TrendWhat rose on GitHub, every day at 07:30 KST · 한국어

GitHub · as of November 15, 2026 · View on GitHub →

google/elephant-goldfish-model

AI coding plugin for design

#9 on GitHub's weekly trending, November 15, 2026 · trending 3 of the last 30 days

Write and review software blueprints with AI agents

An 'elephant' agent that remembers project context writes the design draft, while a 'goldfish' agent that has forgotten previous conversations re-reads it to find hidden assumptions and gaps. This plugin installs into AI coding assistants to support document-based software design. The agents iterate on drafting and review until the user makes final architecture decisions and approves the design.

Design and review with separated context

The core of this tool is separating context. The 'elephant' agent remembers all project requirements and previous decisions while exploring the problem and writing the specification draft. In contrast, the 'goldfish' reviewer reviews without knowing previous conversation content, looking only at the document and referenced source code. This reveals assumptions developers may not be aware of, unclear requirements, and implementation risks. Finally, once the user resolves trade-offs and approves the design, a coding agent implements it.

Flow from design to review

The work begins with idea exploration. The /egm-brainstorm command discusses features and saves the agreed direction to a concepts_brief.md file. Running /egm-draft then converts this brief into design_draft.md, with the agent waiting for user feedback on each section. Once the draft is complete, the /egm-review command has new agents check the document for clarity, alignment with code, and implementation planning. The user can selectively apply review results and give final approval, after which code can be written according to the design.

Checking for inconsistencies between existing documents and code

In addition to creating new designs, you can check how well existing documents align with the current code. The /egm-drift-report command allows the agent to compare documents in a folder with the current source code and report differences. Users can use this report to decide whether to update the documents or modify the code. Additionally, /egm-review-loop allows setting specific reviewer groups and criteria to review other types of documents.

Installation and usage environment

This plugin is written in Python and can be used with various AI coding agents such as Claude Code, Codex CLI, and Antigravity CLI. Codex CLI and Antigravity CLI also provide their own plugin installation commands. After installation, you must start a new session, and using the review feature requires an environment capable of creating new sub-agents. In environments where sub-agent creation is not possible, the review process is interrupted.

Pre-usage checks

This tool is specialized for software specification design, and the review process does not replace testing or human judgment. It is released under the Apache 2.0 license but is not an officially supported product by Google. Users should first verify that the sub-agent contexts generated by the agent work correctly and that their project structure is suitable for a document-based design workflow. In particular, since users remain ultimately responsible for complex architecture decisions, critically reviewing agent output rather than blindly trusting it is essential.

Written by AI from this repository's README on November 13, 2026. For the facts, the original on GitHub is the reference.

By the numbers

GitHub trending history
3 days between Jun 19, when Trend began recording, and Nov 15, 2026 · first on Nov 13

Read from GitHub, as of November 15, 2026.

Language
Python
Topics
agent-skills · ai-agents · antigravity · claude-code · claude-code-plugin · claude-skills · codex · codex-plugin
Stars
19,595
Contributors
36
Forks
1,455
Latest release
v0.1.1 · November 6, 2026
Commits
203
Last commit
November 14, 2026
Open issues
72
Open pull requests
39

Related repositories

View on GitHub → · Homepage