███████╗████████╗ █████╗ ██████╗ ████████╗███╗   ███╗███████╗
██╔════╝╚══██╔══╝██╔══██╗██╔══██╗╚══██╔══╝████╗ ████║██╔════╝
███████╗   ██║   ███████║██████╔╝   ██║   ██╔████╔██║█████╗
╚════██║   ██║   ██╔══██║██╔══██╗   ██║   ██║╚██╔╝██║██╔══╝
███████║   ██║   ██║  ██║██║  ██║   ██║   ██║ ╚═╝ ██║███████╗
╚══════╝   ╚═╝   ╚═╝  ╚═╝╚═╝  ╚═╝   ╚═╝   ╚═╝     ╚═╝╚══════╝

Auto-generated codebase orientation for AI.
One scan. One file. Any model.

startme

Stop letting AI waste tokens scanning your codebase. Give it the answer.

startme scans your project and generates a STARTME.txt — an ultra-compact orientation file that tells any AI assistant exactly what your project is, how to run it, and where everything lives. Under 50 lines, under 200 tokens. Works with Claude, GPT, Gemini, Codex, or anything else.

---

The Problem

Every time an AI opens your codebase:

"Let me scan through the project structure to understand..."
"Looking at package.json..."
"Reading the README..."
"Checking the source directory..."

That's 5,000-50,000 tokens burned on orientation. Every. Single. Session.

The Solution

npx startme --write

Generates a STARTME.txt like this:

wade — AI screen automation with token-efficient capture.
Stack: node, commander, @anthropic-ai/sdk, @modelcontextprotocol/sdk

COMMANDS
  start: node bin/wade.js
  mcp: node src/mcp-server.js
  chat: node bin/wade-chat.js
  test: node --test 'test/**/*.test.js'

ENTRY POINTS
  wade → bin/wade.js
  wade-chat → bin/wade-chat.js

STRUCTURE
  bin/ (2 files)
  src/ (9 files)
  swift-bridge/ (2 files)
  test/ (4 files)

DEPS: 4 runtime, 0 dev

~140 tokens. The AI reads this one file and knows everything. No scanning.

---

Usage

# Scan current directory, print to stdout
npx startme

# Write STARTME.txt to project root
npx startme --write

# Output as a section for AGENTS.md or CLAUDE.md
npx startme --section

# Scan a specific directory
npx startme /path/to/project

# Scan ALL projects in a parent directory
npx startme --all /path/to/projects

# Raw JSON output
npx startme --json

Batch Mode

Scan every project on your drive at once:

npx startme --all ~/projects --write

Output:

  kalshi-betting-bot → STARTME.txt (89 tokens)
  wade               → STARTME.txt (140 tokens)
  claudemail         → STARTME.txt (62 tokens)
  transcript-api     → STARTME.txt (118 tokens)

Total: 18 projects, ~1124 tokens

1,124 tokens for 18 entire projects. That's less than reading one medium-sized source file.

---

What It Detects

Source	What it extracts
package.json	Name, description, scripts, entry points, key deps
pyproject.toml	Name, description, Python stack
Cargo.toml	Name, description, Rust stack
go.mod	Module name, Go stack
Dockerfile	Exposed ports, CMD, docker stack
docker-compose.yml	Compose command
Makefile	Build targets
.env.example	Required environment variables
Source files	Port numbers
LICENSE	License type
Directory tree	Structure with file counts

How It Differs From...

Tool	What it does	How startme differs
CLAUDE.md	Instructions for Claude (rules, preferences)	startme is orientation, not instructions. Model-agnostic.
AGENTS.md	Mixed instructions + context	startme is auto-generated, pure orientation, minimal tokens.
Repomix	Dumps entire codebase into one file	startme is ~200 tokens, not ~50,000.
llms.txt	Website docs for AI consumption	startme is for codebases, not websites.
README.md	Human-readable project docs	startme is machine-readable, no prose.

STARTME.txt vs CLAUDE.md

They complement each other:

STARTME.txt  →  "What is this project and how do I run it?"  (auto-generated)
CLAUDE.md    →  "How should I behave while working on it?"   (hand-written)

Use both. Or paste startme output into the top of your CLAUDE.md with npx startme --section.

---

Zero Dependencies

Pure Node.js built-ins. No npm install. Just npx startme.

---

License

MIT