AI Diff Reviewer
One review methodology, two surfaces — a GitHub Action for CI and a coding-agent skill for your local machine.
AI Diff Reviewer is one review methodology with two surfaces. In CI it runs as a composite GitHub Action — inline comments, suggestion blocks, severity-based gating, and a clean PR timeline. Locally it installs as a coding-agent skill so you can review the current branch before you push. Pin `@v2` on both sides and the default prompt stays in sync. Formerly published as AI PR Reviewer; the old repository name redirects here.
2
surfaces — Action + agent skill
0
infra required — pure GitHub Action
4
LLM providers (Anthropic, Claude Code, Cursor, Codex)
MIT
licensed, end to end
Why it exists
The bits that make it worth adopting.
GitHub Action for every PR
Add one `uses:` line and every pull request gets a structured review — inline comments, native suggestion blocks, severity gating (critical / warning / info), and automatic collapse of previous bot reviews. Listed on the GitHub Marketplace as AI Diff Reviewer.
Coding-agent skill for local review
Install the skill into Claude Code, Cursor, Codex, and friends. Ask “Review my current branch” before you open the PR. Same severity model as CI when you pin the same tag.
One extension file, both surfaces
House rules live in `.review/extension.md` (legacy paths still accepted). CI and the local skill read the same file so team standards do not drift between laptop and Actions.
Severity-based gating
Every finding is `critical`, `warning`, or `info`. Configure which severities gate the check. Ship faster on info-only reviews; block on critical.
Provider choice you control
Anthropic (default), Claude Code, Cursor, and Codex. Bring your own API key — the Action never proxies through Dailybot.
MIT + fully forkable
Fork it, wire a private model, extend the severity contract — the code and licence say yes.
Install
Get started in seconds.
Pick the channel that matches your stack. Every path lands you in the same working state.
name: PR review
on:
pull_request:
branches: [main]
types: [opened, synchronize, labeled]
concurrency:
group: pr-review-${{ github.event.pull_request.number }}
cancel-in-progress: true
jobs:
review:
runs-on: ubuntu-latest
timeout-minutes: 15
permissions:
contents: read
pull-requests: write
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 0
- uses: DailybotHQ/ai-diff-reviewer@v2
with:
api-key: ${{ secrets.ANTHROPIC_API_KEY }}
github-token: ${{ secrets.GITHUB_TOKEN }}
provider: anthropic
model: claude-sonnet-4-6
strictness: block-on-critical
prompt-extension-file: .review/extension.md
applied-label: pr-reviewed
skip-review-label: skip-ai-review
pr-description-mode: autocomplete
complexity-labels-enabled: truenpx skills add DailybotHQ/ai-diff-reviewer --skill ai-diff-reviewerhttps://github.com/marketplace/actions/ai-diff-reviewerIn practice
What teams actually do with it.
01
Fast catch-up on medium-sized PRs
Humans still own architecture and taste. The reviewer clears the class of issues that do not need a human eye — obvious typos, missing null checks, style drift — before anyone opens the tab.
02
Review before you push
Run the skill locally, fix what matters, then open the PR. CI confirms with the same methodology instead of surprising you in the morning.
03
Standards without a Slack fight
Configure severity thresholds and let the check gate the merge. Teams stop arguing about style in chat — the check either fails on critical, or it does not.
04
New-contributor friendliness
First-time contributors get instant, actionable feedback. Nice tone, concrete suggestions, no waiting hours for a maintainer.
At a glance
The short list.
One methodology for CI and your coding agent
GitHub Marketplace Action + `npx skills add` skill
Severity gating with configurable strictness
Anthropic, Claude Code, Cursor, and Codex
FAQ
The questions we hear the most.
Is this the same as AI PR Reviewer?
Yes — renamed. The GitHub repository `DailybotHQ/ai-pr-reviewer` redirects to `DailybotHQ/ai-diff-reviewer`. Use `@v2` and the new name going forward.
Which LLM providers are supported?
Anthropic (default), Claude Code, Cursor, and Codex. Configure via the Action inputs / skill setup. You bring your own key.
Where does the code get sent?
To whichever LLM provider you configure — the Action never proxies through Dailybot. Your API key, your data, your choice of provider.
Does it replace human review?
No. It handles the class of feedback that does not need a human eye so humans focus on architecture, correctness, and taste.
What it is
AI Diff Reviewer is an LLM-driven code reviewer with two surfaces: a GitHub Marketplace Action for pull-request CI, and a coding-agent skill for local “review my branch” passes. Pin the same tag (@v2) on both and you keep one methodology — severity model, default prompt, and house rules via .review/extension.md.
Why open source
Because “which reviewer is looking at my code?” should have an answer you can inspect. MIT-licensed, developed in the open, no black-box behavior. Fork it, wire it to a private model, extend the severity contract — the code and licence say yes.
How it fits the ecosystem
Dailybot uses this reviewer on its own PRs (including ones AI agents open). It pairs with the Dailybot Agent Skill and Deep Work Plan — local review before push, CI review on the PR, same standards in both places.
Ready to try it?
Open source, MIT-licensed, and shipping in production at Dailybot every day. Fork it, wire it in, contribute back.
Other Dailybot open source
Dailybot CLI
The full platform CLI for Dailybot — current release 3.7.3 on PyPI.
Dailybot Agent Skill
Fourteen capabilities that turn your AI coding agent into a real Dailybot teammate.
Deep Work Plan
The gated-plan methodology any AI coding agent can execute.
Deep Work Plan website
The open-source documentation hub for the Deep Work Plan methodology.