Google’s Elephant-Goldfish Plugin Makes Coding Agents Write and Defend a Spec First

Google has open-sourced the Elephant-Goldfish Model (EGM) plugin, a design-and-review layer for AI coding agents, MarkTechPost reports.

How it works

The “Elephant” is a long-running session with all the project context. It writes the spec, not the code. “Goldfish” are fresh subagents with no history that only read the draft and attack it. You settle the trade-offs and approve the design before any code gets written.

• Seven skills, nine reviewer personas, and commands like /egm-brainstorm, /egm-draft and /egm-review.
• Works with harnesses that can spawn fresh subagents, including Claude Code, Codex and Antigravity. It uses your existing model quota.
• Apache 2.0, and Google notes it’s not an officially supported product.
• Downsides: heavy token use, and one open issue reports AGENTS.md leaking into Codex reviewers.

My take

Catching a bad assumption in a Markdown file is a lot cheaper than catching it in a PR. Watch your token bill.

Review before you install: read the release notes yourself, back up your config, and try new versions somewhere you can afford to break first.

God Bless, and Happy hacking!

References & Further Reading

• Google Elephant-Goldfish model plugin (MarkTechPost)