hermes-harness
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides commands to download and execute installation scripts directly from the official Nous Research domain. Examples include "iex (irm https://hermes-agent.nousresearch.com/install.ps1)" for Windows and a curl pipe to bash for Linux/WSL environments.
- [COMMAND_EXECUTION]: The skill documentation includes instructions for using the
hermesCLI with flags like--yoloand--accept-hooks, which are intended to bypass user confirmation prompts for dangerous or sensitive operations during script execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes data returned from the
hermesCLI, which serves as a surface for indirect instructions. - Ingestion points: The host agent reads and interprets stdout and stderr from
hermes -zandhermes chatcommands (documented in SKILL.md). - Boundary markers: No explicit delimiters or instructions to ignore embedded commands within the tool output are provided in the skill instructions.
- Capability inventory: The skill facilitates full execution of the Hermes Agent, which has capabilities for file system access and tool invocation.
- Sanitization: The instructions do not describe any filtering or validation of the CLI output before it is ingested by the agent context.
Audit Metadata