06 — The Prompt Runner

What We’re Building

The prompt runner is the simplest tool in our toolkit: take a prompt.txt from a directory, feed it to Ollama, save the output.

But even this simple tool introduces an important pattern: each prompt lives in its own directory with everything it produces. This keeps your work organized and reproducible.

The Script

This is the run_prompt.sh generated by our one-shot prompt (plan1.md). Copy and save it:

#!/bin/bash

# Prompt Runner — feeds prompt.txt to Ollama and saves output

# Check for directory argument
if [ -z "$1" ]; then
    echo "Usage: $0 <directory>"
    echo "Example: $0 ./my-prompt"
    exit 1
fi

DIR="$1"
PROMPT_FILE="$DIR/prompt.txt"
OUTPUT_FILE="$DIR/output.txt"
MODEL="gemma4:26b"

# Check that prompt.txt exists
if [ ! -f "$PROMPT_FILE" ]; then
    echo "Error: $PROMPT_FILE not found."
    echo "Create a prompt.txt file in $DIR"
    exit 1
fi

# Run the prompt through Ollama and save output
echo "Running prompt in $DIR using $MODEL..."
cat "$PROMPT_FILE" | ollama run "$MODEL" --nowordwrap > "$OUTPUT_FILE"
echo "Done! Output saved to $OUTPUT_FILE"

Setup

# Save the script
nano run_prompt.sh
# (paste the code above, save and exit)

# Make it executable
chmod +x run_prompt.sh

Usage

Create a directory with a prompt.txt file:

mkdir -p my-first-prompt
echo "Write a haiku about programming" > my-first-prompt/prompt.txt

Run it:

./run_prompt.sh my-first-prompt

Output:

Running prompt in my-first-prompt using gemma4:26b...
Done! Output saved to my-first-prompt/output.txt

Check what the model wrote:

cat my-first-prompt/output.txt

Key Design Decisions

Why this approach?

1. Bash, not Python — zero dependencies. Works on any Mac out of the box.
2. One directory per prompt — keeps inputs and outputs together
3. Overwrite, don’t version — (simple! YAGNI!) — we want the latest output, not a collection of old ones
4. Error handling — checks that the directory and prompt file exist before running

The --nowordwrap Flag

You’ll notice ollama run model --nowordwrap. Without this flag, Ollama may wrap the model’s output at a fixed width, which can break code formatting. The --nowordwrap flag preserves the model’s original formatting.

What You’ve Learned

• A simple bash script can run prompts through Ollama
• Keeping each prompt in its own directory organizes your work
• Error handling prevents confusing failures
• --nowordwrap preserves code formatting

Next: 07 — Self-Improving →