I'm Learnding!

💭 The Philosophy

Ralph Wiggum tries stuff. It doesn't work. He tries again. He eats paste. He picks his nose. He tells Lisa he choo-choo-chooses her. And somehow, through pure persistence and obliviousness to failure, things happen.

That's a workflow. That's the workflow. And it gets better when you run seven Ralphs in parallel.

🔄 The Loop

Regular Ralph is a bash loop:

while true; do
    claude prompt.md
    # Stop hook checks: "Are you done? Really done?"
    # If not done, loop continues
done

When Claude tries to exit, a stop hook catches it. Keep going until the task is actually finished. This is beautiful in its stupidity. No complex orchestration. No state machines. Just: keep trying until it works.

⚡ But What If You Had More Ralphs?

One Ralph picks his nose. What if seven Ralphs picked seven noses at the same time?

That's Parallel Ralph. The key word is DIAGONALIZE:

The magic word is DIAGONALIZE. Instead of doing tasks one at a time, you launch N agents in a single message:

# One message. Seven Ralphs. All picking simultaneously.
Task("Audit auth", subagent_type="Explore")
Task("Audit API", subagent_type="Explore")
Task("Audit frontend", subagent_type="Explore")
Task("Audit database", subagent_type="Explore")
Task("Audit tests", subagent_type="Explore")
Task("Audit docs", subagent_type="Explore")
Task("Audit deploy", subagent_type="Explore")

They all run in parallel. They all come back with findings. You collect them, score them, plan fixes, execute in parallel again.

👀 The 8 Perspectives

When Ralph Review looks at code, it doesn't just look once. It looks 8 times, through 8 different eyes:

Each perspective scores /100. You build a matrix. You find patterns. The Engineer score is always lowest (tests, types, magic numbers). The Enthusiast score is often highest (cool features). You fix the gaps.

⚔️ Why This Works

The key insight: Claude Code's Task tool can launch multiple agents in one message. They run in parallel. You're not waiting in sequence. You're diagonalizing the problem space.

💾 The Persistence Trick

Ralph forgets everything between episodes. But Parallel Ralph has Memory MCP:

# Start of session: What did past-Ralph learn?
mcp__memory__read_graph()
mcp__memory__search_nodes("PENDING")

# End of session: Save for future-Ralph
mcp__memory__create_entities([
    {"name": "Auth_Audit_Jan1",
     "entityType": "Audit",
     "observations": ["Score: 72/100", "P0: Token bug"]}
])

Context windows reset. Claude Squad edits your prompt. But Memory survives. Future-you resumes where past-you stopped.

🎮 Spawn Some Ralphs

🎯 The Real Secret

Ralph Wiggum isn't smart. He's persistent. He's oblivious to failure. He keeps trying things until something works, and he's genuinely delighted when it does.

That's the workflow. Keep picking. Keep tasting. Keep trying. But do it in parallel, score what you find, and remember what you learned.