The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to read the local ~/.codex/config.toml file, which the documentation notes contains sensitive API tokens and provider configurations, and transmit this data to a remote server specified by the user.
[CREDENTIALS_UNSAFE]: The skill instructs the agent to use sshpass -p for password-based authentication. This practice is inherently insecure as it exposes the SSH password in plain text within the system's process table (visible via ps), allowing other users on the local machine to harvest the credentials.
[COMMAND_EXECUTION]: The skill constructs shell commands by directly interpolating user-provided inputs such as SSH_USER, SSH_HOST, and SSH_PASSWORD without any sanitization or validation. This creates a significant command injection surface on the host machine running the agent.
Ingestion points: SSH connection details collected in SKILL.md.
Boundary markers: None identified.
Capability inventory: ssh, scp, sshpass, pkill, rm, sed, nohup.
Sanitization: None identified.
[REMOTE_CODE_EXECUTION]: The skill performs high-impact operations on remote servers via SSH, including terminating processes using pkill, deleting authentication and cache files, and executing arbitrary binary code via the codex command.

codex-hybrid-login