The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/clash_tun_doctor.py utilizes subprocess.run to manage the lifecycle of the Clash Verge Rev application on macOS. It executes system utilities such as osascript (to quit the application), open (to restart it), and pgrep (to monitor process status). The arguments for these commands are derived from fixed constants, which effectively prevents command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The direct-list feature ingests and parses external service logs to identify hosts for network routing rules. While this is a potential attack surface for indirect prompt injection, the skill mitigates the risk by validating all extracted hostnames against a strict hostname regex (HOSTNAME_RE) and using safe formatting via json.dumps when writing new configuration rules.
[SAFE]: The skill adheres to security best practices by defaulting to read-only operations. Any modifications to the user's system require the explicit use of the --apply flag. Furthermore, the skill automatically generates timestamped backups in a dedicated directory before modifying any configuration files and includes a built-in rollback mechanism to revert changes.
[SAFE]: Data access within the skill is limited to specific Clash Verge Rev configuration files and a local Unix domain socket for API communication. The script uses targeted parsing to extract only the necessary settings required for diagnosis, ensuring that unrelated sensitive information is not unnecessarily exposed or processed.

lov-clash-tun-doctor