hermes-agent-self-evolution
Pass
Audited by Gen Agent Trust Hub on May 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Downloads the evolution framework from the NousResearch GitHub repository.
- [REMOTE_CODE_EXECUTION]: Generates and executes new skill and prompt variants during the optimization cycles.
- [COMMAND_EXECUTION]: Instructs the user to run installation and execution commands via the terminal.
- [DATA_EXFILTRATION]: Reads session history from the local database at ~/.hermes/sessions.db to use as evaluation data for API-based optimization.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by ingesting untrusted session data that influences the generation of future agent instructions.
- Ingestion points: Reads interaction data from ~/.hermes/sessions.db.
- Boundary markers: None identified in the optimization pipeline documentation.
- Capability inventory: Executes shell commands, generates code variants, and creates GitHub Pull Requests.
- Sanitization: No input sanitization or validation of session data is described.
Audit Metadata