Guide · Agent skill
How to use the EGM Goldfish Design Review agent skill (egm-review)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The egm-review skill is an instruction package that tells an AI agent, such as Claude Code, to review a design document in five phases using separate reviewer subagents. It is written for teams that already follow the EGM Goldfish design process and have the EGM repository files. After this guide you will know what the skill does, how to install it from the ZIP, how to trigger it with a plain request, and what it needs before it will run.
EGM Goldfish Design Reviewgoogle/elephant-goldfish-model · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the egm-review skill ZIP from this page and unzip it.
- 2Place the folder. Move the egm-review folder into ~/.claude/skills/ for personal use, or into your project's .claude/skills/ folder.
- 3Add the EGM repository files. Make sure the persona files in agents/ and the docs such as docs/harnesses.md are available to the agent.
- 4Confirm a subagent tool. Check that your harness has a subagent tool such as Claude Code's Agent tool, since the skill stops without one.
- 5Ask for a review. Request a review of your design document path under the EGM protocol.
- 6Choose a mode. Pick interactive, consolidated, feature spec only, or implementation plan only when the agent asks.
- 7Read the reports. Open the report files the agent writes and answer any chat prompts the review raises.
What an agent skill is and what egm-review makes the agent do
An agent skill is a folder containing a SKILL.md file. The file holds instructions that an AI agent loads when the skill applies to the task. The egm-review skill is for reviewing design documents. Its SKILL.md describes the EGM Goldfish Review Protocol and its five phases: Initial Zoom Out Review, Comprehension Review, Design Critique, Intermediate Zoom Out Review, and Implementation Readiness.
Each reviewer runs as a subagent in a fresh context with no prior conversation, which the source calls a Goldfish review. The agent writes reports to files rather than pasting them into chat, asks you for choices through chat prompts, and is instructed to stress-test the document and avoid flattery.
- Phase 1: Initial Zoom Out Review
- Phase 2: Comprehension Review
- Phase 3: Design Critique
- Phase 4: Intermediate Zoom Out Review
- Phase 5: Implementation Readiness
Install the egm-review skill from the ZIP
Download the ZIP of the skill from this page and unzip it. For a personal install, place the unzipped egm-review folder inside ~/.claude/skills/. For a single project, place it inside that project's .claude/skills/ folder.
In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/egm-review/
- Project: .claude/skills/egm-review/ inside your project folder
- Claude apps: Settings → Capabilities → Skills
Make sure the EGM repository files are available
The skill does not work alone. Its SKILL.md points to other files that are not included in the skill: persona files in agents/, and documents such as docs/harnesses.md, docs/codex.md and docs/claude.md. The reviewers are defined by these persona files, so the EGM repository must be present where the agent can read it.
The skill also requires an agent harness with a subagent tool, such as Claude Code's Agent tool, Codex spawn_agent, or Antigravity invoke_subagent. If your harness has no subagent tool, the skill tells the agent to stop rather than run a single-context review.
- Agent harness with a subagent tool
- EGM repository files, including persona files in agents/
- A target design document file to review
Trigger the review with a natural request
Give the agent the path to your design document and ask for a review under the EGM protocol. The agent will ask you to choose an execution mode if you do not name one. The skill's Inputs are a TargetFile (the design document path), an optional GoldfishContext (a reading list), and an optional Step preset.
If you omit the path, the agent will ask for it in chat as a plain text message. Reviewers always read files that the design document links or references, even when you provide no extra context.
- Ask for a review of a specific path, for example: review docs/my-design.md using the EGM protocol
- Choose an execution mode when prompted: interactive, consolidated, feature spec only, or implementation plan only
Understand the output and the review modes
The skill's artifact-first rule means the agent writes the Discussion Summary, design drafts and all review reports to files, then tells you the file path in chat. Expect several report files, because each phase produces its own report.
The interactive mode runs the five stages one at a time and checks in after each. The consolidated mode runs the full flow without stopping. The two narrower modes run only the feature spec evaluation or the implementation plan evaluation.
- Interactive: step by step, with a check-in after each review
- Consolidated: full flow without stopping
- Feature spec only: evaluates the design without considering implementation
- Implementation plan only: evaluates implementation readiness
Safety notes: scripts and network
Based on the skill listing, this skill does not run scripts and does not use the network. Its files are the SKILL.md instructions and agents/openai.yaml. The work happens through the agent's own subagent and file-writing tools, which means the reports and any edits land in your working directory.
Because the reviews write files and spawn several subagents, check the output locations before you run a review on important work. Review the persona files in the EGM repository as you would any instructions an agent will follow.
- No scripts are run by the skill
- No network use is listed for the skill
- Reports are written to the working directory or the harness artifact directory
Tips for useful reviews
Give the reviewer a complete design document with its links intact, since reviewers read referenced files automatically. Decide before you start whether you want the interactive or consolidated mode, because consolidated runs take longer without a stop. Read the critique first; the protocol leads with what is wrong, missing or risky.
- Keep linked files in place so reviewers can read them
- Use the feature spec or implementation plan modes to narrow a run
- Expect the review to take time and to use significant model tokens
What people use it for
- Get an adversarial second read of a design doc before a team starts implementation.
- Check whether a feature spec is clear enough for a reviewer who has not seen the conversation behind it.
- Evaluate an implementation plan for readiness before writing code.
- Run a structured EGM Goldfish review on a design that a team already keeps in its repository.
Frequently asked questions
How do I install the egm-review skill?
Download the ZIP from this page and unzip it. Put the egm-review folder in ~/.claude/skills/ for personal use or in your project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What do I need to run egm-review?
You need an agent harness with a subagent tool, the EGM repository files including the persona files in agents/, and a target design document. The skill points to other docs and personas that are not included in the ZIP, so it will not run on its own.
Is egm-review free to use?
The skill is licensed under Apache-2.0, as listed on this page. Check the license file in the source repository for the full terms.
Does egm-review run scripts or use the internet?
According to the skill listing, it runs no scripts and uses no network. It works through the agent's subagent and file-writing tools.
What should I do if the skill stops with a message about subagents?
The skill stops when the harness has no subagent tool, because a review in the same context is not a Goldfish review. Use a harness that has a subagent tool such as Claude Code's Agent tool, Codex spawn_agent, or Antigravity invoke_subagent.
How does egm-review compare with a quick informal review?
A quick review is faster and uses fewer tokens. egm-review spawns multiple subagents and writes several reports, so it takes more time and resources, but each reviewer works in a fresh context and is told to stress-test the document.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use egm-review if your team follows the EGM Goldfish process, has the repository files, and wants a structured, adversarial review of a design before implementation. Pick something else for quick informal feedback, or if you do not have the EGM persona files or a harness with a subagent tool.
Download the skill