How a Personal OS
Compounds Over Time

Setup vs. Compounding: What Changes

The biggest shift is moving from telling Claude what to do to Claude already knowing what to do. Here's what that looks like:

Auto Memory (How It Really Works)

This is the part most people miss. Claude Code has two complementary ways to carry knowledge across sessions: CLAUDE.md (instructions you write) and auto memory (notes Claude writes itself). The first guide covered CLAUDE.md. This section is about the second one — the layer that turns Claude into a coworker who learns from how you actually work.

CLAUDE.md vs. Auto Memory

Use CLAUDE.md when you want to instruct. Auto memory lets Claude take notes for itself based on your corrections, without you writing anything.

Where Auto Memory Lives

Each project gets its own auto memory directory:

~/.claude/projects/<project>/memory/

The <project> path is derived from the git repository, so all worktrees and subdirectories within the same repo share one auto memory directory. Outside a git repo, the project root is used instead.

Inside that folder, two things matter:

• MEMORY.md — an index file. The first 200 lines or 25KB (whichever comes first) load into Claude's context at the start of every session.
• Topic files — additional markdown files Claude creates as memories accumulate. These are lazy-loaded: Claude reads them on demand when something in the current conversation makes them relevant.

The /memory Command

Run /memory in any Claude Code session to:

• See which CLAUDE.md and rules files are loaded in your current session
• Toggle auto memory on or off
• Open the auto memory folder in your editor to browse, edit, or delete files

Auto memory is plain markdown. You can read and edit any file by hand — Claude treats your edits as the source of truth.

How Auto Memory Gets Written

Claude decides what's worth remembering based on whether the information would help in a future conversation. The common triggers:

• You correct Claude. "Don't do that again" tends to become a durable note.
• You confirm a non-obvious choice was right. Worth saving so Claude doesn't second-guess the same call next month.
• You teach Claude a durable fact. Vendor relationships, process details, who handles what.
• You say "remember this." Direct trigger.

When you see "Writing memory" or "Recalled memory" in the Claude Code interface, Claude is actively updating or reading from your auto memory folder.

How I Organize Mine

Anthropic doesn't prescribe a file structure or frontmatter format for auto memory — Claude writes whatever's useful. Over time I've settled on a convention that makes the index easier to scan:

The convention is mine, not the system's. Claude doesn't care what you name files — the prefixes just make the index easier to read at a glance.

What a Real Auto Memory File Looks Like

This is one of my feedback memories:

REAL AUTO MEMORY EXAMPLE

---
name: feedback-no-stopping-suggestions
description: Never suggest stopping, pausing, or saving work for later.
  She decides when to stop.
type: feedback
---

Never suggest stopping, pausing, deferring, or "saving for later" —
including soft framings like "clean stopping point," "pause here,"
"call it for tonight," or "fresh eyes tomorrow."

**Why:** I flagged this directly after Claude repeatedly offered
exit ramps during a productive Saturday session. The pattern was
Claude pattern-matching on session length as a signal I should stop.
That signal is wrong. Time of day and accumulated progress carry
zero information about my capacity to continue.

**How to apply:**
- Continue executing. When one piece finishes, offer the next concrete
  option.
- If multiple options exist, present them as choices to make.
- If I want a break, I'll say so directly.

What I Actually Have in Memory

My auto memory folder right now has 60+ files. They cluster into:

The Skills Ecosystem

The first guide showed two skills: /today and /shutdown. A mature Personal OS has 30+. Each one is a folder inside ~/.claude/skills/ with a SKILL.md file describing what it does and when to use it. Claude reads the descriptions and triggers automatically when a request matches. The / prefix is shorthand to force a specific skill to run.

What I Actually Run

Here's how mine cluster, by job:

How a Skill Is Structured

A SKILL.md file looks like this:

---
name: call
description: Create or update a call note in 20_Areas/Calls/.
  Auto-populates People and Companies directories.
allowed-tools: Glob(*), Read(*), Write(*), Edit(*)
---

# /call

Create a structured call note.

## What to do

1. Ask for contact name and date (default today).
2. Check if the person exists in 20_Areas/People/. If yes, read their
   profile for context. If no, create from the template.
3. Check the contact's company in 20_Areas/Companies/. If exists,
   update Engagement Timeline; if not, create from template.
4. Create the call note in 20_Areas/Calls/YYYY-MM-DD_Contact-Name.md
   using the call schema in CLAUDE.md.
5. Surface any prior notes with this contact from past calls.

