project-init

name	project-init
description	Project scaffolding — creates CLAUDE.md, documentation/, datarim/ for new or existing projects. Loaded by /dr-init when project intent is detected.
current_aal	1
target_aal	2

Project Init — Scaffolding Skill

Loaded by: /dr-init (Step 0, when project intent detected) Purpose: Create a standardized project structure with CLAUDE.md, documentation, and Datarim workflow state.

When This Skill Activates

/dr-init loads this skill when the user's input contains project creation signals:

• Keywords: "create project", "new project", "init project", "scaffold project"
• Update keywords: "update project structure", "re-scaffold", "обнови структуру проекта"
• Russian: "создай проект", "новый проект", "инициализируй проект"
• Pattern: /dr-init create project "Name"

If none of these signals are present, /dr-init follows the standard task flow.

Scaffolding Flow

Step 1: Gather Project Info

Ask the user (if not already provided in the prompt):

1. Project name — used in CLAUDE.md header and directory name
2. One-line description — what the project does
3. Project type — determines tech stack (see $HOME/.claude/skills/tech-stack/SKILL.md § Stack Selection Decision Tree)

If the user provided enough context in the initial prompt, extract these values without asking. Only ask for what is missing.

Step 2: Determine Target Directory

• Default: current working directory
• If user specified a name: create <name>/ in current directory (kebab-case)
• If inside an Obsidian vault or monorepo: respect the existing structure (e.g., Projects/<Name>/code/)

Verify the target exists or create it.

Step 3: Determine Tech Stack

Load $HOME/.claude/skills/tech-stack/SKILL.md and match the project type to the required stack. This determines:

• .gitignore contents
• Build commands for CLAUDE.md
• Dependencies and toolchain

If the project type is unclear, ask the user. If the project is documentation/research-only, skip tech stack detection.

Step 4: Create Project Structure

Create the following structure following the Diátaxis Documentation Taxonomy Mandate (skills/diataxis-docs/SKILL.md). Idempotency rule: check if each file/directory exists before creating. If it exists, skip it and report "skipped: already exists". Never overwrite existing files.

<project-root>/
├── CLAUDE.md                    # From template: $HOME/.claude/templates/project-claude-md.md
├── .gitignore                   # Standard for detected stack
│
├── documentation/                        # Diátaxis 4-category split (mandate per skills/diataxis-docs/SKILL.md)
│   ├── tutorials/               # Learning-oriented (newcomer end-to-end)
│   │   └── README.md            # From template: $HOME/.claude/templates/documentation-diataxis/tutorials/README.md
│   ├── how-to/                  # Problem-solving (task recipes)
│   │   ├── README.md            # From template: $HOME/.claude/templates/documentation-diataxis/how-to/README.md
│   │   ├── testing.md           # Legacy stub mapped to how-to per Diátaxis
│   │   ├── deployment.md        # Legacy stub mapped to how-to
│   │   └── gotchas.md           # Legacy stub mapped to how-to
│   ├── reference/               # Information-oriented (lookup, catalogue)
│   │   ├── README.md            # From template: $HOME/.claude/templates/documentation-diataxis/reference/README.md
│   │   └── architecture.md      # Legacy stub mapped to reference (system map)
│   └── explanation/             # Understanding-oriented (background, why)
│       └── README.md            # From template: $HOME/.claude/templates/documentation-diataxis/explanation/README.md
│
├── documentation/ephemeral/              # Transient working material (may be gitignored or committed per preference)
│   ├── plans/                   # Implementation plans
│   ├── research/                # Research notes
│   └── reviews/                 # QA reports and reviews
│
├── datarim/                     # Workflow state (created via standard /dr-init logic)
│   ├── backlog.md               # From template: $HOME/.claude/templates/backlog-template.md
│   ├── activeContext.md          # Active task tracking
│   └── tasks.md                 # Task details
│
└── documentation/               # Long-term archives (committed to git)
    └── archive/                 # Completed task archives

Step 5: Fill CLAUDE.md Template

Read ${DATARIM_RUNTIME:-$HOME/.claude}/templates/project-claude-md.md and replace placeholders:

Placeholder	Source
__PROJECT_NAME__	From user input (Step 1)
__ONE_LINE_DESCRIPTION__	From user input (Step 1)
__DATE__	Current date (YYYY-MM-DD)
__TECH_STACK__	From tech-stack.md detection (Step 3)
__BUILD_COMMANDS__	From tech-stack.md detection (Step 3)
__GITIGNORE_PATTERNS__	From tech-stack.md detection (Step 3)

Diátaxis taxonomy in CLAUDE.md. When the project's CLAUDE.md is generated, include a one-liner reference to documentation/{tutorials,how-to,reference,explanation}/ so that future contributors discover the mandate from the project root, not only from Datarim framework docs.

For placeholders the agent cannot fill (components, terminology, gotchas), leave them as [TODO: ...] markers for the user.

Step 6: Initialize Git (if needed)

• If .git/ does not exist in the target directory: run git init
• If .gitignore was created: ensure datarim/ is listed
• Do NOT create an initial commit automatically — let the user review first

Step 7: Post-Scaffold Report

Output a summary:

Project scaffolded: <project-name>
Location: <target-path>
Stack: <detected-stack or "none (documentation project)">

Created (Diátaxis 4-category split per skills/diataxis-docs/SKILL.md):
  ✓ CLAUDE.md
  ✓ documentation/tutorials/README.md         (learning-oriented)
  ✓ documentation/how-to/README.md            (problem-solving)
  ✓ documentation/how-to/testing.md           (legacy stub mapped to how-to)
  ✓ documentation/how-to/deployment.md        (legacy stub mapped to how-to)
  ✓ documentation/how-to/gotchas.md           (legacy stub mapped to how-to)
  ✓ documentation/reference/README.md         (information-oriented)
  ✓ documentation/reference/architecture.md   (legacy stub mapped to reference — system map)
  ✓ documentation/explanation/README.md       (understanding-oriented)
  ✓ documentation/ephemeral/{plans,research,reviews}/
  ✓ datarim/ (workflow state)
  ✓ documentation/archive/
  ✓ .gitignore

Skipped (already existed):
  - <list of skipped files, if any>

Next steps:
  1. Review and customize CLAUDE.md — fill in [TODO] placeholders
  2. Review .gitignore
  3. git add -A && git commit -m "scaffold: initial project structure"
  4. /dr-init <first task description>  — start your first task

Idempotency Rules

1. Never overwrite an existing file — even if the template has changed
2. Always create missing directories silently
3. Report every skipped file so the user knows what was preserved
4. Update mode: if user says "update project structure" or "re-scaffold", apply the same logic — create only what is missing, skip what exists

Integration with Existing Projects

When run in an existing project (that already has some files):

1. Scan for existing CLAUDE.md — if found, skip it
2. Scan for existing documentation/ — create only missing stubs
3. Scan for existing datarim/ — skip entirely (already initialized)
4. Create only what is missing from the standard structure

This allows updating old projects to the new structure incrementally.

What This Skill Does NOT Do

• Does not install dependencies via the project's package manager — that is the user's responsibility
• Does not create source code files — only project infrastructure
• Does not modify existing files — only creates new ones
• Does not run CI/CD setup — that is a separate task
• Does not add --dry-run or --force flags (future enhancement)