Edventure AI Onboarding Kit

What is AGENTS.md?

Hermes automatically discovers and loads context files that shape how it behaves in a given project. AGENTS.md is the primary one: a Markdown file that tells the agent everything specific to this repo.

What goes in it

Where it lives, and how it’s found

AGENTS.md sits in your project root (and optionally subdirectories) — not in ~/.hermes. At startup Hermes walks up to the git root to find it and loads it into the system prompt. As the agent moves into subfolders during a session, it progressively discovers any deeper AGENTS.md files and pulls them in when relevant.

The loading cascade

Only one project-context file is loaded per session — first match wins:

.hermes.md  →  AGENTS.md  →  CLAUDE.md  →  .cursorrules

So AGENTS.md is the common default, and the format is shared across agent tools (Claude Code reads CLAUDE.md, Cursor reads .cursorrules) — same idea, different filename.

AGENTS.md vs SOUL.md

They load independently and do different jobs:

SOUL.mdAGENTS.md
ScopeGlobal — the agent’s identityProject — this repo’s rules
Lives in~/.hermes/ (HERMES_HOME)the project root
AnswersWho is the agent?How does this project work?

The rule of thumb: if it should follow you everywhere, it’s SOUL.md; if it belongs to a project, it’s AGENTS.md. SOUL.md is always loaded as the identity; AGENTS.md is loaded on top when you’re working in that project.

Why it matters: the agent stops guessing your project’s conventions and starts following them — the single highest-leverage file for getting useful output instead of generic output.

Try it: drop an AGENTS.md into latex-starter/ (open in VS Code at localhost:8080) describing the Flue agent + render tool and the “no Manim” rule, then ask Hermes or Claude Code to extend the app — watch it respect the conventions.

Docs: Hermes — context files.