Skip to content

870tik/brain-codifier

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Brain Codifier

Codify your own brain with Claude Code — a personal Q&A knowledge system.

                  ┌──────────────┐
   ┌─────────────►│   8 domains  ├──────────────┐
   │              └──────────────┘              │
   │                                            ▼
[questions]                              [brain/domains/]
                                                │
                                                ▼
                                        [/synthesize]
                                                │
                                                ▼
                                       [your principles]

A book and a system, generated from itself.


What this is

A method and tooling for handing your own particular logic to AI.

Not "let AI imitate you." Rather: codify the structure of how you decide (what you won't compromise, how you weigh trade-offs, what you find beautiful) so AI can carry it.

The mechanism is simple:

  1. Eight domains, ~240 questions
  2. Save answers verbatim in brain/domains/
  3. After 30+ answers, run /synthesize to extract your principles
  4. Drop the principles into AI's system prompt — outputs start to lean toward you

This repository is the open-source bundle that runs the system on Claude Code. Drop into your project, run /start, and your accumulation begins.


Quickstart

# 1. Clone into your Claude Code project root
git clone https://github.com/870tik/brain-codifier.git
cd brain-codifier

# 2. Or paste files into an existing project
cp -r .claude questions brain CLAUDE.md /path/to/your/project/

# 3. Start Claude Code in that project
claude

# 4. Run the first command
/start

The first question gets thrown. Answer it. Run /start again or /ask. After 30 answers, run /synthesize to get your first principles.


What's in the box

.claude/
├── skills/                      # 8 slash commands
│   ├── start/SKILL.md           # /start — the first question
│   ├── ask/SKILL.md             # /ask [domain] — AI picks a question
│   ├── answer/SKILL.md          # /answer — save (verbatim)
│   ├── recall/SKILL.md          # /recall <query> — search past
│   ├── reflect/SKILL.md         # /reflect <domain> — deep dive
│   ├── connect/SKILL.md         # /connect — find cross-domain patterns
│   ├── synthesize/SKILL.md      # /synthesize — extract principles
│   └── map/SKILL.md             # /map — overall map
├── agents/                      # 3 specialized agents
│   ├── interviewer/AGENT.md     # Sonnet — long sessions
│   ├── synthesizer/AGENT.md     # Opus — pattern extraction
│   └── cartographer/AGENT.md    # Haiku — aggregation
└── rules/
    ├── common/
    │   ├── tone.md              # "a curious friend" voice
    │   └── voice-preservation.md  # don't change user words
    └── brain/
        └── brain-layer.md       # path-scoped rules

questions/                       # ~240 questions across 4 tiers
├── starter.md                   # first 30
├── domains/                     # 20 × 8 = 160
├── deep.md                      # 30 deep questions
└── edge.md                      # 20 edge questions

brain/                           # your accumulation goes here
├── domains/{8 dirs}/
├── principles/ontology.md
└── index.md

CLAUDE.md                        # project-root config

The non-negotiable rule

Never change the user's words.

Every saved answer is the user's. AI may write metadata, derive observations in a separate ## 派生 section, and update the index — but the ## A section is sacred. Don't fix typos. Don't smooth tone. Don't paraphrase.

This is what makes the record meaningful when read 100 years later. Tidied notes show only the tidier.

Details: .claude/rules/common/voice-preservation.md


The four-tier confidence model

Principles aren't declared — they emerge.

Level Meaning Backing
emerging A trace appeared 2–3
forming A shape is emerging 4–6, multiple domains
established Confirmed 6+, three or more domains
core A non-moving self Re-confirmed across 3+ /synthesize runs

Principles can also retire. The retirement is data: what you stopped believing matters as much as what you still believe.


Why Claude Code

  • Skills make operations structured and re-callable (/synthesize over chat)
  • Agents let you assign Opus to extraction, Sonnet to dialogue, Haiku to aggregation
  • Local databrain/ lives on your disk, not in someone else's cloud

The book

This open-source bundle is the runtime of Brain Codifier. The full method, philosophy, one-month case study, and applications (decision oracle, AI self-prompting, automated product pipeline) are in the book:

  • Brain Codifier — 147 pages, $29 individual / $99 team / $299 enterprise
  • Buy on Gumroad →

The book itself was generated from the system it describes. After reading it, you can use the same setup to write your next book.


Status

  • 8 skills
  • 3 agents
  • Voice-preservation rules
  • 240+ question pool
  • One-month case study (in book)
  • /publish skill (automated product pipeline)
  • Optional product-scout agent

Contributing

Issues and pull requests welcome — particularly:

  • Question pool additions in any language
  • Translations of skill / agent files
  • New skill ideas (but don't grow the core 8 — keep it minimal)

The voice-preservation rule is non-negotiable. Any change that risks rewriting user content will be declined.


License

MIT.

The book content (PDF) is not under MIT — that's a separate purchase. This OSS bundle is provided so anyone can run the system without buying the book.


Related

Brain Codifier is Q&A-first, principle-extracting, AI-native — a different cut.


Author

870tik Industrial designer. Exploring AI × design with Claude Code.


"The notes have grown. The thinking has not. This isn't a problem of recording. It's a problem of format." — Brain Codifier, Chapter 1

About

Codify your own brain with Claude Code — a personal Q&A knowledge system

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors