EGM Resolve (Goldfish Review: Resolve)
Triages findings from parallel Goldfish reviewers into contest/escalate/fix/decline, then batch-applies only the edits you approve.
What it does
This is the "resolve" stage of Google's Elephant–Goldfish review pipeline. Given a target file, a line range and an aggregated findings list, it:
- Labels each finding exactly once: CONTEST (must cite evidence the reviewer could actually read), ESCALATE (evidence lives outside the repo — hand it to the human), PROPOSE FIX (valid; define the precise replacement), DECLINE (nit, recorded only)
- Writes a structured
goldfish_review_report.mdordered by severity - Stops at an explicit approval gate (auto accept / continue / interactive breakdown / custom instruction) and calls zero tools while waiting
- Applies only approved edits in one batch, re-validates the markdown, and hands counts back to the caller (e.g.
egm-review-loop)
Anti-bias rules are baked in: "it was intentional" is not evidence, comprehension gaps may not be contested, and severities can't be silently downgraded.
Who it's for
- Teams already running the EGM / Goldfish multi-agent review loop
- Anyone who needs a disciplined way to process review feedback on design docs, RFCs or specs
- Users who insist on a human gate before an agent edits files
Examples
- Run four Goldfish reviewers over
docs/design.md, feed the 20 findings here, then reply "auto accept" to batch-apply every valid fix. - Reply "interactive breakdown" to walk Blockers one by one and "Overrule contest" when the agent's pushback is weak.
- A reviewer misreads an unwritten team convention — the skill marks it ESCALATE and asks whether you want that fact added to the document.
· · · Install guide · · ·
Try it now, no install
Paste this into Claude to use the skill without installing anything.
Read the instructions in this file and follow them to help me: https://raw.githubusercontent.com/google/elephant-goldfish-model/HEAD/distributions/claude/skills/egm-resolve/SKILL.md What I want: (describe your task here)
If Claude can't open the link, open it yourself and paste the contents instead.
↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.
Install in the Claude app (no terminal)
- Download the ZIP with the button below.
- In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
- Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
Install in Claude Code
Let Claude do it — paste this into Claude Code
Install the skill I found on AI Skill Mart. Copy the distributions/claude/skills/egm-resolve folder from the GitHub repo google/elephant-goldfish-model into my ~/.claude/skills/egm-resolve/. When it's done, tell me in one line what this skill can do.
Install with a command instead
git clone https://github.com/google/elephant-goldfish-model.git && mkdir -p ~/.claude/skills && cp -r elephant-goldfish-model/distributions/claude/skills/egm-resolve ~/.claude/skills/⚠ This is a third-party skill. Check the source repository before installing.
- Install Claude Code and open a terminal.
- Clone the repo:
git clone https://github.com/google/elephant-goldfish-model.git - Create the skills folder if needed:
mkdir -p ~/.claude/skills - Copy the skill:
cp -r elephant-goldfish-model/distributions/claude/skills/egm-resolve ~/.claude/skills/ - Also copy the sibling EGM skills (such as
egm-review-loop) and the plugindocs/folder, since this skill referencesdocs/claude.mdanddocs/codex.md. - Restart Claude Code and invoke the skill with
TargetFile,LineRangeandFindingsList; if you omit them it will onboard you one question at a time.