hustcer/deepseek-review

What it solves

DeepSeek Code Review automates the process of analyzing code changes in GitHub Pull Requests (PRs) and local Git repositories. It uses DeepSeek AI models to identify potential bugs, logical errors, security vulnerabilities, and style violations, providing concise improvement suggestions directly within the developer's workflow.

How it works

The tool operates in two primary modes:

1. GitHub Action: It can be integrated as a workflow that triggers automatically when a PR is opened, reopened, or synchronized. It can also be triggered manually via specific labels (e.g., "ai review") or by mentioning a bot in a PR comment. The action then analyzes the diff content and posts the review results as a formal review on the PR.
2. Local CLI: Using Nushell, the tool allows developers to review local commit changes, remote GitHub PRs, or specific git diffs from the command line. It supports streaming output and can save results to Markdown files.

Both modes support customizable system and user prompts, configurable temperature settings, and the ability to to use DeepSeek's V4 and R1 models, as well as self-hosted models or GitHub Models.

Who it’s for

Developers and maintainers who want to automate code reviews to catch errors early and reduce the manual effort required for initial PR screening.

Highlights

• Flexible Triggering: Supports automatic PR triggers, label-based triggers, and interactive comment-based triggers.
• Local & Remote Review: Review code changes either directly in GitHub or locally via a CLI tool.
• Customizable AI: Full control over the model choice (V4, R1, self-hosted), base URLs, and prompts.
• Cross-Platform: Compatible with macOS, Ubuntu, and Windows.
• Granular Control: Ability to include or exclude specific files from the review process via patterns.