded into broader social engineering campaigns targeting cryptocurrency professionals. These include outreach through LinkedIn, impersonation of venture capital firms, and fake video conferencing links. Some attacks lead to deceptive CAPTCHA pages that trick victims into executing hidden commands in their terminal, enabling cross-platform infections on macOS and Windows. These activities overlap with clusters tracked as GhostCall and UNC1069.
Separately, the U.S. Department of Justice has taken action against individuals involved in supporting North Korea’s fraudulent IT worker operations. Audricus Phagnasay, Jason Salazar, and Alexander Paul Travis were sentenced after pleading guilty in November 2025. Two received probation and fines, while one was sentenced to prison and ordered to forfeit more than $193,000 obtained through identity misuse.
Officials stated that such schemes enable North Korean operatives to generate revenue, access corporate systems, steal proprietary data, and support broader cyber operations. Separate research from Flare and IBM X-Force indicates that individuals involved in these programs undergo rigorous training and are considered highly skilled, forming a key part of the country’s strategic cyber efforts.
What this means
This attack does not depend on exploiting a flaw in software. It depends on exploiting trust.
By embedding malicious behavior into tools, workflows, and hiring processes that developers rely on every day, attackers are shifting the point of compromise. In this environment, opening a project can be just as risky as running an unknown program.
.png "Emerging AI Built Malware Used in Targeted Attacks on Blockchain Engineers")
.png)
