google/elephant-goldfish-model
Developer ToolsAn open-source plugin that helps coding agents draft software specifications and runs fresh-context reviewers to expose gaps before implementation.
What it does
Elephant-Goldfish Model (EGM) is a plugin for AI coding assistants that separates design work into two roles. The "elephant" keeps the project context and drafts the specification; "goldfish" reviewers read the draft and its referenced source material without the earlier conversation, which makes unstated assumptions and blind spots easier to surface. The developer resolves trade-offs and approves the design before implementation.
The plugin ships as a set of skills: brainstorming, drafting, resolving review findings, bootstrapping docs for an existing codebase, a five-stage design review, a configurable review loop, and a drift report that compares checked-in documents with current source code How it works and skills. The built-in draft/review workflow targets software specifications; other documents can use /egm-review-loop with chosen reviewers and standards, while bootstrap and drift reporting apply to code README.
Install and first run
EGM is installed through the plugin marketplace of a supported coding agent. Pick one:
- Claude Code:
claude plugin marketplace add google/elephant-goldfish-model, thenclaude plugin install egm@egm. - Codex CLI:
codex plugin marketplace add google/elephant-goldfish-model, thencodex plugin add egm@egm. - Antigravity CLI:
agy plugin install https://github.com/google/elephant-goldfish-model. - ChatGPT / Codex desktop: install the published plugin from the desktop plugin listing; it is desktop-only install commands.
Start a new session after installing. In Antigravity (agy CLI) and Claude Code, commands carry the /egm: prefix; in Codex, choose the matching EGM skill from the skill picker. If prefixed completion confuses your agy version, keep the explicit prefix.
To review an existing design document or a PR plan saved as Markdown against the live code:
/egm:egm-review docs/designs/webhooks.md
Fresh reviewers check the design, its clarity, its fit with the code, and the implementation plan. To check documentation drift instead:
/egm:egm-drift-report docs/designs/
The agent compares each document in the folder with the current source and reports where they diverge; you decide whether to update the document or the code 60-second review.
For a new design, run /egm-brainstorm <feature> (e.g. "Add webhook delivery to this project"), save the agreed direction as concepts_brief.md, run /egm-draft to produce design_draft.md with pauses for feedback, then /egm-review design_draft.md. The full flow, including where approved specs are stored, is in the workflow guide.
Constraints
Brainstorm and review require a session that can create fresh subagent contexts; if the agent cannot, EGM stops the review rather than running with shared context. Support differs by agent and version, and the setup guide records capabilities and known limits alongside a support status section; the agy 1.4.8 command-completion issue is tracked separately issue #196. A review does not replace tests or human judgment.
EGM is licensed under Apache 2.0 and is explicitly not an officially supported Google product license.
HotRepo's take
The project was ranked 5th on the all-language weekly GitHub list for 13 November 2026, when the snapshot recorded 18,504 stars and 1,377 forks, with 5,555 stars gained in that week. No all-language daily ranking was recorded. The specific design — separate a context-rich drafter from context-free reviewers inside existing agent harnesses — fits the current interest in spec-driven and multi-agent workflows rather than introducing a standalone app. Hacker News attention has been thin: several submission threads with one or two points, none with substantive discussion. That makes the weekly ranking the main external signal, and it measures visibility, not whether the review skills improve design quality.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
Hacker News
You might also like
- wshobson/agentsA cross-harness catalog of coding agents, skills and commands so one set of prompts works across Claude Code, Codex, Cursor and others.
- DietrichGebert/ponytailA skill pack that pushes coding agents to reuse existing code and write only what the task needs.
- Gentleman-Programming/gentle-aiWrap the AI coding agent you already use with persistent memory, a sized workflow, and evidence-bound review.
- haydenschmitty/floorplanAn AI plugin for Claude Code or Codex that maps a codebase into an interactive plain-English diagram.
- open-gsd/gsd-coreA phased workflow system that keeps AI coding agents context-lean across research, planning, execution, verification and shipping.