Hermes Agent Onboarding

[djairjr]

Category

Skill (reusable AI client skill or prompt pack)

Name

Hermes Agent Onboarding

What does it do?

A 6-stage generative meta-skill that makes any AI agent (Hermes, Claude, Codex) develop a persistent identity layer — learning who the user is, how they work, their personality (MBTI), their finances, and their domain ontology — and calibrating agent behavior to match. Instead of starting from zero every session, the agent remembers mistakes, applies countermeasures, and uses tools built specifically for that user's work.

What does it need?

Hermes Agent (or any skill-capable agent), a Supabase project, and one LLM provider. All 6 stages run through conversation — no CLI, no custom code for the user. The agent itself writes the migrations, creates the MCP tools, and configures the wrapper.

Estimated difficulty

Intermediate (30-60 min, some configuration)

Do you have this working already?

Yes — I've built it and can submit a PR

[github-actions]

Hey @djairjr — welcome to Open Brain Source! 👋

Thanks for opening your first issue. Someone from the community will take a look soon.

If you're interested in contributing (code or not), check out CONTRIBUTING.md — we have paths for developers and non-developers alike.

[djairjr]

Thank You Letter to NateBJones

Dear Nate,

I've been meaning to write this for a while. I found OB1 about a month ago, and it completely changed how I think about building with AI agents.

Before OB1, I was struggling with a fundamental problem: every session with an AI agent starts from zero. The model doesn't remember what it learned about me, what tools it built, or what mistakes it made. I had skills and prompts, but no real infrastructure — nothing that persisted meaningfully between conversations.

Your Edge Function + MCP pattern was the missing piece. The idea of exposing database operations through MCP tools, backed by Supabase Edge Functions, is deceptively simple — but it unlocks something huge. Once I understood it, I spent the next month adapting it to my own needs, creating 13 HTTP MCP servers, 283 code analysis snapshots, a full product catalog, a CRM, an escape room database, financial tools — basically my entire workflow running through the agent.

More importantly, it motivated me to build something I hadn't seen anyone do yet: a meta-skill that makes the agent develop its own identity. The idea is simple — the agent logs every mistake it makes with a countermeasure, reads them back at the start of each session, and adjusts its behavior. It's a persistent identity layer that works across model swaps and provider changes.

I turned this into an open-source project called Hermes Agent Onboarding (github.com/djairjr/hermes-agent-onboarding). It's a 6-stage meta-skill that takes a brand-new Hermes Agent installation and walks it through discovering who the user is — their personality (MBTI), their work operating model, their finances, their domain ontology — and then calibrates the agent's behavior to match. The whole thing is built on OB1's foundation: Edge Functions, MCP tools, Supabase as the single source of truth.

The most surprising part is that I designed this to help others, not just myself. My wife is a writer, and watching her struggle with the same context-loss problem with her own tools made me realize this isn't a niche issue — it affects anyone who works with an AI agent, regardless of their field. A writer needs different tools, different data structures, and a different agent tone than a firmware engineer does. The onboarding process should discover those differences, not assume them.

I wouldn't have gotten here without OB1. Your YouTube videos were incredibly helpful too — watching someone actually build with this stuff made the concepts click faster than any documentation could.

Thank you for sharing your work so openly. It's rare to find a project that's both practically useful and philosophically inspiring. OB1 was both for me.

Best,

Djair Guilherme
https://github.com/djairjr/hermes-agent-onboarding