The Agent Skills Directory

[OBFUSCATION]: Automated security scans identified evals/evals.json as a malicious dropper (FileRepMalware [Drp]). The file contains encoded PowerShell stagers and malicious patterns frequently used in malware delivery.
[PRIVILEGE_ESCALATION]: Reference materials in references/scripts-and-obfuscation.md explicitly instruct the agent to use sudo for network monitoring (tcpdump) and process tracing (strace), granting the agent elevated permissions on the host system without sufficient safety warnings.
[COMMAND_EXECUTION]: The skill provides commands for executing unverified malicious binaries (./malware) and emulating shellcode using Unicorn engine. Direct execution of untrusted code poses a high risk of host compromise if the agent environment is not strictly isolated.
[INDIRECT_PROMPT_INJECTION]: The skill's core purpose involves processing and deobfuscating untrusted data (malware, PCAP logs, scripts) which constitutes a significant attack surface for indirect prompt injection. The skill lacks explicit boundary markers or sanitization logic for handling these external inputs.
[EXTERNAL_DOWNLOADS]: The skill references an external unpacking tool from an unverified third-party GitHub repository (github.com/Svenskithesource/PyArmor-Unpacker), which could lead to the execution of unvetted code during a malware analysis workflow.

ctf-malware