A RubyGems string-helper package ran a JavaScript credential worm while it was being installed. The malicious release was strutil-core version 1.0.1; its native-extension installer started the worm without an application loading the gem. For teams that build Ruby applications on developer machines or CI runners, the first question is whether that version ever reached an install job.
Socket’s November 6 investigation says the publisher released a clean 1.0.0 on November 1, then published 1.0.1 less than two hours later. Socket flagged the malicious release 70 seconds after publication; RubyGems removed both versions about five hours after the bad release. Socket writes that “strutil-core was a standalone gem, not a compromised existing package.” Its analysis found no public projects depending on it and no confirmed victims or downstream packages infected from this gem. Scope a response to installations of 1.0.1 on developer machines and build runners.
Installation ran the hidden bundle
The RubyGems extension guide explains the legitimate convention: extconf.rb prepares a Makefile when an extension is built during gem installation. The strutil-core release used that convention as a trigger. Socket found that ext/strutil-core/extconf.rb downloaded the legitimate Bun runtime on supported Linux and macOS systems, ran the adjacent 5.48 MB ext/strutil-core/index.js bundle, then wrote a Makefile with empty build targets. The install could finish successfully even after the bundle ran. An application never had to load the gem.
That mechanism makes a Gemfile.lock entry, CI install log or cached gem a better starting point than an application stack trace. The original strutil-core release implemented a small slug helper; the later gem kept that library version string while changing the gemspec to 1.0.1. Socket says the bundled injector updates gem metadata and installs extension files without editing the old library. The installer in the bad release matched output from that injector byte for byte, with SHA-256 c5264f3b9cf2f95c83a70b7dd7430aab8abc262b853dbbcd5386c3a8a9c74769.
Socket cites a Bitbison sandbox run that observed install-time execution and reads from credential stores. Further propagation, AI-tool poisoning and deletion are capabilities Socket identified in decoded code. The sandbox recorded reads from .aws/credentials, .npmrc, .pypirc, .gem/credentials, .docker/config.json, .kube/config, .claude/.credentials.json and .codex/auth.json. The code also targets CI secrets, SSH material, cloud credentials and package-publishing keys. The installer could read credentials available to its process, so check the affected host and runner identities when scoping exposure.
Stolen access would outlive removal
Socket found two ways the bundle tries to make a single installation durable. It registers a Python backdoor as the pkg-runtime user service or com.user.pkg-runtime LaunchAgent, with the script at ~/.local/share/runtimes/sync.py. It also plants workspace-tools MCP server entries, SessionStart hooks and agent instructions into AI coding configurations and repositories. Look for unexpected changes to .claude/settings.json, .cursor/mcp.json, .github/init.js, CLAUDE.md, AGENTS.md and .vscode/tasks.json. A tool opening an affected workspace could execute the planted hook, depending on its trust settings. Inspect these files without opening the suspect workspace in an agent or IDE.
A second persistence path creates the net-health-checker monitor after GitHub token exfiltration. Socket’s code analysis shows a destructive condition: an HTTP 401 or 403 response to checks of the stolen GitHub token triggers deletion of the user’s home-directory files. The monitor checks every 45 seconds for up to 48 hours. That changes the usual first move in credential response. If the monitor is present, isolate the machine and disable its service or LaunchAgent before revoking the token at ~/.config/net-health-checker/token; preserve evidence and recover from a clean host. The monitor’s script path is ~/.local/bin/net-health-checker.sh. Uninstalling the gem alone leaves these planted files and stolen credentials in play.
The bundle’s propagation routines target RubyGems, npm and PyPI, plus code-hosting and CI systems, according to Socket’s static analysis. For a separate but related Python package, the OSV record MAL-2026-18432 lists pystr-codec versions 0.1.0, 0.1.1 and 0.1.2. It describes a .pth startup trigger, credential theft, persistence and destructive behavior, and lists the same collection domain. Socket says the two packages carry separate builds. Check Python exposure separately rather than assuming a Ruby lockfile search covers it. For pystr-codec, OSV now lists pystr_codec-setup.pth with SHA-256 14dc1f185be61f2b0f49dc4118e48e44445d204c62106c01316b163cb64c5eba as a file indicator; compare suspect installations and logs, but do not treat a hash mismatch as proof that other listed versions are safe.
Check installs before rotating keys
Search Ruby Gemfile.lock files, CI dependency logs, caches and container images for strutil-core at 1.0.1; an installed copy warrants host isolation. Examine the installer and bundle paths above, unexpected pkg-runtime or net-health-checker user services or LaunchAgents, and the AI-tool files before opening a suspect workspace in an agent or IDE. The bundle SHA-256 is 306c1deec402e8b4a6478e7f0fe02194bcd224d37d8fc6ba61ad1951e8abd7a9; the gem archive SHA-256 is 3b3f6e1bde0b28b015030a81a6d805b5b5306dc3f021e1102123971cfd83acc0.
For network and repository telemetry, search for the attacker-controlled host paleasbest[.]top and its defanged collection endpoint hxxps://paleasbest[.]top/v1/api. Look for unexpected public GitHub repositories described as project-utils, commits marked AutoSyncComplete or skip-checks:true, and branches named chore/ci-*. The code abuses legitimate shared services: GitHub (github.com) for fallback collection and GitLab (gitlab.com) if that path fails. Correlate a matching repository or commit with host evidence. On CI runners, inspect unusual Bun execution from /tmp/.b_<pid> after a Ruby install and Runner.Worker memory access through /proc/<pid>/mem. Correlate those process and path strings with the runner’s gem-install history.
If 1.0.1 ran, stop its persistence and inspect the host before revoking exposed GitHub tokens; then rotate accessible RubyGems, npm, PyPI, cloud, Git-host, SSH, Vault, Kubernetes and AI-provider credentials from a clean system and rebuild the machine. Check packages the affected publishing identities control for unexplained patch releases or newly added native extensions. Historical install logs, affected host artifacts and publishing activity matter alongside today’s dependency file: the installer could read secrets and write hooks before a package was removed.