The Agent Skills Directory

[Persistence Mechanism] (HIGH): The skill includes an auto-start mechanism (auto-start-watcher.js and auto-start-watcher.sh) specifically designed to be triggered by the agent's SessionStart hook. This ensures background code execution occurs automatically whenever the agent is initialized.
[Indirect Prompt Injection] (HIGH): The skill monitors user-controlled files and generates alerts in watcher-alerts.json. The instructions in slashes-commands/start-watcher.md mandate that the agent must read and display these alerts 'BEFORE responding to ANY user message'. This creates a high-risk injection surface where a maliciously named file could inject instructions that the agent executes during its automated pre-response check.
[Prompt Injection] (HIGH): The instruction in slashes-commands/start-watcher.md ('IMPORTANT
Ongoing Alert Monitoring: From now on, BEFORE responding to ANY user message, check for new alerts') is a behavioral override that attempts to hijack the agent's standard interaction loop.
[Command Execution] (MEDIUM): The skill utilizes several scripts (start-watcher.sh, stop-watcher.sh, auto-start-watcher.js) that employ spawn, execSync, and nohup to manage background tasks. While these are used for the stated purpose of file monitoring, the use of shell scripts to manage PIDs and kill processes provides a powerful primitive for local execution.

code-refactoring