ai-agent-supply-chain-attack

Warn

Audited by Socket on Sep 19, 2026

2 alerts found:

Securityx2
SecurityMEDIUM
SKILL.md

SUSPICIOUS: the install sources shown are mostly legitimate and proportionate, but the skill’s actual purpose is to operationalize offensive AI supply-chain attacks, including persistence and exfiltration. This is internally coherent as a red-team skill, yet high-risk because it equips an agent/operator with attack workflows against external registries, plugins, and retrieved content.

Confidence: 90%Severity: 72%
SecurityMEDIUM
test-cases.md

The supplied content is an offensive AI/ML supply-chain security test plan. It documents and partially implements critical attack techniques, most notably pickle deserialization RCE, credential-targeting prompt injection, model backdoors, scanner evasion, and potential MLflow/Kubeflow compromise. It should be restricted to authorized isolated environments. The document is not obfuscated and does not itself demonstrate execution or exfiltration, but several included snippets are directly dangerous when run.

Confidence: 98%Severity: 90%
Audit Metadata
Analyzed At
Sep 19, 2026, 04:39 PM
Package URL
pkg:socket/skills-sh/brucesongs%2Fkali-claw%2Fai-agent-supply-chain-attack%2F@ef7dd20532dc0a6951fc71431ba5f1bc3b112e6497662930545b190481edd343