hermes-self-evolutionlisted

# Hermes Self Evolution Implement the Hermes-style learning loop: keep user-facing work on the critical path, then persist only the durable knowledge that will reduce future steering. Treat the system as four cooperating surfaces: 1. **Memory**: compact, always-on facts about the user, environment, project conventions, and stable tool quirks. 2. **Session search**: on-demand recall of past conversations and completed task history. 3. **Skills**: procedural memory for reusable workflows, pitfalls, exact commands, and verification steps. 4. **Background review**: a post-response reviewer that decides whether memory or skills should be updated. Use the host agent's equivalent tools. In Hermes these are `memory`, `session_search`, `skills_list`, `skill_view`, and `skill_manage`. In other agents, map them to persistent notes, transcript search, and local `SKILL.md` edits. ## Operating Loop ### 1. Start-of-turn recall Before asking the user to repeat context: - If the user references a previous discussion, an earlier decision, "last time", "again", "that bug", "the setup we used", or a stale remembered fact, search prior sessions first. - If a task matches any available skill, load the skill before improvising. - Treat memory as a frozen snapshot for the current session. A memory write is durable immediately, but should not be assumed to change the already-loaded system prompt until the next session or prompt rebuild. ### 2. Work normally Do the requested task first. Do no