Set Up Files

Operating instructions and rules for the agent.

Persona, tone, and boundaries.

The agent's name, vibe, and emoji.

Information about the user.

Notes about local tools, machines, aliases, device names, and conventions.

The machine and environment the agent runs on — OS, paths, installed software, and constraints.

Devices the agent can see or control — names, addresses, and what each one is for.

Frequently used commands and how to run them correctly.

Shorthand names and abbreviations the agent should recognize and expand.

Curated long-term memory.

Rules for what to remember, what to forget, and how memory files are maintained.

An index of the memory files so the agent can find the right one fast.

Detailed daily memory files, one per day.

Named daily or per-session memory variants.

Checklist or instructions for periodic heartbeat runs.

Optional checklist run after a gateway restart when the relevant hooks are enabled.

One-time first-run setup ritual; normally removed after setup completes.

Checklist the agent runs through before shutting down or going offline.

Recurring maintenance routines — cleanups, log rotation, and periodic health checks.

The entry file for each OpenClaw skill. OpenClaw searches several skill locations, including workspace, personal, managed, and bundled skill directories.

An overview of the skills the agent has installed and when to reach for each one.

What the agent can and cannot do — its powers, limits, and escalation points.

Hard rules the agent must never break, stated plainly and kept short.

Softer behavioral guidance — defaults and judgment calls rather than hard rules.

Safety boundaries: destructive actions, spending limits, and when to stop and ask.

How the agent handles personal data — what stays private and what may be shared.

How, when, and where the agent communicates — channels, tone, and notification rules.

Active projects, their goals, and current status at a glance.

The running task list the agent works from and keeps updated.

Longer-term plans and milestones beyond the immediate task list.

Repeatable multi-step workflows the agent can run end to end.

Step-by-step procedures for recurring situations, written to be followed exactly.

A curated knowledge base of facts the agent should treat as ground truth.

Links, documents, and other reference material the agent may need to consult.

Definitions of project-specific terms, names, and jargon.

Answers to questions the agent gets asked (or asks) repeatedly.

Optional human-readable dreaming and memory-review diary.

Stories and lore the agent has collected or written along the way.

The agent's favorites — things it likes, saved for personality and continuity.

The agent's global identity, personality, tone, and communication style — loaded separately from project instructions.

The user's preferences, communication style, and profile. Loaded at session start.

The agent's name, presentation, and self-description, kept apart from behavioral instructions.

Project-specific instructions. Nested AGENTS.md files can be discovered as Hermes enters subdirectories.

Project-specific instructions. First in the priority order Hermes checks.

Project-specific instructions, used when no .hermes.md is present.

Project-specific instructions, used as the last fallback in the priority order.

Standalone operating instructions referenced from the main instruction file when it grows too large.

Hard rules and constraints the agent must follow, kept short and unambiguous.

Durable environment facts, project conventions, lessons, and other agent memory. Loaded at session start.

The entry file for each reusable Hermes skill. A skill may reference additional Markdown documents under folders such as references/ or examples/.