Google Open-Sources Elephant-Goldfish Model Plugin for Spec-First Coding
Google has open-sourced the Elephant-Goldfish Model (EGM) plugin, a design-and-review layer for AI coding agents. The plugin forces an agent to write a specification before it writes any code, then dispatches fresh subagents with no conversation history to attack that spec. Google reports that thousands of its engineers already use the tool. The premise is deliberately simple: catch wrong assumptions in a Markdown file, not in a pull request.
TL;DR
- Size: 7 skills, 9 reviewer personas, 6 brainstorm lenses, and 5 review stages. No model weights are bundled; your agent supplies the LLM.
- Runs on: any harness that can spawn fresh subagents, including Antigravity, Claude Code 2.1.271+, Claude Cowork, Codex, and ChatGPT. It consumes your existing model quota.
- Performance: no accuracy benchmark has been published. The evidence rests on adoption and end-to-end harness test runs.
- Best result: Codex 0.163.0 completed 15 native fresh-context reviewers in sequence, with two live tests passing in 573.13 seconds.
- Worst result: the Claude Cowork run stopped after stages 5 to 7 due to a usage limit. Issue #237 reports
AGENTS.mdleaking into Codex reviewers. - Bottom line: the strongest feature is forced independent review before code is written. The weakest points are heavy token consumption and unverified isolation on some hosts.
The Core Idea: Specifications Before Code
The EGM plugin shifts quality control earlier in the development lifecycle. Rather than relying on post-hoc code review, it requires the coding agent to produce a written specification first. Independent reviewer subagents then examine that specification with no access to the authoring conversation, reducing the risk of shared blind spots.
This spec-first approach aligns with a broader 2026 industry trend toward agentic workflows that emphasize planning and verification over raw code generation speed.
via MarkTechPost