Install the Core Five

The first guide installed /today and /shutdown. Past those, these five do the most work in a typical week: meeting capture, pre-meeting prep, auto memory hygiene, Monday kickoff, and Friday wrap-up. Paste this into Claude to install all five at once.

Please install five skills for me. Each lives at ~/.claude/skills/<name>/SKILL.md.

1. ~/.claude/skills/call/SKILL.md

---
name: call
description: Create or update a structured call note in 20_Areas/Calls/. Auto-populates People and Companies entries.
allowed-tools: Glob(*), Read(*), Write(*), Edit(*)
---

# /call

Create a structured call note.

## What to do

1. Ask for the contact name and date (default today).
2. Check 20_Areas/People/ for the contact. If found, read their profile for context. If not, create one from the template.
3. Check 20_Areas/Companies/ for their company. If exists, update Engagement Timeline. If not, create from template.
4. Create the call note at 20_Areas/Calls/YYYY-MM-DD_Contact-Name.md.
5. Surface any prior notes with this contact.
6. Append a Relationship History line to the person's profile.


2. ~/.claude/skills/brief/SKILL.md

---
name: brief
description: Generate a one-page pre-meeting brief for a person, company, or topic. Pulls vault context for a structured briefing.
allowed-tools: Glob(*), Read(*), Write(*)
---

# /brief

Generate a one-page brief.

## What to do

1. Ask what or who the brief is about.
2. Search 20_Areas/People/, 20_Areas/Companies/, and 20_Areas/Calls/ for relevant context.
3. Synthesize into a brief with: who/what, relationship history, open items, recommended angles for the meeting.
4. Save to 00_Inbox/YYYY-MM-DD_Brief_Topic.md and tell me where it is.


3. ~/.claude/skills/remember/SKILL.md

---
name: remember
description: Review the current session and save durable learnings to auto memory.
allowed-tools: Read(*), Write(*), Edit(*)
---

# /remember

Review what happened in the session and propose memory updates.

## What to do

1. Scan the conversation for: corrections I gave, durable facts I shared, non-obvious choices that worked, patterns worth keeping.
2. For each candidate, propose a memory file name and a one-line summary.
3. Ask which to save; write the approved ones to ~/.claude/projects/<this-project>/memory/.
4. Update MEMORY.md index with new entries.


4. ~/.claude/skills/weekstart/SKILL.md

---
name: weekstart
description: Monday kickoff. Daily agenda, overdue triage, project health, people follow-ups, weekly focus.
allowed-tools: Bash(*), Read(*), Write(*)
---

# /weekstart

Monday morning routine.

## What to do

1. Run the same scan as /today to build today's agenda.
2. Scan 10_Projects/ for projects with no activity in 7+ days. Flag them.
3. Scan 20_Areas/People/ for entries where last_contact is more than 14 days ago and follow_up is true.
4. Ask me for a one-line focus for the week.
5. Save everything as 00_Inbox/YYYY-MM-DD_Weekstart.md.


5. ~/.claude/skills/weekend/SKILL.md

---
name: weekend
description: Friday wrap-up. Weekly review, unresolved items, next week preview.
allowed-tools: Bash(*), Read(*), Write(*)
---

# /weekend

Friday wrap-up.

## What to do

1. Read this week's daily agenda files in 00_Inbox/.
2. Summarize what moved, what stalled, and what got committed to.
3. Pull open action items into a single list.
4. Propose 3 priorities for next week.
5. Save as 00_Inbox/YYYY-MM-DD_Weekend.md.


Create all five now and confirm when done.

Capture Pipelines: The Input Layer

The first guide treated the vault as a place you write into. The leverage move is to make it a place that fills itself from your real meetings. By the time you sit down to ask Claude "what did I learn this week?", the raw material is already there.

The lowest-effort starting point is meeting transcripts, and the easiest path is Granola.

Granola for Meeting Capture

Granola sits on top of Zoom, Google Meet, and Microsoft Teams and captures transcripts plus AI-generated summaries for every call. It ships with a built-in Claude connector, so once you connect it, Claude can pull recent meetings on demand without you copying anything.

Two parts to set up:

Please install a Granola filing skill for me.

Save this as ~/.claude/skills/granola/SKILL.md:

---
name: granola
description: Pull recent Granola meeting transcripts and file them into 20_Areas/Calls/ as structured call notes. Skip routine internal meetings and solo notes.
allowed-tools: Read(*), Write(*), Edit(*), Glob(*)
---

# /granola

File recent Granola meetings into the vault as call notes.

