container-security-hardening

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). The required workflow includes fetching and executing installer scripts via curl from public GitHub URLs (outsider-authored web content), which then becomes readable text in the agent’s runtime context (e.g., via sed -n '1,160p' "$tmpdir/*-install.sh").

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill contains commands that fetch and execute remote install scripts and manifests (e.g., curl ... -o ... && sh on https://raw.githubusercontent.com/aquasecurity/trivy/main/contrib/install.sh and https://raw.githubusercontent.com/anchore/grype/main/install.sh, plus kubectl/helm commands that apply or install remote charts/manifests such as https://raw.githubusercontent.com/open-policy-agent/gatekeeper/release-3.17/deploy/gatekeeper.yaml, https://kyverno.github.io/kyverno/, and https://falcosecurity.github.io/charts), which will execute external code at runtime.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill includes explicit host-level commands that require elevated privileges and modify system security (e.g., "sudo apparmor_parser -r /etc/apparmor.d/docker-default"), plus instructions to run containers unconfined for syscall auditing, which encourage changing the machine's security posture.