Elephant Goldfish Model

google/elephant-goldfish-modelReviewed by AI

What it does

A plugin with seven skills for coding agents (Antigravity, Claude Code, Codex and others). One agent keeps all your project context and drafts the specification for a feature; other agents read only the document and the code it points to, with none of the earlier chat, and list missing assumptions and risks. You decide what to change and approve the design before any code is written. For developers planning a feature, not for small bug fixes. Nothing to install beyond the coding agent you already pay for, plus one install command.

How easy it is to install

B · installer plus one simple step · Windows

Windows B
macOS B
Linux B

B wherever your coding agent runs: once you have the agent, installation is one or two commands (`claude plugin marketplace add google/elephant-goldfish-model` then `claude plugin install egm@egm`, or `agy plugin install ...` for Antigravity), then you start a new session and type `/egm-brainstorm`. There is nothing else to install, but you do need a paid agent account.

Numbers on GitHub

Stars: 7,777on Oct 31, 2026
Downloads: 14438 this week
Growth: +4,482stars in 4 days

Stars over time

Facts

Runs on: Windows, macOS, Linux
You need to install: Nothing extra
To build from source: Python
Disk space: 100 MB–300 MB · estimate
Accounts or keys: None
Cost: Free, API costs extra · The files are Apache-2.0 and free. Using them needs a coding agent that costs money (Claude Code, Codex, Antigravity and similar), and a review spawns several fresh agents, so one design can use a lot of tokens.
License: Apache-2.0 (Permissive)
Last update: Oct 30, 2026

Risk signals (5)

AttentionVery new repository with many stars
5952 stars in 46 days
automatic check
Infonot an officially supported google product
README, License section: "[Apache 2.0](LICENSE). This is not an officially supported Google product." It sits in the `google` GitHub organisation, but comes with no Google support.
AI review
Inforeview stops on agents that cannot create subagents
README, "How it works": "Goldfish reviews require fresh subagent contexts. If your agent cannot create them, EGM stops the review." The same section adds that "A review does not replace tests or human judgment."
AI review
Infoadoption claim cannot be checked
README: "Thousands of Google engineers use EGM, an open-source plugin, to draft specs with deep-context agents". Nothing in the repository backs that number up.
AI review
Infoearly version
The repository was created on 2026-09-13 and has a single release, v0.1.0 (2026-10-27), with 53 open issues, so the skills and commands may still change.
AI review

Evidence is shown in its original language.

What Radar could not check

Whether the skills install and run as described in each listed agent, including Claude Cowork and ChatGPT Work.
How many tokens, and how much money, a full brainstorm-draft-review cycle uses on your own plan.
Whether the reviewer findings are actually useful on your codebase, or mostly generic remarks.
What the skills read from your project and send to the model during a review.