The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading emails via himalaya message read and listing them with himalaya envelope list. This content enters the agent's context without defined boundary markers or sanitization instructions, creating a surface for indirect prompt injection where malicious email content could attempt to influence agent behavior.
[COMMAND_EXECUTION]: The skill provides a Python script template using smtplib for sending emails with attachments. This involves the agent generating and executing code that performs network operations and handles authentication.
[DATA_EXPOSURE]: The instructions guide the management of sensitive configuration at ~/.config/himalaya/config.toml, which contains email credentials and backend settings. While the skill recommends secure methods like pass, it also documents the less secure backend.auth.raw method for storing passwords in plain text.

himalaya