skills/beita6969/scienceclaw/himalaya/Gen Agent Trust Hub

himalaya

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Recommends installing the himalaya utility via Homebrew, a well-known and trusted package management service.
  • [COMMAND_EXECUTION]: Instructs the agent on how to use the himalaya CLI for legitimate email management tasks, including listing accounts, reading messages, and composing replies.
  • [CREDENTIALS_UNSAFE]: Discusses password management options, correctly recommending the use of secure external commands like pass or system keyrings. While it provides a documentation example for raw password entry, it explicitly notes that this method is not recommended for production use.
  • [DATA_EXFILTRATION]: Facilitates sending and receiving emails via standard IMAP/SMTP protocols. The MML (MIME Meta Language) composition feature allows for attaching local files by their system paths; users should be aware that any content processed by the tool into an email could potentially include such file references.
  • [PROMPT_INJECTION]: Identified a surface for indirect prompt injection due to the tool's ability to process MML tags which can trigger file reads.
  • Ingestion points: Message body content processed via himalaya template send or himalaya message write.
  • Boundary markers: None defined to isolate user content from MML instructions.
  • Capability inventory: Includes the ability to read arbitrary local files (via MML <#part> tags) and transmit them over the network (via SMTP).
  • Sanitization: No sanitization or validation of user-provided email content is specified in the instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 11:31 PM
Security Audit — agent-trust-hub — himalaya