Day 2: Memory: Skills & Self-Improvement

Day 1 got Hermes installed and working. Today is about the two systems that make Hermes different from every other agent platform: the memory system that remembers who you are and what you've done across sessions without burning tokens, and the skill system that lets Hermes write down procedures from its own experience so it gets better the longer it runs.

These aren't add-ons or power-user features. They're the mechanical reasons Hermes is more than a chatbot with tool access. By the end of today you'll understand how memory keeps context cheap, how to write skills by hand that guide Hermes's behavior, how the agent discovers and uses skills at runtime, how the curator manages the growing library in the background, and how the Skills Hub ecosystem lets you share skills with other Hermes users.

What you'll have by the end of today

• MEMORY.md and USER.md configured with the information Hermes needs to work across sessions without repeating yourself
• At least one hand-written SKILL.md loaded and working, giving Hermes a reusable procedure it follows on command
• A mental model of the skill lifecycle, from authoring to runtime loading to curator archiving, so you can debug when skills don't fire
• Understood the trust model and security guarantees of installing skills from the Skills Hub

Lessons

1. Memory Architecture, the two-tier memory system: frozen blocks for cheap, cached identity and preferences, plus session search for finding past conversations. The batch operations API, backends, and the hermes memory CLI
2. Writing Skills by Hand, the SKILL.md format, YAML frontmatter, the skill_manage tool, the write_approval gate, external skill directories, and writing your first skill from scratch
3. The Agent Skill Loop, how Hermes discovers, loads, and applies skills at runtime. Progressive disclosure (Level 0–2), the /learn and /skill commands, and skill bundles
4. The Curator Lifecycle, the background process that tracks usage, detects stale skills, archives old ones, and keeps the skill library healthy. Pinning, consolidation, and the hermes curator CLI
5. Skills Hub Ecosystem, installing community skills, trust levels, security scanning, publishing your own skills, and tapping external repositories

Prerequisites: Hermes installed and connected to a provider (Day 1). Familiarity with the CLI and slash commands helps, but isn't required for the memory architecture lesson.

Next: Day 3, The Gateway, deploying Hermes as a persistent service on messaging platforms.