https://news.ycombinator.com/item?id=50157926
AI code review is not new. Claude can review GitHub pull requests, and Codex can be summoned with @codex review to inspect a diff and post a GitHub review. learn.chatgpt
But tdk-pr-review-buttons caught my attention because it fixes a much smaller—and much more annoying—problem: getting from “I’m looking at a PR” to “an AI agent is reviewing this PR” without copying links, switching apps, or explaining context again.
The idea in one sentence
It adds “Review this PR in Grok / Claude / Codex” buttons to a GitHub repository’s pull-request descriptions, using the same workflow and script as tdk-cli-core.
That’s it. And that’s why it’s brilliant.
Why this UX is so good
Most AI-review setups make the human do the integration work:
- Open the PR.
- Copy the URL.
- Open Claude, Codex, Grok, or another agent.
- Paste the link.
- Explain what kind of review you want.
- Wait.
This skill collapses that into:
- Open the PR.
- Click a button.
The review entry point lives exactly where the reviewer already is. That is the same principle behind good developer-tool design: don’t make people leave the workflow to trigger the next step.
What it actually is
tdk-pr-review-buttons is an agent skill in the tdk-landscape/tdk-skills repository. The repo describes itself as the source of truth for reusable CLI skills and baseline repository rules, with project-specific guidance kept in each repository’s AGENTS.md. github
The skill is listed on skills.sh as:
Add “Review this PR in” Grok, Claude and Codex buttons to a GitHub repository’s pull request descriptions, using the same workflow and script as
tdk-cli-core.
So it is not trying to replace your review process. It is a launcher: it makes AI review immediately accessible from the PR itself.
Why “small” features often win
The best developer tools rarely win because they do everything. They win because they remove one repeated moment of friction.
This feature removes:
- Context switching between GitHub and AI tools
- Manual link copying
- Repeating review instructions
- “Which AI tool should I use for this?” hesitation
- The gap between seeing a PR and starting a review
It also makes AI review more discoverable. If the button is right there, contributors are more likely to use it. If it lives in documentation or a separate dashboard, many people will forget it exists.
How it fits into modern AI review
This is not the only way to bring AI into pull requests. Claude Code offers GitHub-based code review that comments inline on changes, while Codex supports requesting a review by mentioning @codex review in a PR comment. learn.chatgpt
The difference is interaction design:
| Approach | Reviewer action | Strength |
|---|---|---|
Comment command, e.g. @codex review
|
Type a command in the PR | Simple, native to GitHub comments learn.chatgpt |
| Automated CI review | Wait for the pipeline | Consistent, hands-off review |
| PR review buttons | Click the model you want | Fast, visible, reviewer-controlled |
Buttons do not beat automation—they complement it. Automation is ideal for every PR; buttons are ideal when a human wants to deliberately send a specific PR to a specific reviewer agent.
Why multi-model choice matters
The skill includes buttons for Grok, Claude, and Codex, which is smarter than locking teams into one model.
Different models and agents can have different strengths:
- One may be better at explaining architectural trade-offs.
- Another may be stronger at spotting security or correctness issues.
- Another may be faster or cheaper for a first-pass review.
Giving reviewers a choice turns AI review into a practical workflow decision rather than a vendor mandate.
How I’d evaluate it before adopting it
Before installing any third-party agent skill, I’d check four things:
- What exactly gets sent to the AI provider? PR title, description, diff, repository context, secrets, internal URLs?
- Which permissions does it need? Read-only repo access is very different from write access to PRs or comments.
- Where does the review output go? Inline GitHub comments, a new comment, a local terminal, or somewhere else?
- Does it work with your team’s security rules? Especially for private repositories, customer data, credentials, or regulated codebases.
The broader ecosystem is already moving this way: AI agents can review PRs, follow repository guidance, and post structured GitHub reviews. The value of this skill is not inventing that capability—it is making it effortless to start. learn.chatgpt
Final thought
This is the kind of feature that looks almost too simple to be exciting:
Add buttons to a PR.
But simple features that remove repeated friction are often the ones teams actually adopt.
AI review becomes useful when it is available.
It becomes powerful when it is one click away.
And that is exactly what tdk-pr-review-buttons gets right.
Top comments (9)
Why multi-model choice matters
The skill includes buttons for Grok, Claude, and Codex, which is smarter than locking teams into one model.
Different models and agents can have different strengths:
One may be better at explaining architectural trade-offs.
Another may be stronger at spotting security or correctness issues.
Another may be faster or cheaper for a first-pass review.
Giving reviewers a choice turns AI review into a practical workflow decision rather than a vendor mandate.
Final thought
This is the kind of feature that looks almost too simple to be exciting:
Add buttons to a PR.
But simple features that remove repeated friction are often the ones teams actually adopt.
AI review becomes useful when it is available.
It becomes powerful when it is one click away.
And that is exactly what tdk-pr-review-buttons gets right.
cool!
nice!
cool!
wow!
amazing!
cool!
nice!