5.2 Turn the output into a runnable project

Goal: move from text to code

The model’s output is not a project until it exists as files you can run locally. This page is about converting the generated text into a runnable repo with a repeatable run/test loop.

Create the project folder

Create a new folder (and optionally a git repo). Example:

mkdir hello-vibe-calc
cd hello-vibe-calc
git init

If you use Python virtual environments, create one now:

python -m venv .venv
source .venv/bin/activate

Create files exactly as generated

Copy the model output into files. Be literal:

• match file names and folder names exactly,
• don’t “improve” anything yet,
• don’t merge files to save time,
• don’t refactor until it runs.

Quick file-tree check

You should end up with something like:

hello-vibe-calc/
  calc/
    __init__.py
    __main__.py    (may be added if missing)
    cli.py
    parser.py
    eval.py
  tests/
    test_calc.py
  README.md

Run the CLI (SP1)

Try the smallest “it works” command first. Your acceptance criteria from 5.1 should include something like:

python -m calc "2+2"

If python -m calc fails

The most common reason is that the package lacks calc/__main__.py. This is a great example of a tiny follow-up prompt:

Add `calc/__main__.py` so that `python -m calc "2+2"` routes to the existing CLI entrypoint.
Do not change behavior. Show the new file only.

Run tests (SP2)

Run the tests using standard library tooling:

python -m unittest -v

If the project is structured as a package, you may also use discovery:

python -m unittest discover -v

What “tests pass” means here

• the CLI acceptance tests behave as expected,
• invalid input fails in a controlled way,
• you can refactor without fear because you can detect regressions.

Common setup failures (and quick fixes)

Import errors

• Cause: inconsistent module/package names or missing __init__.py.
• Fix: prompt the model: “Fix imports to match this file tree; do not change behavior; show diff-only.”

Tests call the wrong entrypoint

• Cause: tests assume a function exists (e.g., evaluate(expr)) but code exposes something else.
• Fix: choose an API and align tests/code to it. Prefer a small, explicit API like calc.eval.evaluate(expression: str) -> float.

Behavior mismatch (e.g., whitespace, unary minus)

• Cause: parser rules are incomplete.
• Fix: add a test that demonstrates the bug, then prompt: “Make the smallest change to pass this test.”

Lock in with Git

Once the CLI runs and tests pass, make a commit. This is a vibe-coding superpower: you can always reset back to a known-good state.

git add .
git commit -m "Hello Vibe: working CLI calculator"

Where to go next

• 5.3 The first refactor prompt (make it clean, not clever)
• Back to 5.1