k8s-sidecar-attack

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). This skill explicitly instructs capturing plaintext network traffic (tcpdump -A, greps for token/password/secret) and inspecting Authorization headers, cookies and POST bodies, which requires handling and exposing secret values verbatim (high exfiltration risk).

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This skill explicitly instructs an attacker on abusing Kubernetes sidecars and shared network namespaces to sniff traffic, capture credentials/tokens, and perform MITM/exfiltration (tcpdump, socat), and thus constitutes deliberate malicious behavior.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). This skill explicitly instructs active in-pod traffic interception (tcpdump, socat) and even suggests bypassing Istio security, which directs the agent to perform network MITM and sensitive-data exfiltration on the host/pod and thus enables compromising the machine's operational security.