AgenticGoGo

Define your agent workflow as code — YAML or Rust — and let a deterministic outer loop
with incorruptible judges drive it to done.
Loop Engineering · Graph Engineering · Agentic Workflows as Code

Stop typing “go go”.

---

Are you constantly typing **“go go”**, **“continue”**, **“keep going”** to nudge your coding agent through a long plan? Do even spec-driven approaches stall mid-flight, run out of context, or quietly stop one step short — leaving you to babysit a terminal?

**Then AgenticGoGo is for you.**

AgenticGoGo (`agg`) is a framework for **Loop Engineering** (e.g. Ralph Loops), **Graph Engineering** and **Agentic Workflows as Code**. You define the workflow as a versioned artifact — in **YAML** for a sequence that laps, or in **Rust** when the flow has to branch — and `agg` runs it as a deterministic outer **[Ralph loop](https://ghuntley.com/ralph/)** around a stochastic inner coding agent (currently supported: [Claude Code](https://claude.com/claude-code), [OpenAI Codex](https://developers.openai.com/codex/cli), [GitHub Copilot](https://github.com/github/copilot-cli)) — relaunching a **fresh** session each cycle, verifying its work against gates *it can't fake* (the **judges**), and repeating until your goals are actually met. The loop is plain code: it never hallucinates a decision. The agent does the work, inside one step, and never decides when it's done. *(A similar LLM-based approach — generate → verify → keep, as in evolutionary code search — was proposed years ago, outside the Ralph-loop community, by DeepMind's [AlphaCode](https://arxiv.org/abs/2203.07814) and its open-source variant [CodeEvolve](https://arxiv.org/abs/2510.14150).)*

A **judge** is a smal