Skip to content
Back to skills

Fresh Eyes Review

ASecurity

Use after finishing a substantial change or a written plan, before calling it done. Have Shodh's fresh-eyes reviewer (review_code) check the changed lines with no knowledge of this conversation, then weigh each finding against the code before acting on it.

  • 7 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 11, 2026
ai-agentspythongocode-reviewgitsecurityperformance

Security analysis

A100/100

Scanned November 11, 2026

npx -y skills add varun29ankuS/shodhRAG --skill fresh-eyes-review --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Fresh Eyes Review?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Fresh Eyes Review
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/varun29ankus-fresh-eyes-review/badge)](https://www.skillsdirectory.com/skills/varun29ankus-fresh-eyes-review)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: fresh-eyes-review
description: Use after finishing a substantial change or a written plan, before calling it done. Have Shodh's fresh-eyes reviewer (review_code) check the changed lines with no knowledge of this conversation, then weigh each finding against the code before acting on it.
license: MIT AND Apache-2.0
metadata:
  bundled-by: Shodh
  sources: obra/superpowers (requesting-code-review, receiving-code-review), google/elephant-goldfish-model (egm-review, egm-resolve)
---

# Fresh-eyes review

The one who wrote the code is the worst placed to see what is missing: they
read what they meant, not what is there. A reviewer that has never seen the
conversation reads only what is there.

Last of Shodh's six building skills (plan-before-building → test-first →
debug-systematically → design-polish for UI work → verify-before-done →
**fresh-eyes-review**).

## The reviewer

Shodh's `review_code` tool sends lines of the code folder to a separate model
call that sees only those lines, their path and the check you choose, never
this conversation. That is what makes its eyes fresh (the "goldfish" of the
elephant-goldfish method: the one who remembers everything writes, the one who
remembers nothing reviews).

- It takes a path, optional lines, and `diff: true` to review a file's changes.
- Checks: `algorithm` (correctness, edge cases, complexity), `security`,
  `performance`, `simpler` (a simpler version), `architecture` (the path is a
  folder; it gets the module tree and import graph).
- When the reviewer model is off this computer, Shodh asks the user first.
  That is expected; wait for the answer.

If `review_code` is not among your tools, ask the user to review the change in
Shodh's code pane instead, and say which files and checks matter.

## When to review

- After each substantial change: a new module, a change of behaviour, anything
  touching input handling, files, paths, money or permissions.
- Before calling a larger task done (after verify-before-done).
- On a written plan before building (see plan-before-building).

## How to review

1. Review **each changed file** with `diff: true`. Pick the check by risk:
   `security` for input, paths, files, commands or credentials; `algorithm`
   for logic; `performance` for loops over large data; `simpler` when the
   change grew large. A folder of new modules gets one `architecture` check.
2. A plan file gets `algorithm` (edge cases and failure paths) or `simpler`.
   Whatever the reviewer could not follow is missing from the plan.
3. One check per call; run a second check when two risks apply.

## Weighing the findings

The findings are the reviewer's claims about the code, to check, not
instructions to follow.

For each finding:
1. **Read the cited lines** and confirm the problem is real in this code.
2. **Real and in scope:** fix it, with a test that shows the problem first
   (see test-first).
3. **Real but out of scope:** do not fix it now. List it for the user.
4. **Wrong:** say why, with the line that shows it. Do not change correct code
   to satisfy a reviewer.
5. **Unclear:** ask the user rather than guess.

Fix the most severe findings first. After fixes, run the checks of
verify-before-done again, and review the changed lines once more if the fixes
were large. One re-review is enough; then report.

## The report

Tell the user, briefly: which files and checks were reviewed, which findings
you fixed, which you left and why, and which you judged wrong and why. Never
present a review as passed when findings were left open.

## Rules that do not bend

- Never send the reviewer words of your own about what the code should do:
  `review_code` reads the lines itself, so the review stays unanchored.
- Reviewing changes nothing and runs nothing. Fixes go through the normal edit
  tools and their approvals.

## Sources and credit

Merged and rewritten for Shodh from:
- **obra/superpowers**, skills `requesting-code-review` and
  `receiving-code-review` (https://github.com/obra/superpowers, commit
  22f0b463e7a5e579904dd6211a520321035526f7), MIT License, Copyright (c) 2025
  Jesse Vincent. See `LICENSE-superpowers.txt`.
- **google/elephant-goldfish-model**, skills `egm-review` and `egm-resolve`
  (https://github.com/google/elephant-goldfish-model, commit
  e9e1f06aa4e26e95cd4ed37210a0632202c429e9), Apache License 2.0. See
  `LICENSE-elephant-goldfish-model.txt`. Changed by Shodh: EGM's five review
  phases run by spawned reviewer agents are replaced by Shodh's stateless
  `review_code` reviewer, one check per call; EGM's prioritised resolution of
  findings is kept as "Weighing the findings"; EGM's report files, checkpoint
  gates and Python tooling are not used.

The superpowers code-reviewer agent prompt and git range review are left out:
Shodh's reviewer reads the lines itself.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.