The Agent Skills Directory

[COMMAND_EXECUTION]: Vulnerability in scripts/handler.sh. User-supplied remote_path from the /sync add command is used in the string ssh ... "mkdir -p $remote_path/$sync_path". This allows execution of arbitrary commands on the remote system via shell metacharacters in the path.
[COMMAND_EXECUTION]: Weak security configuration in scripts/handler.sh. The ssh utility is invoked with -o StrictHostKeyChecking=no, which disables verification of the remote host's identity and exposes the connection to man-in-the-middle attacks.
[DATA_EXFILTRATION]: The skill is designed to move sensitive files (memory/, MEMORY.md, USER.md) to external network addresses. While this is the primary purpose (synchronization), the capability can be misused to send sensitive agent data to unauthorized endpoints.
[EXTERNAL_DOWNLOADS]: The references/setup.md file includes instructions to download and execute a script from Tailscale's official domain using a curl | sh pipe.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface (Category 8).
Ingestion points: Files are pulled from remote peers via rsync into the local $WORKSPACE in scripts/handler.sh.
Boundary markers: None; the files are synchronized as raw content.
Capability inventory: The script uses ssh, rsync, and mkdir.
Sanitization: No validation of the content synchronized from remote peers is performed.

clawdbot-sync