The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): Found in SKILL.md and scripts/check_setup.sh: curl https://rclone.org/install.sh | sudo bash. Piping remote scripts to a shell with root privileges is a high-risk pattern that allows unverified code execution.
[COMMAND_EXECUTION] (HIGH): The skill frequently uses sudo for installation and configuration (e.g., sudo apt install, sudo bash), increasing the attack surface for privilege escalation.
[CREDENTIALS_UNSAFE] (HIGH): The skill instructs the user to provide access_key_id and secret_access_key for various cloud providers (AWS, Cloudflare, Backblaze). These secrets are handled in plaintext commands, creating a risk of exposure in logs or shell history.
[EXTERNAL_DOWNLOADS] (MEDIUM): Downloads scripts and software from rclone.org. Since this domain is not in the trusted sources list, it is treated as an unverified external dependency.
[PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection. Ingestion points: File paths and remote targets in rclone copy/sync/ls (SKILL.md). Boundary markers: Absent. Capability inventory: rclone (file/network access), sudo (privilege escalation), bash (shell execution). Sanitization: Absent. Maliciously crafted filenames could potentially lead to command injection if the agent interpolates them directly into shell commands.

rclone