---
name: respond-to-code-review-prompt
description: A prompt that makes an AI coding agent handle review feedback properly - restate each comment, check it against the real code, ask about anything unclear before changing a line, push back with reasons when a suggestion is wrong, then fix one item at a time. Use when you paste reviewer comments into an AI, or the agent agrees with every comment without checking.
license: MIT
metadata:
  category: coding-prompts
  tags: "agents,quality"
  author: johnny-bravo
  version: "1.0"
  updated: "2026-11-02"
  status: published
  aliases: "receiving code review, review feedback, push back on review"
  source_url: "https://github.com/obra/superpowers/tree/2ce8aa15d05bd57213e6f746f953aafb7ded7a8b/skills/receiving-code-review"
  source_license: MIT
  copyright: "Copyright (c) 2025 Jesse Vincent"
  original_author: "Jesse Vincent (obra/superpowers, receiving-code-review)"
  modified: "Adapted by TheSkillVault: condensed the response pattern into a single paste-in prompt with an original reply format; dropped agent-harness specifics; added steps, adapt notes, secrets note and done-when."
  imported_on: "2026-11-02"
---

# Respond to code review (prompt)

Agents tend to say "Great point!" and change code they never checked. This prompt makes them verify first and answer like an engineer.

## When to use
- You got review comments on a pull request and want an AI to work through them.
- The agent keeps agreeing with every suggestion, even wrong ones.
- Some comments are vague and you want questions, not guesses.

## Steps
1. Copy the review comments, with file and line where possible.
2. Paste them into the prompt below in place of {{REVIEW COMMENTS}}.
3. Read the triage table before letting the agent change anything.
4. Answer its questions, then let it fix items one at a time with tests.

## Prompt
```text
You are responding to code review comments on my change. Do not change any code yet.

For each comment:
1. Restate what the reviewer is asking for, in one sentence.
2. Check it against the actual code in this repo. Quote the relevant lines.
3. Decide: AGREE (correct for this codebase), PUSH BACK (wrong or harmful here, give the technical reason), or UNCLEAR (say exactly what you need to know).
4. Note whether it could break other code, tests or platforms.

Rules:
- No praise or filler such as "great point" or "you're absolutely right".
- If any comment is UNCLEAR, stop after the table and ask your questions. Do not fix the clear ones first; comments are often related.
- Skip suggestions for features nobody uses (YAGNI) and say why.

Output a table: comment | restated ask | verdict | reason | risk.
After I confirm, fix AGREE items one at a time, run the tests after each, and give a one-line reply per comment I can post.

REVIEW COMMENTS:
{{REVIEW COMMENTS}}
```

## How to adapt
- Reviewer is your tech lead and you trust them? Change rule 3 to default to AGREE unless the code proves otherwise.
- For a single comment, drop the table and ask for the verdict and reason only.

## Gotchas
- Remove tokens, keys and private customer data from comments or diffs before pasting them into any AI tool.
- An agent that pushes back can still be wrong. Check its quoted lines yourself.

## Portability
Works with any AI coding agent (Claude Code, Cursor, Codex, Copilot) and any git host. No tool lock-in.

## Done when
- Every comment has a verdict with a reason.
- Unclear comments were asked about before any code changed.
- Fixes landed one at a time with tests passing.

## Example prompt
"Here are the five review comments on my PR. Triage them before you change anything."

---
Adapted from "receiving-code-review" in obra/superpowers (MIT, Copyright (c) 2025 Jesse Vincent). See LICENSE.txt.