The Agent Skills Directory

[COMMAND_EXECUTION]: The script status.sh unsafely interpolates shell variables into Python code blocks executed via python3 -c. The functions reverse_file and snapshot_fixture_get insert variables such as $file and $path directly into single-quoted Python strings, which can lead to code injection if these variables contain single quotes or malicious snippets.
[REMOTE_CODE_EXECUTION]: There is a potential for arbitrary code execution through log poisoning. In the check_snapshotter_health function, the script extracts a timestamp from the wiki-snapshot.log file and places it inside a Python execution block: start = datetime.fromisoformat('$completion_ts'...). Since an attacker may be able to append data to the system logs, they could inject Python code that executes when the agent runs the health check.
[DATA_EXFILTRATION]: The skill performs outbound network operations to verify reachability. It uses git ls-remote, rclone lsf, and ssh to probe connectivity to GitHub, S3 remotes, and snapshotter hosts. While these are documented health check features targeting well-known services, they constitute external network activity.

vault-sync-status