## What to do

1. Use the Granola connector to list meetings since the last run (default: past 7 days).
2. For each meeting, decide whether to file or skip:
   - SKIP: solo notes, recurring 1:1s with my manager, routine internal ops syncs, social events, held meetings with no transcript
   - FILE: external calls, partner meetings, customer calls, new contacts, strategic discussions
3. For each meeting to file:
   - Pull the transcript and summary via the Granola connector
   - Check 20_Areas/People/ for the participants; create or update profiles
   - Check 20_Areas/Companies/ for their org; update Engagement Timeline
   - Create a call note at 20_Areas/Calls/YYYY-MM-DD_Meeting-Name.md
4. Report a summary: filed N, skipped M, and why each was skipped.

Confirm when the skill is installed.

The Voice System

Without a voice system, Claude writes in Generic Professional. With one, Claude writes in your voice — with your cadence, your word choices, your structure preferences, even your banned phrases. This is the difference between Claude as a draft generator and Claude as a ghostwriter who actually sounds like you.

Where Voice Lives

My voice system sits in 99_System/Context_Library/ as a small set of files:

99_System/Context_Library/
│
├── speaking_positions.md     # My actual positions on the topics I'm known for
├── quotable_moments.md       # Lines that landed on stage or in writing
├── email_voice_guide.md      # How I actually write emails (tone, structure)
├── audience_delight_profile.md # What I'm like as a speaker
└── slide_playbook.md         # Visual and verbal patterns for decks

The Files That Carry the Most Weight

Build Your Voice Files

The most accurate source of your voice is your own sent mail. The prompt below tells Claude to use an email connector (Outlook, Gmail, or whatever you have wired up) to read your actual sent emails and produce the voice files from real evidence.

I want you to build my voice system in 99_System/Context_Library/.

First: confirm you have access to an email connector (Outlook, Gmail, Microsoft 365, or similar). If no email connector is enabled in this Claude environment, tell me which one I should set up and walk me through enabling it. Do not proceed until email access is working.

Once you have email access:

1. Read 30 to 50 of my actual sent emails from the past 6 months. Prioritize emails I wrote from scratch over forwards or one-line replies. Include a mix of audiences: internal team, peers, external partners, vendors, and any leadership.

2. Analyze the patterns. Look for:
   - Voice and tone characteristics
   - Sentence and paragraph rhythm
   - Vocabulary I favor and vocabulary I avoid
   - Punctuation habits (em-dashes, parentheses, exclamation marks)
   - Salutations and closings I actually use
   - How my tone shifts across audiences

3. Create these four files in 99_System/Context_Library/:

   - email_voice_guide.md: how I write emails. Tone, structure, salutations, closings, banned phrases. Include 2-3 short before/after examples showing how generic professional writing would get rewritten in my voice.

   - speaking_positions.md: start with section headers for the topics that show up most in my outgoing mail (Strategy, Industry POV, Team Philosophy, etc.). Leave each section mostly empty with a one-line note that it will be filled in over time from calls and webinars.

   - quotable_moments.md: start blank with a brief explanation of what belongs here (lines that landed, on stage or in writing).

   - audience_delight_profile.md: start with what you can infer from my emails about how I show up with different audiences. Flag that this file will get richer once speaking transcripts are added.

4. Report back: how many emails you analyzed, what patterns stood out, and the four files you created.

Companies as Wiki Entries

This is an architectural decision the first guide didn't cover. In my vault, every external organization that touches my work gets a Company file in 20_Areas/Companies/. That file is the canonical wiki entry for the org — the place that consolidates partnership context, products, history, and strategic fit. People files are individual profiles that link into the Company file.

Why This Pattern

People come and go. Companies persist. When someone changes employers, I update the company: field on their People file — their relationship history stays intact, but the wiki entry stays attached to the org. When someone new joins an existing partner, I just add a row to that Company's Key People table and they inherit all the context.

