Skip to content

customization.md

Latest commit

 

History

History
215 lines (138 loc) · 6.89 KB

customization.md

File metadata and controls

215 lines (138 loc) · 6.89 KB

Customization

The whole point of forking this repo is to make it yours. Here's the spectrum of customization from light to deep.

Light: tune what's there

These changes don't add anything new — they make the default setup fit your life.

1. Persona

Edit the Persona section of CLAUDE.md. The default is "direct, warm, slightly cheeky." Variations:

  • Softer: "warm, encouraging, patient. Asks rather than tells. Reframes setbacks as data."
  • Sharper: "blunt, unsentimental, allergic to excuses. Names patterns you'd rather not see."
  • More technical: "engineer-coded. Roots arguments in first principles. Skeptical of consensus."

2. Tier system

The default tiers (Family / Meaningful work / Health) are placeholders. Replace with what actually matters to you:

  • Founder: Cofounders / Investors / Side project
  • Parent of young kids: Family / School community / Self
  • Career-builder: Mentors / Direct reports / Sponsors

The tier system controls filtering everywhere — high-tier people surface first in briefings.

3. Scheduled task timing

The defaults (5am, 9pm, etc.) match a typical morning person. If you're a night owl, shift everything 2–3 hours later. The morning brief should arrive 60–90 min before your alarm.

4. Tracked projects

CLAUDE.md has a #1 Tracked Project slot. Use it for the one long-running thing you most fear losing track of. Examples:

  • A fundraise
  • An immigration case
  • A medical situation for a family member
  • A house renovation
  • A book you're writing

Update it when the project changes.

Medium: edit the briefings

Each briefing in briefings/ is a standalone prompt. Edit them.

Common edits

Remove a section you don't care about. e.g., if you don't use Fitbit, delete Step 8 from morning-brief.md.

Add a section for your situation. e.g., a founder might add to morning-brief.md:

## Step 9 — Today's narrative

In one sentence: what's the story we'd tell about today if today went well? Surface this at the top of the email under "🎯 Today's three."

Change the email format. The default is markdown-ish. You can switch to plaintext, formal HTML, or whatever fits your reading habits.

Adjust the output length. If briefings feel too long, add to the prompt:

Hard cap: 600 words. If exceeded, ruthlessly cut. The cap is the cap.

If they feel too thin, raise the cap and unlock more sub-sections.

Medium: add a per-person memory file

For someone in your life with ongoing complexity — a kid, an aging parent, a co-founder, a high-touch report — create memory/<name>.md.

Template:

# <Name>

## Profile
Age, role/relationship, location, current life stage.

## What's current
What's happening with them right now (the things you'd update monthly).

## Recurring themes
Patterns you want to remember (the things that don't change month to month).

## History
Important past events that explain the present.

## Open threads
| Status | What | Notes |

## Notes
Free-form. Anything else.

Tell CLAUDE.md about this file:

## Per-Person Memory Files
- `memory/<name>.md` — Read when this person is the subject of a conversation or surfaced in a briefing.

This is one of the highest-leverage extensions. It lets the exocortex maintain genuine context on the 3–5 people who need it.

Medium: add a new briefing

Want a daily reflection on your business? A weekly review of your fitness data? A custom monthly review?

Add a new file in briefings/. Use the existing ones as templates. Follow the recipe:

  1. Step 0: Acknowledgment scan
  2. Step 0.5: Health check
  3. Step 1: Memory load
  4. Steps 2..N: Do the work
  5. Step N+1: Update memory + send

Then ask Claude:

"Set up a scheduled task at on that runs the prompt in briefings/.md."

Common additions:

  • Daily journaling prompt — emails you 3 questions to answer
  • Weekly investing review — reads your portfolio + relevant news
  • Monthly relationship check-in — deeper than the weekly review
  • Quarterly OKR check — pulls quarterly goals from life.md, asks "honestly: are these still real?"
  • Daily "what moved the needle" reflection — founder-mode

Deep: add a new skill

Skills are on-demand commands. They live in skills/<name>/SKILL.md. The format follows Claude's Skills convention.

Template:

---
name: <skill-name>
description: <one-line description for skill matching>
---

# <Skill Name>

<Prompt the skill executes when invoked.>

Ideas:

  • /standup — generate today's standup update for your team
  • /decide — runs a structured decision framework on whatever question you bring
  • /playdate — finds 2–3 plausible play dates this weekend from your network
  • /onepager — turns a memory file or a conversation into a polished one-pager doc
  • /airbnb-it — quick analysis: should this trip be Airbnb or hotel?
  • /journal — interactive journaling, writes to a private journal file

Deep: add a new data source

Connect a new MCP. Then:

  1. Add to CLAUDE.md > Data Source Priorities.
  2. Add health-check entry to Step 0.5 of relevant briefings.
  3. Decide which briefings read it. Usually morning brief, possibly a domain-specific briefing.
  4. Map fields to memory files. New financial data → finances.md; new wearable data → health.md.

See docs/data-sources.md for full guidance.

Deep: change the architecture

If you want to depart from the three-layer model, the choices are:

Replace markdown memory with a database. Trade-offs: gain queryability, lose human-editability. Only do this if your memory is bigger than ~500KB across files (it shouldn't be).

Replace the acknowledgment loop with something else. Voice replies, a check-off app, automatic inference from data. Each has trade-offs (see docs/acknowledgment-system.md). The mailto: approach is the local optimum but not the only option.

Run briefings on the server instead of the client. Trade-offs: gain consistency (no laptop-asleep problem), lose privacy (memory leaves your machine). Usually not worth it.

What NOT to customize (until you've used it for a month)

These are tempting but premature optimizations:

  • Don't tighten the prompts on day one. Let the exocortex over-share initially. You need to see what it surfaces before you know what to filter.
  • Don't connect 6 data sources right away. Start with 2. Add one per week.
  • Don't write a new briefing until you've used the default 7 for a few weeks. You don't know what's missing until you've felt it.

When you've extended in cool ways

Open a PR. Specifically interesting:

  • New per-role configurations (founder edition, student edition, retiree edition)
  • New connectors (Apple Notes, Things, Roam, Obsidian, etc.)
  • New briefing types
  • New skills

The base repo stays minimal. The community library of extensions is where the depth lives.