Skip to content
Get packs

Goldfish Review Loop

Runs a multi-agent document review loop that audits docs and can apply approved fixes

Source Download skill
Repository
google/elephant-goldfish-model
Stars
9,965
Licence
Apache-2.0
Files
2 files

What it does

Once installed, the agent can take a design doc, README, feature spec or implementation plan and have several separate AI reviewer subagents audit it against written quality and structure standards. The reviewers run in fresh, isolated contexts so they catch unstated assumptions, and the main agent then evaluates their findings and can apply fixes it judges approved, repeating in rounds. The skill is an orchestration guide; the reviewer personas and standards it points to are separate files that were not included in the reviewed material.

Great for

  • Teams who want a structured second opinion on design docs or feature specifications before review meetings
  • Engineers preparing implementation plans who want checks for readiness and missing codebase details
  • Authors who want completeness and tone checks on READMEs

Skip it if

  • Quick one-off proofreading, since the multi-round subagent loop is slow and token-heavy
  • Situations where automatic edits to important documents are unwanted, because the loop can apply fixes to the target file after confirmation and the reviewed files are only partly visible here
  • Environments without a subagent tool, since the skill itself says a review in the same context is not a valid Goldfish review

Try asking

  • “Run the egm review loop on my feature spec at docs/search-spec.md using the feature-spec standard”
  • “Review the implementation plan in plans/migration.md with implementation readiness and codebase details reviewers”
  • “Audit the section S1.12 of this design doc for missing assumptions and then apply the approved fixes”
  • “Check this README for completeness and tone”
Step-by-step guide: how to use Goldfish Review LoopInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 2 original files, the Apache-2.0 licence and an attribution note.

    egm-review-loop.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip egm-review-loop.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip egm-review-loop.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Confirm the proposed reviewer and standard configuration in Step 1 before the loop starts, and check any fixes it applies to the target file.
  • Specify a precise SemanticTarget such as a section header so the line range is located correctly, or use full file deliberately.
  • Keep the original document under version control so applied edits can be reviewed and reverted.

What's inside

  • agents/openai.yaml
  • SKILL.md

Redistributed unmodified from google/elephant-goldfish-model under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-11-02. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.