The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Python scripts for watchlist validation and state management. The agent executes these scripts via a shell, using double-quoting for user-provided variables like '<watchlist_path>'. This demonstrates a standard and relatively secure method for shell-based tool integration within the agent's environment.
[EXTERNAL_DOWNLOADS]: Web search and content extraction are performed exclusively through the Tavily API tools. The skill includes strict design principles that forbid the use of more general-purpose tools such as 'curl' or full browsers, thereby limiting the agent's external data access to the specific requirements of the monitoring task.
[DATA_EXFILTRATION]: All sensitive data, including search results and tracking state, is handled within a restricted local workspace ('/sandbox/.openclaw/workspace'). The instructions explicitly direct the agent not to expose full digest contents in the final chat response, providing a layer of protection against accidental data leakage.

watchtower