Wigg

Overview

Wigg is a lightweight TypeScript harness for running AI coding agents in an iterative loop instead of a one-shot prompt. It keeps state between runs, feeds prior verification failures back into the next iteration, and keeps going until the task is genuinely complete.

The core idea is simple: if an agent makes partial progress, trips a test, or exits early, the harness should capture that context and give the next iteration a better shot instead of forcing the operator to manually restate everything.

Wigg is my implementation of the Ralph Wiggum technique described by Geoffrey Huntley, and it was directly inspired by Th0rgal/open-ralph-wiggum, which helped turn the technique into a concrete, reusable CLI workflow.

What It Does Well

• Wraps multiple agent CLIs so the same workflow can target Claude Code, OpenCode, or Codex.
• Runs verification commands after each iteration and feeds failures back into the loop.
• Tracks task state and recent history so the agent can resume with more context and less repetition.
• Works cross-platform with explicit Windows shell handling to avoid the usual CLI quoting and stdin edge cases.
• Auto-commits progress when desired so each iteration can leave behind a reproducible checkpoint.

Why I Built It

I wanted a small tool that treats agentic coding like an engineering loop instead of a chat session. That means preserving context, enforcing verification, and making it cheap to keep iterating until the output is good enough to trust.

The result is a tool that feels closer to a pragmatic operator workflow: define the mission, choose the agent, add checks, and let the loop keep tightening the output. In that sense, Wigg is not trying to invent a new pattern so much as carry the Ralph Wiggum technique into a TypeScript-first implementation shaped around the workflows I actually use.

Acknowledgements

• Geoffrey Huntley for articulating the Ralph Wiggum technique.
• Th0rgal/open-ralph-wiggum for proving out the pattern as a practical open-source tool and helping shape the direction for Wigg.

Usage Snapshot

pnpm dlx @cbnsndwch/wigg "Refactor the auth flow" \
  --agent claude-code \
  --verify "lint:pnpm lint" \
  --verify "test:pnpm test"

Wigg also supports task mode for multi-step missions, mission files for longer prompts, and loop context injection while a run is active.

Current Direction

The project is already published to npm and has moved past the first release with improvements around terminal output, package publishing, and cross-platform execution. The next layer of value is making the loop even more observable and dependable for day-to-day coding work.