The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill's primary workflow (in SKILL.md) instructs users to execute a remote shell script directly into a root shell using the highly insecure pattern curl -fsSL ... | sudo bash. This allows the script's author to execute arbitrary code on the remote server with full administrative privileges.
[CREDENTIALS_UNSAFE] (HIGH): The scripts/remote.sh script automatically creates a system user, generates a temporary password, and stores it in cleartext. This password is then transmitted in a JSON payload via an unencrypted HTTP POST request.
[DATA_EXFILTRATION] (HIGH): Sensitive connection details, including the generated username and cleartext password, are sent to a webhook exposed via bore.pub. Because bore.pub is a public tunneling service, this sensitive data is transmitted through third-party infrastructure where it can be intercepted, logged, or exfiltrated by malicious actors.
[EXTERNAL_DOWNLOADS] (HIGH): The skill downloads the bore binary from an untrusted GitHub repository (ekzhang/bore) and executes it. In scripts/remote.sh, the binary is moved to /usr/local/bin and given executable permissions, which requires and utilizes elevated privileges without verifying the binary's integrity via checksums.
[COMMAND_EXECUTION] (MEDIUM): The scripts/server.js and scripts/remote.sh files make extensive use of dangerous system commands including useradd, chpasswd, execSync, and spawn to manipulate system users and process states, which could be exploited if inputs are not properly sanitized.

opentunnel-connect