Hermes Agent — Self-Improving AI Agents

Overview

Inspired by NousResearch/hermes-agent, this skill helps you build agents that grow with usage — capturing feedback, reflecting on their own behavior, and updating their instructions over time.

Unlike static assistants, a Hermes-style agent maintains a living system prompt. After each interaction, it evaluates its own performance, extracts lessons, and writes improvements back to its configuration.

Core Concepts

• Self-reflection loop: After each task, the agent evaluates what went well and what didn't
• Instruction update: Agent proposes changes to its own system prompt based on feedback
• Memory layer: Facts about the user and context are persisted into future conversations
• Feedback signals: Explicit (thumbs up/down) and implicit (did the user ask for clarification?)

Architecture

User Message -> [Memory Retrieval] -> [Agent + System Prompt] -> [Response]