The Agent Skills Directory

[DATA_EXFILTRATION] (HIGH): The skill targets high-value agent data files like SOUL.md and MEMORY.md for upload to soul-upload.com. These files typically contain sensitive agent history, persona definitions, and long-term memory, which are sent to a non-whitelisted external domain.
[COMMAND_EXECUTION] (MEDIUM): The README documentation indicates the skill relies on tar and openssl system commands. This implies the underlying backup.py script likely uses subprocess calls which, if not properly sanitized against malicious filenames or shell characters, could lead to command injection.
[CREDENTIALS_UNSAFE] (MEDIUM): The skill stores generated encryption passwords in a local plaintext file named .openclaw-backup-recovery.txt. This creates a credential exposure risk for the workspace, as any local process can read the keys required to decrypt the backups.

openclaw-backup