The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/gdrive_to_repo_sync.py is vulnerable to shell command injection. It uses subprocess.run(shell=True) and interpolates command-line arguments directly into shell command strings without proper escaping. Specifically, in the commit_and_push function, the commit_message argument is interpolated into a git command: run_command(f"git commit -m '{commit_message}'", cwd=repo_path). A malicious commit message containing a single quote could terminate the argument and execute additional shell commands. The gdrive_folder argument is also interpolated into the rclone command: cmd = f"rclone copy manus_google_drive:{gdrive_folder} {temp_dir}", posing a similar risk.
[PROMPT_INJECTION]: The skill implements an automated pipeline for transferring files from an untrusted sandbox environment to a local machine and then to a GitHub repository. Ingestion points: Files are read from the Manus sandbox via Google Drive. Boundary markers: None provided; the skill does not instruct the user to review files before the local script commits them. Capability inventory: The script performs file system writes and git push operations. Sanitization: No validation or sanitization of file content or names is performed, allowing malicious content to be automatically committed to a repository.
[EXTERNAL_DOWNLOADS]: The script initiates downloads from Google Drive using rclone. While the remote is user-configured, the folder to download is determined by the agent's output.

autonomous-sync-script