hermes-core

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill provides official installation commands for the Hermes Agent, which involve downloading and executing shell scripts from the project's official domain (hermes-agent.nousresearch.com). This is a standard and documented deployment method for the software this skill is specifically designed to manage.
  • [PROMPT_INJECTION]: Documentation within references/persona-context-files.md contains examples of malicious phrases (e.g., "ignore previous instructions") to advise users on patterns that may be blocked by the agent's internal safety scanners. These are educational examples and do not constitute an active injection attempt against the agent.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes a framework for managing project-specific context files (.hermes.md, SOUL.md) that are ingested by the agent to define behavior.
  • Ingestion points: Context files are loaded from the profile directory or discovered via the filesystem walk (up to the git root) as described in references/persona-context-files.md.
  • Boundary markers: The agent implements an internal security scanner that identifies injection attempts in these files and replaces the content with a safety marker.
  • Capability inventory: The agent environment managed by this skill includes access to file system operations, shell terminal commands, and network requests.
  • Sanitization: The skill documents that external content is validated against safety patterns before being processed into the session context.
  • [COMMAND_EXECUTION]: The skill includes extensive documentation for the hermes CLI tool, including health diagnostics (hermes doctor), configuration updates (hermes config set), and service management. These operations are within the expected administrative scope of the agent's core management functions.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 09:24 AM