Auggie Wiggum - A Ralph Wiggum Loop for Auggie CLI

README.md

Auggie Wiggum 🔄

A robust Bash wrapper for the auggie agent that implements the Ralph Wiggum Technique.

Features

• Autonomous Looping: Runs the agent repeatedly until it outputs "I'm done".
• Persistent Context: Reads previous iteration logs so the agent can "remember" mistakes and self-correct.
• Safe Logging: Stores all thought processes in a hidden .auggie_wiggum/{timestamp}/ folder to avoid breaking Git hooks or polluting your commit history.
• Pure Autonomy: No external test runners required; relies on the agent's own internal verification logic.

Usage

1. Create a PROMPT.md file with your task.
2. Run ./auggie_wiggum.sh.

auggie_wiggum.sh

#!/bin/bash

# ==========================================
# RALPH WIGGUM LOOP (Pure Autonomy Mode)
# "The Agent is the boss. The script is the memory."
# ==========================================

# --- CONFIGURATION ---
TASK_FILE="PROMPT.md"
MAX_ITERATIONS=15
ITERATION=0

# Logging Setup (Timestamped)
# Creates a unique folder for this session: .auggie_wiggum/2026-01-09_14-30-00
CURRENT_DATE_AND_TIME=$(date +"%Y-%m-%d_%H-%M-%S")
LOG_DIR="./.auggie_wiggum/$CURRENT_DATE_AND_TIME"

mkdir -p "$LOG_DIR"

# --- PRE-FLIGHT CHECKS ---
if [ ! -f "$TASK_FILE" ]; then
    echo "❌ Error: $TASK_FILE not found."
    exit 1
fi

echo "🚀 Starting Auggie (Autonomous Mode)..."
echo "📂 Session saved in: $LOG_DIR"
echo "-----------------------------------"

# --- THE LOOP ---
while [ $ITERATION -lt $MAX_ITERATIONS ]; do
    echo "🔄 Iteration $ITERATION"

    # 1. BUILD CONTEXT (MEMORY)
    # The script reads everything the agent "thought" and "did" in previous iterations.
    HISTORY_CONTEXT=""
    
    if [ $ITERATION -gt 0 ]; then
        echo "   (Reading memory from previous iterations...)"
        for (( i=0; i<ITERATION; i++ )); do
            PREV_FILE="$LOG_DIR/iteration_$i.txt"
            if [ -f "$PREV_FILE" ]; then
                STEP_CONTENT=$(cat "$PREV_FILE")
                HISTORY_CONTEXT+=$'\n'"--- HISTORY (Iteration #$i) ---"$'\n'"$STEP_CONTENT"$'\n'
            fi
        done
    fi

    # 2. CONSTRUCT THE PROMPT
    # We instruct the agent to use its own tools to verify the work.
    FULL_PROMPT="
    $(cat "$TASK_FILE")

    ====== SHORT-TERM MEMORY (What you already tried) ======
    $HISTORY_CONTEXT
    ========================================================

    LOOP INSTRUCTIONS:
    1. You are running in an autonomous loop.
    2. Analyze the history above. If you tried something and it failed, try a different approach.
    3. YOU are responsible for ensuring the code works. Run your own internal checks/tests if possible.
    4. If the task is 100% COMPLETE and TESTED, write exactly \"I'm done\" on the very last line.
    5. If not finished, briefly describe your progress and what you will do next.
    "

    # 3. RUN AUGGIE
    # --print: captures output to file | --quiet: suppresses system logs
    OUTPUT=$(auggie --print --quiet "$FULL_PROMPT")

    # 4. SAVE LOG (The "Memory")
    CURRENT_LOG_FILE="$LOG_DIR/iteration_$ITERATION.txt"
    echo "$OUTPUT" > "$CURRENT_LOG_FILE"
    echo "📝 Thought process saved: $CURRENT_LOG_FILE"

    # 5. CHECK IF AGENT DECIDED TO STOP
    # Looks for "I'm done" on the last line (case insensitive)
    LAST_LINE=$(echo "$OUTPUT" | tail -n 1 | xargs | tr '[:upper:]' '[:lower:]')

    if [[ "$LAST_LINE" == *"i'm done"* ]]; then
        echo "-----------------------------------"
        echo "✅ Agent reported completion."
        echo "📂 Full history available at: $LOG_DIR"
        exit 0
    fi

    ((ITERATION++))
    
    # Safety pause to avoid API rate limits
    sleep 2
done

echo "🛑 Max iterations reached without completion signal."
exit 1