hermes-skill-system

name	hermes-skill-system
description	Set up and continuously evolve a skill/agent system the disciplined way. Tracks MANY skill systems per project in a registry (`.agents/tracked-systems.md`). Three modes — DEFINE maps each tracked system (every skill + the orchestrating workflow + which node relies on what) and registers it; OBSERVE answers "what do we track, and where does each stand?" from the registry + git, changing nothing; OPTIMIZE turns a spotted flaw into one atomic, revertible `skillsys(<id>)` change via capture→route→edit→verify→approve→commit (human approves before it lands, decides whether to rerun after). The iteration log is git, not a file. Use when you spot a flaw in a skill or workflow, when a finding recurs across runs, when bootstrapping the convention into a repo, or when reviewing/consolidating recent edits. The human is the quality eye; every edit must generalize across ALL future runs — never hard-code one case, never write reward-hackable tests.
version	0.2.0
author	animation-test (informed by Nous Research "Hermes Agent" research, 2026-06-08)
metadata	{"hermes":{"tags":["meta","skill-system","self-improvement","governance","workflow"],"related_skills":["transform-workflow-to-pi","capability-registry-harness","capability-gap-filler"]}}

hermes-skill-system

The method for stewarding the skill systems that produce real artifacts. It does not grade those artifacts — . Its whole job: make every land in the right place, generalize across all future runs, and stay trackable as one atomic, revertible commit.