Scaffold New Project

Or ask Claude:

"Set up the full combo stack for this new project."

Step-by-Step Execution

When this skill is invoked, Claude MUST execute the following steps in order. Do not skip steps. Confirm completion of each before proceeding.

Step 1 — Detect project root

pwd
ls -la

Confirm you are in the correct project root. If package.json, pyproject.toml, go.mod, or similar exists, note the tech stack for CLAUDE.md generation.

Step 2 — Create CLAUDE.md

Create ./CLAUDE.md with this lean template. Replace <STACK> with detected tech stack. Keep it short — Caveman Compress will tighten it further.

## Stack
- <STACK> (e.g. Python 3.12, FastAPI, PostgreSQL)
- Tests: <TEST_FRAMEWORK>

## Graphify
- Knowledge graph at graphify-out/GRAPH_REPORT.md
- Read GRAPH_REPORT.md before any architecture questions
- Query graph before Glob/Grep (hook already installed)

## Output Rules
- No greetings, sign-offs, or preamble
- No narrating what you're about to do — just do it
- Code blocks: always include language tag
- Errors: show exact message + fix only

## Shorthand (see RULES.md)
- crt = create, upd = update, del = delete
- ref = refactor, tst = test, doc = document

Step 3 — Create RULES.md

Create ./RULES.md with a starter shorthand dictionary. Extend with project-specific abbreviations as the codebase grows.

# RULES.md — Shorthand Dictionary

## General
- crt  = create
- upd  = update
- del  = delete
- ref  = refactor
- tst  = test / write tests
- doc  = document / add docstring
- chk  = check / validate
- fix  = fix bug
- rev  = review
- dep  = dependency / dependencies

## Architecture
- svc  = service
- repo = repository (data layer)
- mdl  = model / data model
- api  = API endpoint/handler
- mw   = middleware
- cfg  = config / configuration
- env  = environment / .env
- db   = database
- q    = query
- res  = response
- req  = request

## Usage example
# "crt svc for user auth w/ jwt" 
# = "create a service for user authentication with JWT"

Step 4 — Create .claudeignore

Create ./.claudeignore to hide irrelevant files from Claude's context:

# Dependencies — never read these
node_modules/
vendor/
.venv/
__pycache__/
*.pyc

# Build outputs
dist/
build/
out/
.next/
.nuxt/

# Generated / compiled
*.lock
package-lock.json
yarn.lock
poetry.lock
Pipfile.lock
*.min.js
*.min.css

# Stack-generated files (Graphify)
graphify-out/

# Test fixtures / snapshots (large, noisy)
__snapshots__/
fixtures/
testdata/

# Logs
*.log
logs/

# IDE / OS
.DS_Store
.idea/
.vscode/
*.swp

Step 5 — Create / update .gitignore

Append the following to ./.gitignore (create the file if it doesn't exist). Do NOT remove existing entries:

# ── Graphify (regenerated per machine) ────────────────────
graphify-out/

# ── Caveman backup (personal, not for sharing) ────────────
CLAUDE.original.md

# ── Claude Code local settings (machine-specific) ─────────
.claude/settings.local.json

# ── Commit these (teammates need them) ────────────────────
# CLAUDE.md        ← keep
# RULES.md         ← keep
# .claudeignore    ← keep

Step 6 — Install Graphify integration

graphify claude install

This writes the Graphify directive into CLAUDE.md and installs the PreToolUse hook in ~/.claude/settings.json. If the command fails, check that graphifyy is installed globally (pip show graphifyy).

Step 7 — Build initial knowledge graph

graphify .

This indexes the project. For large repos this may take 1–2 minutes. Output appears in graphify-out/GRAPH_REPORT.md.

If the project has docs/PDFs/images you want indexed semantically:

graphify . --deep

Step 8 — Compress CLAUDE.md

Run in the Claude Code session:

/caveman-compress

This compresses CLAUDE.md by ~46% and saves a backup to CLAUDE.original.md. The backup is already in .gitignore — do not commit it.

Step 9 — Print setup summary

After all steps complete, print this checklist for the user:

✅ CLAUDE.md         created + compressed
✅ RULES.md          created (extend with project shorthands)
✅ .claudeignore     created
✅ .gitignore        updated (graphify-out/, CLAUDE.original.md, settings.local.json)
✅ Graphify hook     installed in ~/.claude/settings.json
✅ Knowledge graph   built at graphify-out/GRAPH_REPORT.md
✅ Caveman           auto-activates via SessionStart hook

Next steps:
  • Extend RULES.md with project-specific abbreviations
  • Run /graphify . after major structural refactors
  • Run /update after adding new files
  • Optional: set up ~/notes/ (or Obsidian vault) for cross-session decision memory

Memory Layer (Optional — No Obsidian Required)

For cross-session decision memory, you have two options:

Option A — Plain Markdown (recommended, zero dependencies):

mkdir -p ~/notes/<project-name>/{architecture,decisions,logs}

Add to ~/.claude/CLAUDE.md:

## Memory
- Project notes at ~/notes/<project-name>/
- Read logs/ for recent decisions before starting a session

Option B — Obsidian vault:

graphify . --obsidian --obsidian-dir ~/vault/graphify/<project-name>

Files Created / Modified