Vibefile Documentation

The task runner for the AI era — declare what you want, let the machine figure out how.

What is Vibefile?

A Vibefile is a task runner configuration file. It describes what tasks do and when they run — not how they are implemented. The how is figured out at runtime by an LLM that reads the Vibefile alongside relevant context from your repo.

The format is inspired by Makefiles. The dependency model is identical. The only thing that changes is the recipe: instead of shell commands, you write a plain-English description of intent.

# Vibefile

model = claude-sonnet-4-6

build:
    "compile and bundle for production"

test:
    "run the full test suite with coverage"

deploy: test build
    "deploy to fly.io and verify the app came up healthy"

How it works

1. You write a Vibefile with targets described in plain English
2. On vibe run <target>, the CLI collects relevant context from your repo
3. An LLM generates the implementation (a shell script)
4. The generated script is cached — subsequent runs are instant and free
5. Dependencies are resolved and executed in topological order

The mental model

source code → Vibefile (intent) → compiled script (implementation)

first run:    Vibefile → LLM → .vibe/compiled/build.sh → execute
next runs:    Vibefile → .vibe/compiled/build.sh → execute (no LLM)

The Vibefile is your source code. The generated shell script is the binary. Just like a compiler, the expensive step happens once and the output is reused.

Key concepts

Quick start

# Install the CLI
go install github.com/vibefile-dev/vibe@latest

# Initialize a Vibefile for your project
vibe init

# Run a target
vibe run build

# See all targets
vibe list