scrollcraft

Warn

Audited by Socket on Oct 2, 2026

2 alerts found:

Anomalyx2
AnomalyLOW
SKILL.md

SUSPICIOUS. The core build, file access, and local verification steps are broadly consistent with a scroll-page design skill, and the npm/Playwright usage is normal developer tooling. The main concern is the optional kie.ai integration: it sends prompts, assets, and a paid API credential through a third-party gateway rather than official model-provider APIs, which is a real data-flow and trust expansion not strictly necessary for the stated purpose. Overall this looks more like a legitimate but higher-risk creative workflow skill than a malicious credential harvester.

Confidence: 89%Severity: 56%
AnomalyLOW
scripts/kie.mjs

No overt malicious/backdoor behavior is evident in the provided file (no eval/exec, no persistence, no stealthy activity). However, the module implements an explicit local-file upload (base64 exfiltration) capability to a hardcoded third-party endpoint whenever local paths are supplied, and it performs unvalidated downloading from API-returned URLs and arbitrary file writes to a user-provided output path. The security posture is therefore dominated by data-exfiltration and operational misuse risk rather than by clear malware. Careful trust-model review and input/output validation (file allowlists/size limits, URL allowlisting, safer output path handling) are recommended for supply-chain scenarios.

Confidence: 66%Severity: 60%
Audit Metadata
Analyzed At
Oct 2, 2026, 04:04 PM
Package URL
pkg:socket/skills-sh/nousresearch%2Fhermes-agent%2Fscrollcraft%2F@e7606ddb21db165539730ad7be56891f52f4e67e7aba3dcbb7dc3cd96eec7383