The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in SKILL.md use the high-risk pattern curl https://sliver.sh/install | sudo bash. This executes a remote shell script directly with root privileges without verifying its contents.
[EXTERNAL_DOWNLOADS]: The skill makes extensive use of the armory command (e.g., sliver > armory install all), which downloads and installs third-party extensions and Beacon Object Files (BOFs) from remote repositories at runtime.
[COMMAND_EXECUTION]: The skill facilitates arbitrary command execution on remote targets via shell and execute commands. It also includes tmux automation scripts that execute shell commands and capture their output, which can lead to command injection if session variables are manipulated.
[DATA_EXFILTRATION]: Explicit commands for data exfiltration are documented, including download /remote/file /local and memory dumping tools like procdump, nanodump, and mimikatz for harvesting credentials and sensitive system information.
[DYNAMIC_EXECUTION]: The skill documents advanced techniques for memory-only code execution, such as execute-assembly (.NET assembly loading), execute-shellcode (process injection), and coff-loader for running BOFs. These methods bypass traditional disk-based security monitoring.
[PRIVILEGE_ESCALATION]: Several mechanisms for acquiring elevated permissions are provided, including the getsystem command and impersonate USERNAME for token manipulation.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection in the Phase 1: 环境准备 section. The agent uses tmux capture-pane -t sliver -p to read output from the C2 console. If the C2 server or a compromised implant returns malicious instructions, they would be ingested directly into the agent's context.

sliver-c2