What a Company File Looks Like

---
type: company
name: [Org Name]
website: [domain]
relationship: [strategic-partner | partner | vendor | press | other]
first_contact: YYYY-MM-DD
last_contact: YYYY-MM-DD
---

# [Org Name]

## What They Do
[One paragraph. Plain description.]

## Relationship Snapshot
[Why this org matters to my work right now.]

## Key People
| Name | Role | Contact | Note |
|------|------|---------|------|
| [[Person-Name]] | [Role] | [email] | [one line] |

## Internal Counterparts
[Who on my side owns this relationship.]

## Products / Workstreams
[Concrete things in play.]

## Engagement Timeline
| Date | Touchpoint | Call Note |
|------|------------|-----------|
| 2026-05-20 | Intro call | [[2026-05-20_Org-Name]] |

## Strategic Fit
[How this org sits relative to competitors and our strategy.]

## Open Items
- [ ] [item] — @owner — due YYYY-MM-DD

## Related
[Wikilinks to projects, parallel companies, reference memories.]

The Compounding Rule

Three cross-references happen automatically at creation time, enforced by my CLAUDE.md:

• New person at an existing company → update that Company's Key People table in the same edit.
• New company that already has People profiles → backfill the company: field on every existing person who works there.
• New Call note involving an external org → link to the Company in the call's Related section, and add the row to the Company's Engagement Timeline.

Install the Templates

The prompt below installs the three templates (People, Companies, Calls) and adds the filing conventions to your root CLAUDE.md so Claude enforces the cross-reference rules automatically.

I want to install the relationship-tracking pattern in my vault. Please:

1. Create 20_Areas/People/_template_person.md with this frontmatter:

---
type: person
name:
company: "[[Company-Name]]"
role:
relationship: # client | colleague | partner | vendor
first_contact:
last_contact:
follow_up: false
tags: []
---

## Context
<!-- Role, how we met, what they care about -->

## Relationship History
<!-- Key interactions, auto-appended from calls -->

## Personal Details
<!-- Family, interests, preferences -->

## Open Items
<!-- Commitments made, questions to follow up on -->


2. Create 20_Areas/Companies/_template_company.md with this frontmatter and sections:

---
type: company
name:
website:
industry:
relationship: # strategic-partner | partner | vendor | press | other
first_contact:
last_contact:
tags: []
---

## What They Do
[One paragraph.]

## Relationship Snapshot
[Why this org matters to my work right now.]

## Key People
| Name | Role | Contact | Note |
|------|------|---------|------|

## Internal Counterparts
[Who on my side owns this relationship.]

## Products / Workstreams
## Engagement Timeline
| Date | Touchpoint | Call Note |
|------|------------|-----------|

## Strategic Fit
## Open Items
## Related


3. Create 20_Areas/Calls/_template_call.md with:

---
type: call
contact: "[[Person-Name]]"
company: "[[Company-Name]]"
date:
tags: []
---

## Context
## Discussion
## Decisions
## Action Items
## Follow-up


4. Add a "Filing Conventions" section to my root CLAUDE.md that explains:
   - Companies are wiki entries; the canonical hub for an org.
   - People link to their company via the company: frontmatter field.
   - Calls link to both.
   - When a new person is added at an existing company, also update that Company's Key People table.
   - When a new company is added and People profiles already exist for that org, backfill the company: field on those People files.
   - When a new call note is created, also update the Company's Engagement Timeline and append a line to the person's Relationship History.

Create all three templates and the CLAUDE.md addition now.

Shipping From the Vault

The vault isn't just a filing system. It's also a workshop where I produce branded artifacts and deploy them to the web.

What I Ship From Here

Why Shipping Belongs in the Vault

Two reasons:

• Source material is already here. The transcript I want to turn into a webinar page is in 20_Areas/Keynotes/. The brand logo files are in 30_Resources/. The voice that should narrate the page is in 99_System/Context_Library/. Building elsewhere means dragging all that context to a new tool.
• Skills can compose. A skill like /add-webinar uses the voice system to draft the page copy, the brand resources for visuals, and the deploy infrastructure to ship. Each piece was built for a different reason; the skill stitches them together.

Build Your Own Shipping Skill

Shipping skills are the most personal part of a Personal OS — they depend on what you publish, your brand, and your deploy infrastructure. The prompt below works for any recurring publishing workflow you've done manually two or three times.

I have a recurring publishing workflow I want to turn into a skill. Walk me through the build.

Ask me these questions one at a time:

1. What's the artifact I'm shipping? (one-pager, slide deck, webinar page, social post, branded PDF, etc.)
2. Where does the source material live? (vault folder, Drive, transcript service, etc.)
3. What does the artifact need from my voice or brand system? (writing voice, logos, colors, templates)
4. Where does it get published? (Vercel, GitHub Pages, S3, LinkedIn, Drive, intranet, etc.)
5. What manual steps do I currently do, in order?

