Editor Starter

A forkable kit for a personal AI editor that already knows your voice. Same spine, infinite variations — the taste comes from your own ~/.claude/CLAUDE.md.

Paste any piece of writing in. Get three views back: a structured critique, a tightened version, and three rewrites in your own voice. Sign-in, history, multi-device — the boring infrastructure that turns a demo into a tool you bookmark.

This is the day-to-day evolution of Editor, the product from Part 2 of Rethink enablement. The workshop variant shipped in forty-five minutes and treated the URL as the access token. This starter takes that spine and adds the two features that turn a demo into a tool you open three times a day: authentication and memory.

Two people forking this from scratch should ship two genuinely different products — different voice, different mood, different critique register — without writing different code. The spine is the engineering. The variation is the taste. The taste comes from your global CLAUDE.md, which you already have.

---

Quick Start

# 1. Clone the kit into a new project
git clone https://github.com/v60samurai/editor-starter.git my-editor
cd my-editor && rm -rf .git && git init

# 2. Read the worked example (10 min)
cat docs/PRD.md docs/BRAND.md docs/PERSONALIZATION.md

# 3. Open Claude Code and paste the build prompt (see below).
#    Claude Code generates the Next.js app from these docs;
#    no app code is checked in by design.

# 4. Ship to Vercel (after the build sessions complete)
vercel

Total time from clone to deployed Editor: two to three hours end to end. Forty-five minutes are for the three core agentic features. The rest is auth, memory, history UI, and personalization audit.

No global ~/.claude/CLAUDE.md yet? Read docs/PERSONALIZATION.md Section 8 before you start — the workflow shifts from extraction to interview.

---

What's in here

editor-starter/
├── README.md                          ← you are here
├── LICENSE
├── .gitignore
└── docs/                              ← the worked example: Editor
    ├── PRD.md                         ← v2: includes Auth + Memory sections
    ├── BRAND.md                       ← the editorial-workshop spine
    └── PERSONALIZATION.md             ← the spine-vs-variation contract; agent trigger for the build

This kit is documentation-first by design. No app code is checked in. The four documents (docs/PRD.md, docs/BRAND.md, docs/PERSONALIZATION.md, plus a filled-in Implementation Guide from builder-os) are everything Claude Code needs to generate the Next.js app session by session.

Blank templates — the PRD template, the brand-guide template, and the three-document implementation-templates set (IMPLEMENTATION_GUIDE.md, SESSION_PLAYBOOK.md, FINAL_PUSH.md) — ship in builder-os, the companion repo.

docs/ is one product (Editor) filled in. docs/PERSONALIZATION.md is what makes your fork look nothing like anyone else's — read it before you start.

---

How to fork this into your own product

1. Copy this directory into a fresh repo. cp -R editor-starter ../my-editor && cd ../my-editor && git init.
2. Read the worked example. Open docs/PRD.md, then docs/BRAND.md. This shows you what good looks like.
3. Run the Personalization Pre-Flight. Open docs/PERSONALIZATION.md Section 5. Ten minutes. Skip this and your fork will look like everyone else's.
4. Spin your own PRD. Overwrite docs/PRD.md with your own — use the blank PRD template from builder-os, or mirror the structure of the Editor PRD. Decision Rationale rule applies: every non-obvious choice gets a one-line defense.
5. Spin your own brand guide. Overwrite docs/BRAND.md with your own — use the blank brand-guide template from builder-os. Fill in five sections minimum for v1.
6. Generate the implementation pack — point Claude Code at docs/PERSONALIZATION.md first. The Agent Instructions in Section 6 tell the agent how to pull your taste from ~/.claude/CLAUDE.md into the Implementation Guide. Then run Shipwright on your PRD, or fill the three implementation-templates files (IMPLEMENTATION_GUIDE.md, SESSION_PLAYBOOK.md, FINAL_PUSH.md) from builder-os by hand.
7. Hand the four documents to Claude Code. Build the product session by session.
8. Run the Personalization Audit before shipping. Five questions in docs/PERSONALIZATION.md Section 7. Pass all five or iterate.

If your fork is also a personal-tool-with-auth-and-memory product, the schema in step 6 should include the users, sessions, and entries tables described in Section 8 of docs/PRD.md.

The one-line build prompt

When you are ready to generate the Implementation Guide, paste this into Claude Code:

Read docs/PERSONALIZATION.md end to end, then follow the Agent Instructions in Section 6. Pull my taste from ~/.claude/CLAUDE.md, propose first passes for the six personalization slots with one line of reasoning per slot, then generate the Implementation Guide and a draft of src/persona/CLAUDE.md. Run the Personalization Audit at the end and report any failures.

---

The features Editor ships with

Feature	Agentic pattern	What it does
Roast	Tool use	Critique any paste, located by line, in your voice
Tighten	Sequential agent loop	Cut a paragraph in half without losing meaning, with a third-call meaning-validator
Voice Match	Parallel few-shot	Rewrite the same paste in three of your own voices (terse, lyrical, formal — whatever three you pick)
Auth	Infrastructure	Email magic link via Lucia + Arctic + Oslo. Sessions last thirty days. Default: single-account-per-deploy.
Memory	Infrastructure	Every paste + output written to entries. /history lists them newest-first, filterable. Re-run any past entry against the current studio.

The three core features are unchanged from the workshop. Auth + memory are the two additions that earn the "day-to-day" label.

---

What this is not

• Not a SaaS template. No billing, no teams, no admin UI. Add those in your fork if you need them.
• Not a hosted product. You deploy it yourself (Vercel one-click works). The kit is for builders who want to own their tools.
• Not opinionated about the editor's personality. The persona file plus voice corpus do all the personality work. The kit is the wiring; the soul is what you put in src/persona/CLAUDE.md and src/corpus/voice-samples.md.

---

Stack

• App: Next.js 16 App Router, React 19, TypeScript strict, Tailwind 4
• AI: Vercel AI Gateway → Anthropic Claude, AI SDK v6, OIDC auth (no provider keys)
• Auth: Lucia + Arctic + Oslo (passwordless email magic link)
• DB: PostgreSQL via Drizzle (queries) + Prisma (migrations)
• Email: Resend (swap to any SMTP via src/lib/email.ts)
• Deploy: Vercel

The full architecture and the schema scaffolds live in the Implementation Guide template (in builder-os) — fill it in for your fork.

---

Roadmap (post-v2)

• v2.1: Postgres trigram search on entries.input so you can find a critique by text content.
• v2.2: Studio Swap dropdown UI.
• v2.3: Mobile-perfect responsive design.
• v3: Streaming token-by-token UI for the Tighten flow's three phases.

These are noted in docs/PRD.md as out-of-scope-for-now. Fork and ship them if you want.

---

Contributing

This kit is meant to be forked, not maintained as a shared codebase. The expected workflow is:

1. Fork the repo.
2. Build your own Editor on top of it.
3. If you discover an improvement to the spine (auth, memory, agentic patterns, deploy story), open a PR back here.
4. If you discover an improvement to a slot (your persona, your tools, your studio mood), keep it in your fork. That is by design — variations stay personal.

PRs that touch docs/PERSONALIZATION.md are especially welcome. Templates-pack PRs go to builder-os.

---

License

MIT — see LICENSE. Use it, fork it, sell what you build with it. Attribution appreciated but not required.

---

Credits

Built by Harshit Badiger. Spun out of Part 2 of Rethink enablement. The three-document Shipwright workflow (Implementation Guide, Session Playbook, Final Push) that ships in builder-os was first published at https://github.com/v60samurai/shipwright.

---

Editor Starter v2.0 · 2026-05-17