The AI agent tool Hermes created a new subagent workflow that utilizes a separate model to investigate and critique the output of a primary conversation. The /review command captures the last 10 messages and a user prompt to deploy an auxiliary model that can inspect code, check documentation, and review research before returning a verdict to the main session. This functionality enables cross-model configurations, such as using GPT for creation and Claude for auditing.
Users can direct the reviewer to focus on specific areas, including security vulnerabilities, factual claims, or requirement verification. The update aims to eliminate the need for users to manually copy and paste work between different AI sessions for verification, providing a more robust alternative to asking a single agent to verify its own work.
Key sources
- SOURCE@teknium“send a new subagent using that model tasks to review the work presented to it”x.com
- SUPPORT@hermeswatcher“GPT does the work. Claude audits it.”x.com
- SUPPORT@paulallendotuk“a proper second pair of eyes without the copy-paste ceremony”x.com
- SUPPORT@nousresearch“You can now set a new auxiliary model for Review”x.com