Guide · Agent skill
How to use the Goldfish Review Resolver agent skill (egm-resolve)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The egm-resolve skill is a set of instructions that teaches an AI agent to work through a list of problems flagged by reviewers of a design or technical document, using the Goldfish review format from the Elephant Goldfish model project. This guide explains what the skill does, how to install it from a ZIP file, how to trigger it with plain requests, and what to check before letting it edit your file. You will be able to set it up in Claude Code or the Claude apps and run a first-pass triage on your own findings.
Goldfish Review Resolvergoogle/elephant-goldfish-model · Apache-2.0 Download the skill Quick start
- 1Download the skill ZIP. Get the egm-resolve ZIP from the directory page. It contains the skill folder with SKILL.md and agents/openai.yaml.
- 2Install the skill. Unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ in your project. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
- 3Gather your inputs. Have the target document path, the line range or full file, and the Goldfish-format findings list ready.
- 4Ask the agent to resolve the findings. Describe the task in plain words and provide the three inputs, or let the agent prompt you for each one.
- 5Read the report. Open goldfish_review_report.md in the working directory and check each category and proposed edit.
- 6Choose how to proceed. Reply with auto accept, continue, interactive breakdown, or a custom instruction. Nothing is edited until you accept.
What is an agent skill?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a task matches them. The skill does not add new software to your computer; it changes how the agent approaches a specific kind of task.
The egm-resolve skill is licensed under Apache-2.0 and is distributed as part of the Goldfish review tooling in the google/elephant-goldfish-model repository.
What does the egm-resolve skill make the agent do?
The skill takes a list of findings from Goldfish reviewers and sorts each one into exactly one category: CONTEST (the critique is inaccurate and you can cite evidence a reviewer could read), ESCALATE (the critique is inaccurate but the evidence lives outside the files, such as a team convention or a verbal decision), PROPOSE FIX (the critique is valid and the agent drafts a precise text replacement), or DECLINE (the finding is a Nit and is recorded but not fixed).
The agent then writes a markdown report named goldfish_review_report.md in the working directory. It pauses and waits for your choice. It edits the target document only after you accept fixes.
- Reads the target file at the line range you give it and parses the findings list.
- Writes a prioritized report with an AI suggestion, justification and proposed edit for each finding.
- Lists ESCALATE items first, because they need your knowledge.
- Does not change a finding's severity; if it disagrees, it contests and explains why.
How do you install the skill?
Download the ZIP of the skill from the directory page. The skill folder is distributions/claude/skills/egm-resolve and includes agents/openai.yaml and SKILL.md.
For Claude Code, unzip the folder into ~/.claude/skills/ to make it available for all your projects, or into a project's .claude/skills/ folder to make it available only in that project.
For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your repository
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
What do you need before you start?
The skill needs three things. First, an agent runtime that loads SKILL.md skills, such as Claude Code, the Claude apps, or a compatible agent. Second, a Goldfish-format findings list from reviewers; without it the skill is not useful. Third, access to the Goldfish docs the skill refers to, docs/claude.md or docs/codex.md, in the repository.
The skill is designed for a document you wrote in a persistent conversation, because the agent uses earlier reasoning from that session to write better fixes.
- An agent runtime that loads SKILL.md files
- A Goldfish-format FindingsList from reviewers
- The Goldfish docs (docs/claude.md or docs/codex.md) in the repository
How do you trigger the skill with a natural request?
Ask for the job in plain words and give the agent the three inputs: the target file, the line range (or full file), and the findings list. If you leave any of them out, the skill starts a guided setup and asks for one piece of input at a time.
After the report is written, the agent stops and waits. Reply with one of the choices it offers. Typical replies include "auto accept" to apply all proposed fixes, "continue" or "no changes" to apply nothing, "interactive breakdown" for an item-by-item walkthrough, or any custom instruction.
- Example request: "Resolve the Goldfish findings in design.md, lines 1-200, using the findings list I paste below."
- Example reply: "interactive breakdown"
What happens during the interactive breakdown?
In interactive breakdown, the agent goes through every item except DECLINE, one at a time. For each one it shows the scope, description, its suggestion, the explanation and the proposed edit, then asks for your decision. The options depend on the category: Accept proposed fix, Skip / ignore or Discuss for PROPOSE FIX; Skip / agree with contest, Overrule contest (propose fix) or Discuss for CONTEST; Confirm (no change), Add the fact to the document or Discuss for ESCALATE.
The agent records your decisions and does not edit any file until every item is decided. Then it applies the accepted edits in one batch.
What should you check before accepting fixes?
Read goldfish_review_report.md before you accept anything. The skill itself warns that the agent may misjudge a finding and edit the file directly, so auto-accepting on an important document without reading the report is a poor choice.
The skill also bypasses CONTEST, ESCALATE and DECLINE items during auto accept, so only PROPOSE FIX items are applied in that mode. Check the ESCALATE items yourself, since they need facts only you may have.
- Read the report and the proposed edits first.
- Use interactive breakdown for important documents.
- Keep a copy of the document under version control so you can undo an edit.
What people use it for
- Triaging a list of reviewer findings on a design document before a team review meeting.
- Walking through contested and escalated items in interactive breakdown on a technical spec.
- Drafting precise text replacements for valid gaps flagged by a multi-reviewer Goldfish run.
- Producing a consistent, reviewable report of how each flagged item was handled.
Frequently asked questions
Is the egm-resolve skill free to use?
The skill is licensed under Apache-2.0, as stated on its directory page. Check the license file in the repository for the full terms.
What do I need to run the egm-resolve skill?
You need an agent runtime that loads SKILL.md skills, such as Claude Code or the Claude apps, plus a Goldfish-format findings list from reviewers. The skill also needs the Goldfish docs (docs/claude.md or docs/codex.md) in the repository.
Where do I put the egm-resolve skill in Claude Code?
Unzip the skill folder into ~/.claude/skills/ for all projects, or into .claude/skills/ inside one project. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
Does the egm-resolve skill run scripts or use the network?
According to the directory listing, it does not run scripts and does not use the network. It reads and writes files in your working directory and edits the target document only after you accept fixes.
Can I use egm-resolve for general proofreading?
No. The skill expects a specific Goldfish findings list format and is not meant for editing a document with no reviewer findings. Use a general editing tool or a different skill for that.
Will the skill change my document without asking?
No. It writes a report and waits for your reply. It edits the document only after you choose auto accept or accept specific items, and it bypasses CONTEST, ESCALATE and DECLINE items in auto accept.
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-resolve if you are an author with structured Goldfish review findings on a design or technical document, or a team running the Elephant/Goldfish multi-reviewer workflow and wanting consistent first-pass triage. Pick a different tool if you only need general proofreading or do not have a reviewer findings list, and do not auto-accept fixes on important documents without reading the report first.
Download the skill