Context Kit — Setup Guide

What This Is

A four-file system for giving AI persistent context about who you are, how you sound, and how you work. Based on Nav Toor's three-file framework with a manifest layer for multi-brand operators.

File Structure

📁 context-kit/
├── _MANIFEST.md       ← Load order + conflict rules       (update quarterly)
├── about-me.md        ← Identity + current priorities      (update weekly)
├── brand-voice.md     ← Rules, banned words, examples      (update monthly)
└── working-style.md   ← Collaboration defaults             (update monthly)

Setup (30 Minutes)

Option A: Fill templates directly. Open each file, replace comments with your info. about-me.md (10 min) → brand-voice.md (15 min) → working-style.md (5 min).

Option B: AI-guided builder. Paste the Builder Instructions (end of this doc) into a fresh session. The AI interviews you and assembles the files.

Where to Put the Files

Platform	Location
Claude Projects	Upload to Project Knowledge
Claude Cowork	Dedicated folder, set as default context directory
ChatGPT Custom GPTs	Instructions field or Knowledge uploads
Any other tool	Paste contents into system prompt at session start

Verification Prompt

After the AI loads your files, run this:

Read all the files in this folder completely. Then summarize: who I am, how I sound, and how we work together.

If the summary is wrong, the files are wrong. Fix the files, not the prompt.

Key Principles

Rules beat adjectives. "Professional and friendly" teaches nothing. "Sentences under 20 words, contractions always, no passive voice" teaches everything.

Anti-examples are load-bearing. The off-brand example in brand-voice.md eliminates more bad output than any positive instruction.

Banned words are the highest-leverage section. 15-20 entries cuts editing time 40-60%.

Brevity is a feature. If about-me.md exceeds two pages, cut it.

Correct in the file, not the prompt. Prompt corrections die at session end. File corrections persist forever.

Load only what's relevant. For multi-brand operators: Tier 2 files load only when that brand is active.

Common Mistakes

Mistake	Fix
Resume instead of identity	`about-me.md` should sound like you talking
Adjectives instead of rules	Convert every tone descriptor into a mechanical instruction
Skipping the off-brand example	Write it. It's the highest-value anti-pattern.
One giant file	Four files = separation of concerns
Set and forget	Update weekly or output goes stale
Correcting in prompts	Move recurring corrections into files
Loading all brand contexts	Only load Tier 2 files relevant to current task

AI-Guided Builder Instructions

Paste everything below into a fresh AI session to build your context files through conversation.

You are helping me build a four-file context management system.
Your job: ask targeted questions, capture my answers in my own
words, and assemble the files. Do not polish my language — the
raw version IS the brand voice.

## Rules
- Ask questions one at a time. Wait for my answer before moving on.
- Don't rephrase my answers into corporate-speak.
- If I give a short answer, don't pad it.
- After each phase, show me the assembled file for review.

## Phase 1: Identity → about-me.md

1. "What do you do? Explain it like you would to someone
   sitting next to you at a dinner party."

2. "What businesses or projects are you running right now?
   For each: what is it, who does it serve, what makes it different?"

3. "What's your background? Not your resume — the path that
   led you here and what makes your current work credible."

4. "What are you actually working on right now? 3-5 things
   on your plate this week or month."

5. "Do you have specific frameworks, methods, or mental models
   you use regularly? Things with names?"

6. "What's one thing you've built or shipped that represents
   your quality bar?"

→ Assemble about-me.md. Show it. Get approval before moving on.

## Phase 2: Voice → brand-voice.md

1. "Send me 2-3 things you've written that sound the most like
   you. Emails, posts, messages. Unpolished > published."

   [Analyze silently for: sentence length, vocabulary, punctuation,
   paragraph structure, tone, recurring patterns. Then present.]

2. "Here's what I see in your writing style: [analysis].
   Sound right? What would you change?"

3. "What words or phrases make you cringe? Especially AI-generated ones."

4. "Do you write differently for different contexts? Social vs.
   email vs. sales vs. internal — walk me through it."

5. "Write one paragraph about anything that sounds exactly like you."

6. "Show me or describe writing that represents what you NEVER
   want to sound like."

→ Assemble brand-voice.md. Convert preferences into mechanical
  rules. Show it. Get approval.

## Phase 3: Collaboration → working-style.md

1. "When you get AI output, what do you edit most often?"

2. "Do you want the AI to just go, or ask questions first?"

3. "What format do you usually want? Plain text, markdown,
   formatted docs? Headers and bullets, or paragraphs?"

4. "Pet peeves about working with AI — things that bother
   you even when the content is technically fine."

5. "When the AI thinks you're wrong, what should it do?"

6. "Where do you go for information? Trusted sources?
   Sources you avoid?"

→ Assemble working-style.md. Show it. Get approval.

## Phase 4: Assembly

1. Generate _MANIFEST.md. If multiple brands/projects were
   mentioned, add them as Tier 2 entries.
2. Present all four files as a complete set.
3. Verify: summarize who they are, how they sound, how we work.
4. Ask: "Does this sound like you? Anything wrong or missing?"
5. Final edits. Remind them: update about-me.md weekly,
   review all files monthly.