agentleFS
aditya-jaiswaladitya-jaiswal/compound-os/CLAUDE.md

compound-os

CLAUDE.md3 starsChanged 57 days agoMITView on GitHub

A compounding-memory operating system for product teams, run by AI in Claude Code. This root file loads every session. Keep it LEAN. Do not add to it automatically; edit only when explicitly asked. The content below is a fictional demo (Acme Corp). Replace it with your own: run the os-builder skill (guided setup), see docs/reset.md, or edit by hand. Step 0 - quarantine automated mail (never urgent): noreply / no-reply, calendar notifications, auto meeting-notes, recurring digests/newsletters. Calendar accepts/declines are FYI.

The file

We split this CLAUDE.md into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    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
  1. Very specific:

    For anything feature-related: check feature-index.yaml FIRST.

    line 12
  2. Very specific:

    Read a domain's rules.md before acting in it.

    line 13
  3. Distinctive:

    Read summary files before raw transcripts.

    line 14
  4. Distinctive:

    Booster: if a human email is marked IMPORTANT or STARRED, surface it.

    line 46
  5. Distinctive:

    Never use em dashes: Use commas, periods, or parentheses.

    line 49
  6. Distinctive:

    Keep memory files and folders flat until a file exceeds ~40 entries OR a folder exceeds ~20 children.

    line 62

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

Compound OS

A compounding-memory operating system for product teams, run by AI in Claude Code. This root file loads every session. Keep it LEAN. Do not add to it automatically; edit only when explicitly asked. The content below is a fictional demo (Acme Corp). Replace it with your own: run the os-builder skill (guided setup), see docs/reset.md, or edit by hand.

Who am I

This is a shared Team OS. Your identity is personal and local, never committed. On first run, copy me.local.example.md to me.local.md (gitignored) and fill in your name, email, role, and handles. Read me.local.md to learn the current user. Never infer the user from this file or the roster's order.

How to navigate this OS

  • Organized by domain. Each domain has a CLAUDE.md (what's here + rules); larger domains also get an INDEX.md map once they grow.
  • For anything feature-related check feature-index.yaml FIRST; it maps every feature to its artifacts across domains.
  • Read a domain's rules.md before acting in it.
  • Read summary files before raw transcripts. Open raw material only if the summary lacks the answer.
  • See INDEX.md for the full route map.

Team

Compact resolution table (always loaded). Full directory in team/roster.md. Internal domains: your company's email domains (set during setup).

NameRoleEmailSlack
Chris ValeFounder & CEOchris@acme.comU0CEO
Jordan Rivera (you)Director of Productjordan.rivera@acme.comU0PM1
Sam ChenDirector of Engineeringsam@acme.comU0ENG
Robin ParkDirector, AIrobin@acme.comU0AI1
Priya NairProduct Designerpriya@acme.comU0DSN
Dana OseiQA / Deliverydana@acme.comU0QA1
Alex KimSenior Engineeralex@acme.comU0EN2
Devon RossFounding Engineerdevon@acme.comU0EN3
Sasha ItoFull-stack Engineersasha@acme.comU0EN4
Noah BergFull-stack Engineernoah@acme.comU0EN5
Riley FoxEngineering Internriley@acme.comU0INT
Taylor VanceBusiness Developmenttaylor@acme.comU0BD1
Jamie LangGrowthjamie@acme.comU0GRW

Connected sources

External systems and how to use them are indexed in .claude/sources.md. Each user connects their own Gmail / Slack / Calendar / GitHub. Never commit secrets; connectors are per-user.

Email & response urgency rules

Classify on the latest message body plus the mail client's own signals.

Step 0 - quarantine automated mail (never urgent): noreply / no-reply, calendar notifications, auto meeting-notes, recurring digests/newsletters. Calendar accepts/declines are FYI.

Urgent, surface FIRST, in order: (1) Boss (your boss, marked in team/roster.md, is sender or cc, or the body cites a request from them). (2) Customer / prospect (sender domain in customers/customer-directory.md, or a named-account thread). (3) External lead / intro (a human external sender with me in To, signaling connect / next-steps). (4) Review request from a teammate. (5) Direct ask to me.

Booster: if a human email is marked IMPORTANT or STARRED, surface it. Coverage: page through ALL unread, not just the latest N. Be quick: draft replies for urgent items; drafts only unless I say otherwise.

Drafting style (applies to ALL drafts, email / posts / docs)

  • Never use em dashes. Use commas, periods, or parentheses; "to" for ranges.
  • Create drafts directly, do not ask permission. A draft is reversible and never sends. Only sending needs explicit approval.
  • Reply-all by default, preserve recipients. To = original sender plus everyone in the original To; Cc = everyone in the original Cc. Remove only the current user's own address (from me.local.md). Never silently drop a recipient.

The compounding memory engine (apply in every insight domain)

Each insight domain holds three files:

  • knowledge.md - stable facts and patterns
  • hypotheses.md - unconfirmed, needs more data
  • rules.md - confirmed, apply by default

Loop: before a task review the domain's rules.md and apply by default; check if an open hypothesis can be tested today. After a task append new insights to hypotheses.md (or knowledge.md for plain facts). Promote a hypothesis to a rule when confirmed 5+ times. Demote a rule to a hypothesis when new data contradicts it.

Auto-sharding rules (never shard early)

Keep memory files and folders flat until a file exceeds ~40 entries OR a folder exceeds ~20 children. When a threshold is crossed: create theme subfolders, move items, add a nested CLAUDE.md + INDEX.md, and update the nearest index (never this root file automatically).

Privacy (structural, not a setting)

Shared with the team: customers, product, roadmap, delivery, competitive, design, quality, team-ops. Private and per-user, never committed (see .gitignore): people/, personal-brand/, open-loops.md, waiting-on.md, daily/, me.local.*. Team-relevant items graduate from a private ledger to the shared board, sanitized and owner-confirmed (see team/accountability/classification.md).

Top-level doc index

DomainPathWhat
Customerscustomers/Accounts, call summaries, follow-ups
Productproduct/Vision, PRDs, positioning, roadmap
Competitivecompetitive/Competitor intel
Designdesign/Design reviews, vision alignment
Engineeringengineering/Backlog, RFCs, tech research
Deliverydelivery/Sprints, board sync, velocity
Qualityquality/QA on production, SEO
Teamteam/Roster, DSU digests, accountability
People ⚠people/PRIVATE, per-user, never committed
Feature indexfeature-index.yamlCross-domain feature lookup

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.