openghost-skill

The skill is purpose-aligned and not overtly malicious, but it is a high-risk offensive-security skill because it enables an AI agent to run pentest tooling, browser attacks, and autonomous assessments against target systems. Supply-chain trust is only moderate due to limited public release verification for the bundled launcher, but there is no clear evidence of credential theft, covert exfiltration, or deceptive behavior in the provided content.

The fragment appears to be a legitimate penetration-testing sandbox launcher, not malware. It intentionally executes user-supplied Bash/Python and performs network scans, but those operations are directed through Docker and ZAP as core functionality. The principal supply-chain risk is pulling an unpinned mutable GHCR latest image, while the principal security risks are broad workspace mounting, host Docker control, weak Bash blocklisting, and unescaped user values in generated ZAP YAML. No direct credential theft, persistence, reverse shell, cryptomining, or covert exfiltration is evident in the shown portion.

The code is a browser evidence-capture utility rather than apparent malware. It intentionally visits a user-supplied URL and records extensive browser data. It should only be run against trusted or authorized targets, in an isolated environment, with TLS validation and sandboxing enabled where possible, and with strict protection of the output artifacts because they may contain session credentials and private content.