hermes-outbound
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references the official Hermes Agent repository (github.com/NousResearch/hermes-agent) for version checks and documentation. These are legitimate resources for the agent's ecosystem.
- [COMMAND_EXECUTION]: The skill instructs the user or agent to perform standard maintenance commands (
hermes --version,hermes update --check) to ensure environment compatibility. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection as it processes outbound drafts and attachments (e.g., PDFs). However, the skill is explicitly designed to mitigate this risk by implementing a mandatory "trace scan" and checklist to identify and remove machine-generated artifacts, internal notes, and system traces before they reach a human recipient. It uses clear boundary markers in its preview template to separate internal commentary from the final payload.
- [DYNAMIC_EXECUTION]: A Python script (
scan.py) is provided as a template for scanning text files for internal traces. The script uses only the Python standard library (re,argparse,pathlib) and performs safe string pattern matching without the use of dynamic evaluation or network access.
Audit Metadata