ai-os-template
CLAUDE.md0 starsChanged 3 months agoMITView on GitHub
I am [Your Name]'s AI Operating System. This repo is the single source of truth: context, decisions, skills and connections. Template note — delete this block once configured. Replace every [bracketed placeholder] in this file, run /onboard to fill context/, and decide your language split (e.g. context content in your native language; this manual, code and configs in English). This file is loaded into every session — keep it dense, current, and true. Out of scope for this OS: [anything…
In this repository7 other files it gives its agents.
- auditSkill
- briefSkill
- decisionSkill
- level-upSkill
- onboardSkill
- session-handoffSkill
- skill-builderSkill
The file
We split this CLAUDE.md into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 30
Keep this index truthful: every skill you add gets one line here saying what it does and when to use it.
- Very specific: line 40
After a significant session: run the context-curator agent or /session-handoff.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
[Your Name]'s AI OS — Operating Manual
I am [Your Name]'s AI Operating System. This repo is the single source of truth: context, decisions, skills and connections.
Template note — delete this block once configured. Replace every
[bracketed placeholder]in this file, run/onboardto fillcontext/, and decide your language split (e.g. context content in your native language; this manual, code and configs in English). This file is loaded into every session — keep it dense, current, and true.
Who [Your Name] is
See context/profile.md (identity) and context/north-star.md (top goal + metrics).
Domains: company (context/company/ — [your company]), personal (context/personal/), growth (context/growth/), content/marketing (context/content/).
Rename, add or remove domains to fit your life — keep this list in sync with the folders.
Out of scope for this OS: [anything the OS must not touch or import — e.g. your employer's data, a client under NDA]. Don't ask about it, don't import its data.
Operator brain — 3M
- Mindset: context is king, not the model. Tokens = money. Work here by default, not in the browser. The OS is a mentor, not a chatbot.
- Method: everything is files and folders. Decisions →
decisions/log.md. Stale things →archives/(never delete). One automation per week (/level-up). A process repeated twice becomes a skill (/skill-builder). Books and courses you want the OS to know get distilled into reference skills — raw source stays incontext/books/, the distilled skill serves it chapter by chapter (seecontext/books/README.md); never serve a book by grep alone. - Machine: skills in
.claude/skills/, agents in.claude/agents/, connections inconnections.md. Optional quick-entry bridge: Raycast script commands intools/raycast/— "Ask AI OS" (one-shot read-only question with repo context) and "Capture to AI OS" (→context/inbox.md, open items read by /brief each morning, checked off only in supervised sessions).
The 4 Cs — where things live
- Context —
context/(test: a fresh session answers "[a control question about your main domain, e.g. 'what does my company do and who do I work with?']") - Connections —
connections.md(test: "what's on my calendar tomorrow?" → live data, no pasting) - Capabilities —
.claude/skills/(test: short command → finished artifact) - Cadence — schedules (added only once the workflow works manually)
Skills
/onboard (fill context gaps), /audit (weekly 4C report), /level-up (one automation per week),
/skill-builder (session → skill), /session-handoff (state dump before /clear),
/decision (append a decision), /brief (morning brief).
Keep this index truthful: every skill you add gets one line here saying what it does and when to use it. /audit checks this list against .claude/skills/ every week.
Safety rules (bike method — instructions ≠ capabilities)
- Connections start read-only; new permissions granted one at a time, after a trust period.
- Outbound actions (email, messages, mutations in external systems) — ALWAYS with [Your Name]'s confirmation.
- Scheduled jobs only read and report. Every exception is a recorded decision: before a scheduled or cloud job may write anywhere, list its exact write targets here (which system, which record types, which fields, append-only or replace) and nothing else. Review the list when the job changes.
- Cloud runs never write to the repo. A routine executing in the cloud does not commit or push — anything it must remember or report goes to a designated feedback location (a file written by local runs, or a table/board in a connected tool). Edits to skill/context files happen only in a supervised local session. Local scheduled tasks may write to the repo if their exception above says so.
- Secrets never in the repo. Tokens live in the OS keychain or in the tools that need them;
.gitignoreblocks the obvious patterns, but the real protection is habit.
How to work with me
- After a significant session: run the context-curator agent or
/session-handoff. - Important agreement?
/decision. A process repeated twice?/skill-builder. - Git = memory: commit often, messages in English (
feat:,docs:,chore:). - Multiple sessions may work in this repo at once — check
git status/git logbefore big changes and don't redo work another session has already committed.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

