222 GitHub Repositories Linked to Fake Go Package Malware Operation

Pierluigi Paganini
July 10, 2026

Researchers uncovered 222 GitHub repositories spreading malware through fake Go packages, delivering loaders, stealers, RATs, and cryptominers.

Socket’s security research team started with the investigation of a single malicious Go module: github[.]com/kaleidora/dnsub-scanning-tool, which presented itself as a DNS and subdomain scanning utility. Pulling on that thread exposed something significantly larger: a network of 222 confirmed GitHub repositories across 190 accounts, all built to make malicious or deceptive software projects look active, recently maintained, and worth running.

Socket tracks the operation as Muck and Load, named after the Muck-themed infrastructure it runs on and the staged loading chain it uses to deliver malware.

The confirmed payload set found across these repositories includes trojan loaders, Vidar infostealer, dropper and spyware payloads, and Monero cryptominers tied to XMRig. This wasn’t a collection of empty lure pages. Some repositories directly distributed malware.

The dnsub-scanning-tool module impersonated a legitimate open-source subdomain enumeration project. Its README described real scanner functionality. Its code did something else. Before any scanner logic could run, the module’s main() function launched a hidden PowerShell command that downloaded content from muckcoding[.]com, saved it as api.db, decoded it using certutil, wrote the result as L.ps1, and executed it with execution-policy bypass and a hidden window. As Socket’s report describes it:

“The loader’s execution mode further reinforces malicious intent. It launches PowerShell with a hidden window and then invokes the decoded script with: -ExecutionPolicy Bypass.” reads the report published by cybersecurity firm Socket.

“That flag does not exploit PowerShell by itself, but it is commonly used by malware to avoid local script-execution policy restrictions. In this case, it is paired with hidden execution and a freshly decoded script from an external source, making the intent clear: the Go module is acting as a first-stage Windows loader.”

One more anomaly made the module stand out before anyone even ran it: it had over 1,200 published versions, more than 700 of them malicious, for a small scanner tool created only months earlier. The version volume wasn’t normal release engineering. It was the result of the threat actor’s own GitHub Actions workflow repeatedly generating timestamp commits to manufacture fake development activity.

L.ps1 is a multi-layer PowerShell loader that uses Base64 encoding and XOR decryption to peel through several stages before reaching its actual function. One of the intermediate layers contains a comment in Turkish: “Direkt calistir, baska adim gerekmez,” which translates as “Run directly, no other step is needed.” The final decoded script is a resolver that retrieves encrypted payload-location material from public platforms rather than hardcoding a direct download URL.

The dead-drop sources include Pastebin, a service called Rlim, Muck-themed infrastructure, and fallback locations on YouTube, Instagram, Telegram, Google Docs, and GitCode. The script searches content retrieved from these platforms for a marker string called LastW, extracts the encrypted blob following that marker, decrypts it with a hardcoded key, and resolves the actual payload URL. As Socket explains:

“Rather than hardcoding only the final payload URL, the loader retrieves text from these public locations and searches the returned content for the marker LastW. In the observed dead-drop material, LastW appears as a marker string appended to or placed after the encoded/encrypted blob.” continues the report. “The script uses that marker to identify and extract the encrypted payload-location data, then decrypts the recovered blob with the hardcoded key. The encrypted payload-location blob recovered from the dead-drop material has SHA-256 51cada347262d7b2bcde70552fcdae221625ad75435cee8a9c3e7b67cc47a807.”

The decrypted URL resolves to a GitHub release: a password-protected .7z archive called Quixo.7z. The loader downloads it, extracts it using a 7-Zip binary it stages under C:\ProgramData\zipathh\7zrr.exe, and drops the contents into C:\ProgramData\Windows.Microsoft.Photos, a directory name that mimics a legitimate Microsoft application path without being one. The main executable, Microsoft.exe, launches from that location with a hidden window. It’s signed by Exodus Movement, Inc., not Microsoft, which suggests the threat actor repackaged or renamed a signed Electron-style application component while relying on surrounding scripts for the malicious behavior.

The infrastructure pivot came from a single GitHub Actions workflow signature. Across all 222 confirmed repositories, the same workflow combined a threat actor-linked email address, ischhfd83@rambler[.]ru, with force-push automation, a schedule running every minute, and synthetic commit activity that repeatedly rewrote a timestamp or log file and pushed the result. The visible commit username changed with each repository owner, but the underlying email stayed constant.

“That split between email and username is the fingerprint. The visible commit name changes with each repository owner, but the underlying email remains constant across the cluster.” states the report. “This gives the threat actor owner-specific commit activity at scale while leaving a reusable pivot for defenders.”

The repositories don’t all need to host payloads to be useful to the operation. Their function is to look active and plausible long enough for a target to clone, build, or follow setup instructions. The lure themes were chosen to attract users already inclined to run untrusted code: cryptocurrency and Web3 tooling, wallet integrations, seed-phrase protection utilities, Telegram bots, game auto-farm tools, game cheats, crypters, and offensive automation. The same repository names appeared under multiple accounts, consistent with scripted generation rather than organic development.

Some repositories did host payloads directly. One Exodus-themed repository used a right-to-left-override character in a filename to disguise a Windows .scr executable as something else. One PUBG-themed repository hosted Loader.exe in the source tree, associated with Vidar infostealer. One Warzone-themed repository delivered its payload as a GitHub release asset rather than a source file. A single Loader.exe binary appeared byte-identically across four separate repositories.

The downstream payload activity maps to AsyncRAT, Quasar, and Remcos-style remote access tools alongside infostealer-like behavior. The payload chain modifies Microsoft Defender and UAC settings, establishes persistence through scheduled tasks and services, accesses browser profile data, prepares screenshot collection, and communicates with Telegram or other public web services. API calls including WriteProcessMemory and SetThreadContext appear in dynamic analysis, supporting the assessment that this is a staged post-compromise environment with remote access and credential-theft capabilities.

Socket notes that this activity overlaps with previously reported repository-backdoor campaigns linked to the same ischhfd83 email address, documented earlier by Sophos researchers. The overlap isn’t defined by the email alone.

“We assess with high confidence that Operation Muck and Load is part of the same repository-backdoor activity cluster previously reported around ischhfd83. That assessment does not depend on one mutable indicator.” concludes the report. “The email address and Muck-themed domains are useful pivots, but the stronger basis is the repeated combination of repository automation, lure design, staging architecture, payload delivery, and post-execution behavior.”

The researchers pointed out that the email address, Muck-themed domains, and specific dead-drop URLs are all replaceable. The operational pattern is harder to change.

Socket reported the malicious Go module to the Go security team, which blocked it from the Go module proxy. The GitHub infrastructure was reported to GitHub’s security team. Neither action eliminates the underlying tradecraft, but both reduce the surface available to developers who encounter these repositories through normal search and discovery.

The 222-repository count is intentionally conservative and represents only repositories where the threat actor email and the synthetic GitHub Actions workflow appeared together in the same context, not every suspicious or lure-themed repository the investigation surfaced.

Follow me on Twitter: @securityaffairs and Facebook and Mastodon