If this package is installed
Remove it, treat the machine (or CI runner) as compromised, and rotate every secret, token and key it could access — from a different, clean machine.
Details
Source: amazon-inspector (2b799ca087f4396cb295cd53012a92d10dea0c763a41c8b85e94437108ea24e4)
Package @kuperka/chainguard-sdk@1.0.2 exposes a single payload, window.__VEIL__.init(), that on load harvests document.cookie, every localStorage/sessionStorage entry, all form input values, meta CSRF/auth tokens, and Web3 wallet identifiers (window.ethereum, window.solana), base64-encodes the blob, and transmits it to the hardcoded endpoint https://webhook.site/b9155136-5f6d-4084-b19e-33c694a44616 via both Image().src GET and fetch POST. It additionally scans document.documentElement.outerHTML with credential-shaped regexes for Stripe live keys (sk_live_*), AWS access keys (AKIA*), GitHub personal access tokens (ghp_*), JWTs, and OpenAI keys (sk-*), including any matches in the exfil payload. A capturing keydown listener buffers keystrokes typed into any INPUT/TEXTAREA/SELECT along with field context (name/id/placeholder) and periodically posts them to the same endpoint. window.fetch is monkey-patched to record every outbound URL, method, Authorization header, and up to 2000 bytes of request body, forwarding captured bearer tokens and request bodies to the same webhook. After exfiltration the code renders a full-viewport overlay branded 'ChainGuard SECURITY & SIMULATION ENGINE' displaying scripted 'Initializing/Fetching/Simulating/Sandbox emulation complete' log lines and a hardcoded 'Overall Risk Score: 0 / No Threats Detected' result, concealing the theft from the user. The package name and 'chainguard' branding impersonate a legitimate security tool.