Init

Project Initialization Skill

Sets up a .memory/ structure for persistent multi-session tracking, and a .claude/CLAUDE.md entry point that loads memory automatically every session.

Workflow

Step 1: Choose mode

Use AskUserQuestion to ask whether this is a project or workspace:

• Project — a focused effort with a clear scope, spec, and phased plan. Examples: building a feature, a new app, a library.
• Workspace — an ongoing environment where tasks are largely independent. Examples: dotfiles, personal scripts, a monorepo with unrelated work streams.

Step 2: Interview the user

Ask conversationally — one or two questions at a time, not all at once. Adapt based on what the user has already shared in the conversation.

1. What are you building / what is this workspace for?
2. What problem does this solve / what's the goal?
3. What's explicitly out of scope? (things you are NOT doing)
4. What tech stack, languages, or frameworks?
5. Any key constraints or non-negotiables?

Do NOT use AskUserQuestion for these — they need free-form text answers.

If the user has already described the project, skip what you already know and confirm.

Step 3: Create the structure

Both modes

Create or update .claude/settings.json to include:

{
  "autoMemoryEnabled": false
}

If the file already exists, merge — preserve any existing keys and add/update only this.

Create .memory/ containing:

• MEMORY.md — index file (template below)
• status.md — current state snapshot (template below)

Create or update .claude/CLAUDE.md with the @.memory/MEMORY.md import.

Project mode additionally

Create:

• .memory/spec.md — project goal (template below)
• .memory/plan.md — phases and open questions (template below)
• .memory/features/ — directory for per-feature tracking (add .gitkeep)

Add spec.md to the MEMORY.md index.

Workspace mode additionally

Create:

• .memory/tasks/ — directory for per-task files (add .gitkeep)

No root spec.md or plan.md — those belong inside individual task files.

Step 4: Populate the files

Use the templates below. Replace {{placeholders}} with interview content.

.memory/MEMORY.md

# Project Memory

- [Project Status](status.md) — {{one-liner current state}}

Project mode also adds:

- [Specification](spec.md) — {{one-liner project goal}}

Keep this index concise — one line per file, ~150 chars max per entry. Add entries as new topic files are created. diary.md is NEVER listed here.

.memory/status.md

---
name: Project Status
description: {{one-liner current state}}
type: status
---

**Phase**: Not started
**Focus**: Project setup complete. Ready to begin.
**Last updated**: {{today's date}}

## In Progress

_Nothing yet._

## Blocked

_Nothing._

## Next Up

- [ ] Review and refine spec
- [ ] Break spec into phases in plan.md
- [ ] Identify first task to work on

## Recent Completions

- [x] Project initialized

.memory/spec.md (project mode)

---
name: Specification
description: {{one-liner project goal}}
type: spec
---

# {{Project Name}} — Specification

## Overview

{{1-2 sentence description}}

## Goals

{{Primary goal and sub-goals, as bullet points}}

## Non-Goals / Out of Scope

{{What we are NOT building, as bullet points}}

## Tech Stack

{{Languages, frameworks, platforms, key dependencies}}

## Constraints

{{Non-negotiables, compatibility requirements, performance budgets}}

.memory/plan.md (project mode)

---
name: Plan
description: Phases and open questions
type: plan
---

# {{Project Name}} — Plan

## Phases

_Break the spec into sequential phases here. Each phase should be achievable
in a few sessions._

### Phase 1: {{Name}}

- [ ] {{Task}}

## Open Questions

_Questions that need answers before or during implementation._

.claude/CLAUDE.md

If .claude/CLAUDE.md does not exist, create:

# {{Project Name}}

{{1-2 sentence description}}

## Project Memory

@.memory/MEMORY.md

_Topic files are loaded on demand — read them when relevant, not every session._

## Memory Routing

When the user says "remember X", "note that", "record this", or anything similar,
always handle it by invoking the `/cpr:memory` skill. Never use the built-in
auto-memory system — it is disabled for this project.

If .claude/CLAUDE.md already exists, add the @.memory/MEMORY.md import under an appropriate heading.

Step 5: Confirm and suggest next steps

Mention that .claude/settings.json was written with auto-memory disabled.

Tell the user what was created. Suggest 2-3 concrete next steps:

• Project mode: review spec.md, define phases in plan.md, identify first feature
• Workspace mode: create the first task file in tasks/

Also mention how to use the project going forward:

To work on this project, use these skills:

• /cpr:start — Begin a session and see current state
• /cpr:end — End a session and save progress
• /cpr:memory — Record persistent information
• /cpr:research — Investigate topics or explore code

Keep it to 3-4 lines — don't overwhelm.