The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Executes a remote shell script directly from an untrusted GitHub repository (mhsanaei/3x-ui) using bash process substitution (curl | bash). This grants the external script full control over the target system and execution environment.
[COMMAND_EXECUTION]: Performs numerous high-privilege system modifications, such as editing /etc/ssh/sshd_config to disable password authentication, modifying kernel parameters via sysctl for BBR, and installing system packages using apt-get.
[CREDENTIALS_UNSAFE]: Accesses and manages highly sensitive authentication data, including SSH private/public keys, root passwords, and administrative credentials for the 3x-ui panel.
[DATA_EXFILTRATION]: Performs network requests to external services like api.ipify.org to retrieve the server's public IP and fetches configuration files from remote subscription URLs, which could potentially expose server presence or metadata.
[PROMPT_INJECTION]: Exhibits a vulnerability surface for indirect prompt injection through untrusted data processing.
Ingestion points: Processes data from remote command outputs (e.g., ssh results in SKILL.md) and external subscription URLs.
Boundary markers: Absent; no delimiters or safety instructions are used to separate external data from system instructions.
Capability inventory: High-privilege command execution via ssh, bash, and curl, along with file writing operations in SKILL.md.
Sanitization: Absent; no filtering or validation is performed on ingested external content.

ray-vps