Google open-sources its Elephant-Goldfish Model plugin

Google describes the plugin as a way to put the Elephant-Goldfish Model into practice at scale[1]. It includes customizable Markdown rules, defined subagent personas and structured review workflows[1]. The plugin automates context transfer between Elephant and Goldfish sessions, validation loops and design-document rendering[1]. Its workflow takes users through three co-design phases[1]. Developers load system context, follow a no-code rule, co-design features, debate trade-offs and challenge assumptions[1].

The Elephant is a context-rich, long-horizon session that maintains architectural vision and project memory[1]. Google describes it as a phase dedicated to design and intent, with no code written during that phase[1]. Goldfish sessions are stateless and have zero context, so they can review specifications without sharing the designer’s blind spots[1]. Developers iteratively draft a Markdown design document covering the problem, technical plan, alternatives and specific files to change[1]. Stateless AI instances then probe the design for weaknesses[1].

The plugin automates a three-step validation sequence[1]. If Goldfish cannot accurately summarize the plan, the document is ambiguous and needs further drafting[1]. Multiple skeptical reviewers look for edge cases, hidden dependencies and logical gaps[1]. An implementation-readiness check verifies that file paths, data structures and function signatures are present and unambiguous[1]. Google says the intended workflow leaves architectural decisions with the human, who approves the specification before implementation[2]. After implementation, the project site advises comparing the code with the specification and reporting divergences[2].

The plugin is released under the Apache 2.0 license[1]. The project site lists Antigravity as an installation option[2]. However, it also says the plugin is not an officially supported Google product[2]. Google says the plugin was first developed internally for its software engineers before being open-sourced[1], and that thousands of employees use it every day to ship production code[1]. Those details offer context on its internal use, but the support caveat remains relevant for teams assessing adoption.