The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to offer to run terminal commands (specifically 'wake' commands for the browser extension) on the user's computer if the extension is detected as offline. The instructions mandate that the agent must ask for user permission before executing these commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from various sources, including error codes, tool responses from the 'bridge_status' tool, and direct user input ('personSaid'). This data influences subsequent agent actions, such as the steps suggested to the user or the content of a problem report. The skill includes a 'capability inventory' involving shell execution and network reporting based on these inputs.
[DATA_EXFILTRATION]: The skill facilitates sending problem reports to the vendor's team. It includes a specific safety instruction to prevent the exposure of sensitive data, explicitly telling the agent not to include logs, tokens, or keys in the report, as the system attaches its own records.

troubleshoot