control-in-app-browser

Warn

Audited by Socket on Aug 24, 2026

2 alerts found:

Securityx2
SecurityMEDIUM
agents/openai.yaml

No traditional malware or obfuscation is evident, but the configuration is highly security-sensitive: it explicitly authorizes an agent to inspect sensitive browser-stored secrets (including passwords/session stores) and to perform high-impact actions such as deploy/publish and DNS/service changes without additional confirmation gates, including bypass-oriented behavior around authentication blocks. If used in any environment with access to authenticated contexts, it materially increases the likelihood of credential harvesting and unauthorized operational changes.

Confidence: 76%Severity: 86%
SecurityMEDIUM
SKILL.md

SUSPICIOUS: the skill’s browser-control purpose is coherent and its runtime appears same-org bundled rather than third-party, so supply-chain risk is low to moderate. However, its scope is unusually broad: it permits access to sensitive browser credential stores and enables autonomous consequential actions in authenticated sessions without per-action confirmation. That makes it high-risk as an agent capability even without clear evidence of malware or external credential exfiltration.

Confidence: 87%Severity: 76%
Audit Metadata
Analyzed At
Aug 24, 2026, 12:16 AM
Package URL
pkg:socket/skills-sh/samihalawa%2Fopenai-bundled-unrestricted%2Fcontrol-in-app-browser%2F@a131ae05846c4e910af0ba55f0fbfdc1dc77fa07143cb83955a7143c275ca24f