hermes-model-management
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill provides a diagnostic 'Verification Recipe' that executes a Python script via
python -cusing the agent's virtual environment. This script dynamically modifies the import path to access internal modules and calls validation functions to verify model availability. This is a controlled execution of local code for troubleshooting purposes. - [EXTERNAL_DOWNLOADS]: The documentation describes how the agent fetches a curated model catalog from
hermes-agent.nousresearch.comandgithub.com/NousResearch. These are official sources for the Hermes agent project. - [INDIRECT_PROMPT_INJECTION]: The skill defines workflows where the agent accepts model IDs from user input to modify its configuration or perform validation.
- Ingestion points: User input in chat requesting specific models (processed via
SKILL.mdinstructions). - Boundary markers: Not explicitly defined in the instructions.
- Capability inventory: The agent can modify local configuration files (
~/.hermes/config.yaml) and execute Python validation logic. - Sanitization: The skill notes that the internal validation logic automatically rejects model IDs containing whitespace, which serves as a basic safety filter.
Audit Metadata