The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to collect and exfiltrate highly sensitive data, including private SSH keys (e.g., ~/.ssh/id_vultr), environment variables containing API keys (.env), and authentication databases (auth.json). This data is packaged and uploaded to external destinations like GitHub and Baidu Wangpan. While this is the skill's stated purpose, the handling of private cryptographic keys and secrets for off-server storage significantly increases the attack surface.
[REMOTE_CODE_EXECUTION]: The restore.sh script provides a --from-url flag that downloads a compressed archive from a remote URL and extracts its contents directly to the system root (tar -xzf ... -C /). This allows for arbitrary file overwrites and potential remote code execution if the source URL is malicious or compromised.
[PRIVILEGE_ESCALATION]: The skill documentation includes instructions to use sudo for system-level operations, such as installing the GitHub CLI (gh) using the system package manager.
[EXTERNAL_DOWNLOADS]: The skill installs external dependencies from public registries, specifically the bypy Python package via pip and the gh utility via apt-get. It also utilizes git clone to fetch external repository content.
[PERSISTENCE]: The skill configures automated cron jobs to run backup scripts (e.g., backup.py) at regular intervals (every 12 hours), ensuring the backup and exfiltration process persists across system reboots and sessions.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands and subprocesses to perform administrative tasks, including archiving files with tar, managing repositories with gh and git, and executing network requests with curl.
[OBFUSCATION]: The skill explicitly mentions using a combination of XOR encoding and Base64 to "encrypt" the .env file during the backup process. While described as a security measure, this technique masks the content of sensitive files during transit.

disaster-recovery