The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Provides templates and one-liners for establishing reverse shells using multiple languages (Bash, Python, Perl, PHP, Ruby, Node.js, and Lua) to connect back to attacker-controlled infrastructure.
[EXTERNAL_DOWNLOADS]: Instructs the agent to download and execute scripts directly from public GitHub repositories using insecure patterns like piped execution (curl | sh) and manual execution of unverified binaries (e.g., LinPEAS, pspy, deepce, and linux-exploit-suggester).
[DATA_EXFILTRATION]: Contains detailed instructions for harvesting sensitive information, including SSH private keys, VPN certificates (WireGuard/OpenVPN), database credentials, cloud metadata (IMDS), and system shadow files, with methods for transferring them to external listeners.
[COMMAND_EXECUTION]: Employs high-privilege system commands and techniques to manipulate the target environment, such as abusing sudo permissions, using mount to access host filesystems, and directly modifying sensitive system files like /etc/passwd.
[DYNAMIC_EXECUTION]: Includes steps for runtime compilation of C source code using gcc to generate malicious shared libraries for LD_PRELOAD injection and to build kernel-level exploit binaries.
[PRIVILEGE_ESCALATION]: Offers a comprehensive library of techniques for breaking out of container isolation (e.g., Docker socket abuse, cgroup escapes), abusing Kubernetes RBAC and ServiceAccounts, and escalating from low-privileged users to root on Linux systems.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external tool outputs and configuration files (e.g., process snooping logs, LinPEAS reports, Terraform state files) while maintaining high-privilege capabilities like file modification and network access, without specifying sanitization or explicit boundary markers.
Ingestion points: Reads output from pspy, LinPEAS, system logs, and configuration files (SKILL.md, references/linux-privesc.md, references/container-breakout.md).
Boundary markers: Absent; the skill does not use delimiters or warnings to ignore embedded instructions in external data.
Capability inventory: Full shell execution, file-write access, network operations, and runtime compilation across all reference files.
Sanitization: Absent; external content is processed directly by the agent.

privilege-escalation