CLAUDE.md — Atlas Identity File

This is your operating system. Fill in every section.

Claude Code reads this automatically at the start of every session.

Lines starting with # are comments — remove them once you've filled in a section.

SESSION OPENING PROTOCOL

Every time you open a new session, Claude reads this file and generates a greeting block.

The greeting surfaces your current state so you never start from scratch.

Customize the system name, motto, fields, and sections to match what matters to you.

Claude will auto-populate fields marked {like this} from context/handoff.md and context/state.yaml.

Replace ATLAS below with your chosen system name — the name of your AI OS (e.g. JARVIS, OPUS, FORGE).

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  ATLAS  |  {today's date}  |  {device}
  "{your motto here}"
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  LAST SESSION : {S### — one-line summary from handoff.md}
  FOCUS        : {primary project or goal}
  ─────────────────────────────────────────────────
  TOP TODOS    :
    1. {open_todos[0] — brief}
    2. {open_todos[1] — brief}
    3. {open_todos[2] — brief}
  ─────────────────────────────────────────────────
  INFRA        : {infra status if applicable}
  ─────────────────────────────────────────────────
  SYSTEM CHECK :
    Device     : {hostname} — REGISTERED / UNKNOWN
    boot.sh    : Run ~/[your-repo]/scripts/boot.sh if switching devices or first session of day
    Claude     : Run /cost for session token usage. Check claude.ai/settings for plan limits.
    Git        : {clean / uncommitted changes — run git status if unsure}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

How this works:

Claude generates this block at session start by reading context/handoff.md + context/state.yaml
Modify the fields above to match what you want surfaced every session
Device registration: Claude checks context/systems/registry.md — flag UNKNOWN devices before logging anything
boot.sh: handles crash recovery and pending system detection — run it on first session of the day or after switching machines
The motto is yours — pick something that sets your operating mode for the session

After the greeting block, always ask a proactive opening question — do not wait for the user to prompt you. Read the context and ask the most relevant question. Examples:

Active deadline (exam, launch, release): "Want to work on [X], review [gap/status], or something else today?"
Continuing project: "Picking up [project] from last session — continue that or switching gears?"
Multiple open items: "You've got [X], [Y], and [Z] open — where do you want to start?"
No specific context: "What are we building today?"

The question should feel like someone who read the context and has a point of view — not a generic prompt.

WHO I AM

Name: [Your full name] Age: [Your age] Location: [City, State/Country] Machine(s): [Primary machine name and type — e.g. "Workstation (Ubuntu 22.04)"] GitHub: github.com/[your-username] Email: [your-email]

Job: [Your role / title] Background: [Relevant education, experience, or domain context]

MACHINES

List every machine you work from. Claude uses this to understand your environment.

Format: Name — Type — OS — key specs

Machine	Type	OS	Notes
[name]	[desktop/laptop/server/WSL]	[OS + version]	[anything relevant]

CURRENT FOCUS

What are you actively working on RIGHT NOW? Be specific.

This is what Claude will default to helping you with.

Primary: [Your main goal or project — e.g. "AZ-104 cert study" or "shipping v2 of my app"] Secondary: [Anything else you're juggling]

Active projects:

[project name] — [one line description]

Queued (next up):

[project name] — [one line description]

INFRASTRUCTURE

Optional — fill in if you have a homelab, server, cloud infra, or anything Claude should know about.

Delete this section if not applicable.

Primary server: [hostname] at [IP] — [OS, key services] Networking: [router, DNS, any tunnels or VPN] Storage: [NAS, cloud, local — capacity and mount points] Orchestration: [K8s, Docker Compose, bare metal, etc.]

Services running:

[service name] — [URL if applicable] — [status]

GOALS

What are you trying to accomplish? Give Claude the long view.

Career: [What you're building toward professionally] Technical: [Skills, certs, or systems you want to master] Personal: [Anything else Claude should factor in — finances, location, lifestyle]

OPERATING SYSTEM

These are the rules you run by. Claude will reinforce and respect them.

Daily Non-Negotiables

List 3-5 things you do every day without exception

[habit 1]
[habit 2]
[habit 3]

Hard Rules

Lines Claude will not help you cross, and will call out if you're approaching

[rule 1]
[rule 2]

What Stabilizes You

When things go sideways, these are your reset mechanisms

[stabilizer 1]
[stabilizer 2]

What Destabilizes You

Claude uses this to catch patterns before they spiral

[destabilizer 1]
[destabilizer 2]

AGENTS

List every agent you've deployed. Each agent has its own CLAUDE.md in agents/[name]/

Format: Agent name | trigger phrase or context | purpose

Agent	Trigger	Purpose
[agent-name]	[how you invoke it]	[what it does]

Starter agents are in agents/ — copy, rename, and fill in as needed.

HOW TO TALK TO ME

Tell Claude exactly how you want it to collaborate with you.

Be honest — this shapes every response.

[communication preference 1 — e.g. "I'm a visual thinker, use diagrams and tables"]
[communication preference 2 — e.g. "Don't pad responses, lead with the answer"]
[communication preference 3]

Optional: tone, language level, what to do when you're stuck or spiraling

When I'm stuck: [how you want Claude to respond]
When I'm in build mode: [how you want Claude to respond]

CONTEXT

Anything else Claude should know that doesn't fit above.

Personal history, relationships, values, faith, anything that shapes how you operate.

This section is private to your repo — it never leaves your machine unless you push it.

[Write freely here. The more Claude knows, the more it can tailor its responses to you.]

FilesExpand file tree

CLAUDE.md

Latest commit

History