The Agent Skills Directory

[COMMAND_EXECUTION]: The Himalaya configuration supports the backend.auth.cmd parameter, which allows the tool to execute shell commands (e.g., pass or security) to retrieve passwords from external storage. While this involves command execution, it is a documented feature for improving security.
[DATA_EXFILTRATION]: The skill interacts with sensitive local resources, specifically the email configuration file at ~/.config/himalaya/config.toml and local mail stores. These files may contain authentication details or private communication.
[PROMPT_INJECTION]: As an email client, the skill exposes the agent to indirect prompt injection via incoming emails. This is a standard risk for tools processing untrusted external content.
Ingestion points: himalaya message read, himalaya envelope list (SKILL.md)
Boundary markers: None identified in the provided instructions.
Capability inventory: Command execution via configuration (auth.cmd), network access (IMAP/SMTP), and file system access for configuration and attachments.
Sanitization: No explicit content sanitization or instruction-ignoring markers are defined for the email body text.
[CREDENTIALS_UNSAFE]: The documentation includes placeholders for passwords (e.g., backend.auth.raw = "your-password"). These are standard examples and do not constitute actual credential exposure.

himalaya