The Agent Skills Directory

[SAFE]: The skill is a legitimate integration for the open-source Himalaya email CLI (github.com/pimalaya/himalaya). Its functionality matches its stated purpose.
[DATA_EXFILTRATION]: While the skill accesses sensitive data (emails and account configuration), this is its primary intended function. There is no evidence of data being sent to unauthorized third-party domains; network activity is restricted to the user's configured IMAP/SMTP servers.
[CREDENTIALS_UNSAFE]: The configuration documentation provides examples for setting up authentication but correctly uses placeholders (e.g., 'your-password') and explicitly recommends secure alternatives like system keyrings or the 'pass' password manager.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of incoming email content. This constitutes a standard surface for indirect prompt injection common to all email-integrated AI tools.
Ingestion points: Email bodies and headers retrieved via himalaya message read and himalaya envelope list.
Boundary markers: The instructions do not define specific delimiters for separating email content from agent instructions.
Capability inventory: The agent has the ability to write/send emails, delete messages, and download attachments using the CLI tool.
Sanitization: Not explicitly defined; relies on the underlying CLI tool's handling of MIME content.
[COMMAND_EXECUTION]: The skill uses the himalaya binary to interact with email servers. All shell commands described are standard usage patterns for the tool and do not involve unsafe interpolation of untrusted input into administrative shell contexts.

himalaya