Once I've answered all five:

1. Propose a skill name and short description.
2. Draft a SKILL.md that orchestrates the full workflow: pulling sources, applying voice and brand, producing the artifact, deploying or filing it, confirming success.
3. Flag any external setup I need (API keys, account access, MCP connectors).
4. Save the skill to ~/.claude/skills/<name>/SKILL.md and tell me how to trigger it.

MCP Servers: What's Actually Worth Connecting

MCP (Model Context Protocol) servers let Claude reach beyond your filesystem — into project management tools, calendars, CRMs, BigQuery, your inbox. After months of use, only a few are load-bearing.

What I Actually Use

Two Environments, Two Configs

If you use both Claude Code (in VS Code) and Claude Desktop, MCP servers are configured separately:

Writing Rules & Guardrails

Beyond the voice system (which teaches Claude how you write), guardrails are rules that prevent Claude from producing content you'd have to heavily edit or that could embarrass you. These live in your CLAUDE.md or in feedback memories.

Communication Guardrails

If you work in an industry where AI hype language is common, write a rule that pulls drafts back to specifics:

## Communication Guardrails

When drafting communications, presentations, or talking points
for internal audiences, keep AI framing practical and grounded.
Focus on specific tools and wins. Flag if a draft leans too heavy
on sweeping transformation language.

Internal vs. External Content

Claude can easily mix internal language (project codenames, partner pricing details, internal tool nicknames) into content meant for external audiences. The fix is a one-line rule:

## Content Creation

When creating content for external audiences (partners, clients,
press, industry contacts), always ask "Is this internal or external?"
if it's unclear. Never include partner pricing, financial figures,
internal program names, or compliance-sensitive language in external
deliverables.

Data Honesty

If Claude is helping you with reports, analytics, or research, this rule prevents quiet fabrication:

## Data & Reports

Never fabricate or extrapolate data. If I provide three months
of data, do NOT create year-over-year comparisons. Verify
relationships, geography, numbers, and frameworks before writing
them into scripts or briefings. When source data is insufficient,
flag it instead of filling the gap.

Hard-Won Lessons

These come from months of daily use. Each one represents a problem that took time to diagnose.

The Weekly Rhythm

A mature system benefits from a weekly maintenance pass. About 15 minutes. Most of it runs through skills.

The Vault After Six Months

Here's what mine looks like now. Items in green are additions beyond the foundation setup. They all emerged from real workflows.

Personal_OS/
│
├── 00_Inbox/                          # Daily agendas, transcripts, captures
│
├── 10_Projects/                       # Active work, each with its own CLAUDE.md
│
├── 20_Areas/
│   ├── Calls/                         # Meeting notes (per call)
│   ├── Companies/                     # Wiki entry per external org (NEW)
│   ├── People/                        # Individual profiles, linked to Companies
│   ├── Keynotes/                      # Speaking engagement archive (NEW)
│   ├── Events/                        # Conference session notes (audience-side) (NEW)
│   ├── Market_Research/               # Ongoing research area (NEW)
│   └── Press_Coverage/                # Media mentions log (NEW)
│
├── 30_Resources/
│   ├── Research/AI/                   # Sources, Digests, Stats_Library (NEW)
│   ├── Research/Tourism/              # Same structure, different domain (NEW)
│   ├── Webinar_Transcripts/           # With TOPIC_INDEX for voice retrieval (NEW)
│   ├── Brand_Resources/               # Logos, colors, fonts, usage rules (NEW)
│   └── AI_Generated_Images/           # Reusable image library (NEW)
│
├── 40_Archives/
│
├── 50_Personal/                       # Personal projects, kept separate
│
├── 99_System/
│   ├── Context_Library/
│   │   ├── speaking_positions.md      # Voice system (NEW)
│   │   ├── quotable_moments.md        # Voice system (NEW)
│   │   ├── email_voice_guide.md       # Voice system (NEW)
│   │   ├── slide_playbook.md          # Deck design system (NEW)
│   │   └── recurring_meetings/        # Standing roster files (NEW)
│   ├── Profiles/
│   ├── Scripts/                       # Automation scripts and cron jobs (NEW)
│   └── claude_code_config.md
│
├── Apps/                              # Software projects deployed from here (NEW)
│
└── CLAUDE.md

The structure isn't the system — the structure is the residue of the system. Folders appear because workflows need a home. Skills appear because patterns repeat. Memories appear because lessons stick. None of it was planned upfront. All of it was built one piece at a time.