---
title: "Write a dismissal reply that someone else can check | Proofline"
description: "A reply that dismisses an AI review comment should let a later reader confirm the dismissal without redoing your work."
url: https://proofline.sh/learn/reply-to-dismiss-ai-review-comment/
---

[Learning center](/learn/)

# Write a dismissal reply that someone else can check

A reply that dismisses an AI review comment should let a later reader confirm the dismissal without redoing your work. Name the claim you reject, give the evidence against it, say what you checked and what you did not, and only then resolve the thread.

Code review The Proofline team November 22, 2026 5 min read

## Who reads the reply

The bot does not need your reply. Three people do: the maintainer deciding whether to approve, the next author who touches the same code and wonders why it looks this way, and anyone who later audits the review, for example to [score a dismissed finding](/learn/reconstruct-dismissed-review-finding/) in a reviewer trial. Each of them arrives after the context is gone. A resolved thread collapses, so for most readers the reply is the only part of the exchange they will see.

That sets the bar. "Bot is wrong" asks the reader to trust you. A reply with a file, a commit, a test or a reference lets them check you in a minute.

## Four parts of a checkable reply

1.  The claim, in one clause. Restate what the bot said that you reject, since its comment may be long, hidden or partly right.
2.  The evidence. Point to something the reader can open: a line at a commit, a test, documentation for the language or library, or a project rule.
3.  What you checked, and what you did not. "The test passes" is a fact about one run. Say which part of the claim it covers.
4.  What happens next. Resolve the thread, leave it open for a maintainer, or link an issue.

Not every reply needs all four at length. A two-line reply can carry them, as the second example below shows.

## A reply with all four parts

In September 2026, Copilot reviewed a Home Assistant pull request that added entity links to the KNX integration. Its summary said: "Validation errors can escape their handler, and attribute-only state updates generate redundant KNX telegrams." [Copilot's review](https://github.com/home-assistant/core/pull/181281#pullrequestreview-5110681324)

The author answered under the heading "One Copilot comment is wrong":

> The claim that equal status and command addresses escape the handler does not hold: `homeassistant/__init__.py` calls `install_as_voluptuous()`, so in a running HA process `vol.Invalid is probatio.Invalid` is `True`. The self-loop takes the handled path and returns `success: False`, and there is a test for it. The two classes only differ if `voluptuous` is imported before `homeassistant`. Flagging it so it does not get "fixed" into a translation layer. The Probatio style point stands either way, and is done.

[Author's reply](https://github.com/home-assistant/core/pull/181281#issuecomment-5620427637)

Every part is there. The claim is narrowed to one case, equal status and command addresses. The evidence is a named file and a named test. The reply states the condition under which the bot would be right, when `voluptuous` is imported first, which is the limit of the check. It separates the rejected point from an accepted one. It also names the harm of not answering: someone could "fix" the code into an unnecessary translation layer.

The reader can verify it. Home Assistant's `homeassistant/__init__.py` on the `dev` branch calls `install_as_voluptuous()` before anything else imports `voluptuous`, with a comment saying it must run first. [homeassistant/\_\_init\_\_.py](https://github.com/home-assistant/core/blob/dev/homeassistant/__init__.py) The pull request's diff contains `test_switch_link_rejects_self_loop`, which sends equal addresses and asserts `success` is `False`. [Pull request diff](https://github.com/home-assistant/core/pull/181281/files) We read both and did not run the test. The pull request was later closed as stale without merging, so the reply never had to stand up in a merged change, but it settled the question for the review.

## A short reply that names its source

In May 2026, Copilot commented on this line in a Home Assistant dependency bump: `except GTIError, ClientConnectorError:`. The comment's text is hidden in the resolved thread. The author's reply shows its subject: "Had to fight with this, because when I added parens ruff removed them. Seems copilot is wrong, because of [https://peps.python.org/pep-0758/](https://peps.python.org/pep-0758/)" [Author's reply](https://github.com/home-assistant/core/pull/172613#discussion_r3328694206)

The reply is short and still checkable, because it names its source. [PEP 758](https://peps.python.org/pep-0758/) allows unparenthesized `except` expressions "only when not using the `as` clause," starting with Python 3.14. We ran a two-exception `except A, B:` block. Python 3.14.6 caught the exception, and Python 3.13 rejected the file with `SyntaxError: multiple exception types must be parenthesized`. Home Assistant's `pyproject.toml` on `dev` requires Python 3.14.2 or later.

Two additions would have saved a reader those checks: the Python version the project requires, and the fact that the formatter enforces the form. For example:

> Valid since Python 3.14 (PEP 758, no `as` clause), and the project requires 3.14.2. Ruff strips the parentheses when I add them, so they would not survive formatting. Resolving.

The maintainer merged the pull request on 2 June 2026 with the unparenthesized form. Copilot's review of the last commit attached new threads to the same `except` lines, whose text is also hidden. A reply in the first thread does not stop a bot from raising the point again. It gives you a link to answer the repeat with.

## When to resolve the thread

GitHub lets you resolve a conversation "if you opened the pull request or if you have write access to the repository." When you do, "the entire conversation will collapse and be marked as resolved." [Resolving conversations](https://docs.github.com/en/pull-requests/how-tos/review-pull-requests/commenting-on-a-pull-request#resolving-conversations) A repository can also require conversation resolution, which "requires all comments on the pull request to be resolved before it can be merged to a protected branch." [Protected branches](https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/managing-protected-branches/about-protected-branches#require-conversation-resolution-before-merging) Under that rule, resolving is what lets the pull request merge, so the reply has to come first.

| Situation | Resolve? |
| --- | --- |
| Your evidence settles the claim, and you own the change | Yes, after the reply |
| A maintainer has to accept the reasoning | No. Leave it open and ask them in the thread |
| The claim is real but out of scope | Yes, after linking a follow-up issue |
| The claim was right about an earlier commit | Yes, after naming the commit that changed it, as in [the outdated comment guide](/learn/ai-review-outdated-comment-after-push/) |
| The bot repeated a claim you already answered | Yes, with a link to the earlier reply |

Resolving without a reply leaves a collapsed thread with the bot's claim and nothing against it. A later reader has to treat that comment as unanswered.

## One month of free code review.

New organizations get one calendar month of hosted code review for private and public GitHub repositories. No card or cloud connection required. [Eligibility and fair-use terms](/free-code-review/).

[Start free with GitHub](https://proofline.sh/orgs/plan?offer=free_month)

## Related reading

-   ### [Read a dismissed code review finding before you score it](/learn/reconstruct-dismissed-review-finding/)
    
    Code review The Proofline team November 20, 2026 4 min read
    
-   ### [Does an outdated review comment still apply to your branch?](/learn/ai-review-outdated-comment-after-push/)
    
    Code review The Proofline team November 22, 2026 4 min read
