hermes-agent-framework
Pass
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill serves as an educational guide and reference for an open-source framework. No malicious instructions, obfuscation, or credential harvesting patterns were found.
- [EXTERNAL_DOWNLOADS]: The skill contains instructions to clone the framework from its official GitHub repository (github.com/NousResearch/hermes-agent) and install necessary dependencies. These references target official project resources and represent standard developer workflows.
- [PROMPT_INJECTION]: The documented architectures describe agents that ingest untrusted data (web content, PDFs, file systems) and utilize high-impact tools (code execution, database access). This identifies a potential attack surface for indirect prompt injection in agents developed using these instructions.
- Ingestion points: Web search results, local file system reads, and PDF document processing as described in SKILL.md.
- Boundary markers: While the documentation mentions the use of constraints, it does not provide explicit implementation examples for delimiting untrusted data in the code snippets provided.
- Capability inventory: The framework supports arbitrary code execution (Python, JavaScript, Bash), file system modifications, and persistent database operations.
- Sanitization: The documentation explicitly mentions and recommends the use of sandboxed environments for code execution tools.
Audit Metadata