OpenShrike

Code generation is cheap. Code review is the new bottleneck.

AI writes code faster than humans can review it

LLMs generate pull requests in minutes. Reviewing them thoroughly still takes hours. The bottleneck has shifted from writing code to understanding and verifying it.

Teams drown in review or switch to YOLO mode

Developers either spend their entire day reviewing AI-generated code, or they stop reading it altogether. Neither option is sustainable.

Existing tools don't solve the real problem

Linters catch syntax. Tests check behavior. But nobody is systematically verifying that AI-generated code follows your team's architectural decisions, security practices, and engineering standards.

A third option: predictable, automated code review

OpenShrike gives you semi-deterministic checks and an automated code review process. Instead of reviewing everything line by line, it brings your focus to the aspects that actually need human attention.

Requirements as Markdown

Architectural requirements live in your repo as Markdown documents. They are versioned, reviewable, and evolve with your project — no external dashboards or config languages.

Automated enforcement

OpenShrike materializes your requirements into checks and enforces them automatically on every PR. You know exactly what is being verified every time.

Focus on what matters

Not every line needs a human eye. OpenShrike highlights the parts that actually need review, so developers spend time where it counts.

Everything you need to enforce engineering standards

Markdown-driven checks

Materialize higher-level requirements into Markdown documents and enforce them automatically during code review. No proprietary DSL — just files in your repo.

Best practice library

Use the provided best practice library to bootstrap your project's review checks. Start with proven defaults, then customize to fit your team.

Beyond linters and tests

Enforce code best practices that are impossible to catch with linters, unit tests, integration tests, or any other kind of traditional testing.

Full control over the process

Easily relax, harden, or introduce new requirements just by editing a Markdown file — or asking your coding agent to edit it for you.

Auto-fix with agents

Automatically spawn an agent to fix failing checks and ensure the PR meets your encoded best practices before it reaches a human reviewer.

BYOK — Bring Your Own Key

Use the LLM and provider of your choice: OpenAI, Anthropic, LMStudio, Ollama, Zen, Bedrock, Azure, and more. OpenShrike uses OpenCode as its agent harness.

How OpenShrike compares

	/review skill	CodeRabbit	OpenShrike
Predictable checks	✗	✗	✓
Open source	Varies	✗	✓
Runs locally / CI	✓	✗	✓
Bring your own key	Sometimes	✗	✓
You control review process	✗	✗	✓

/review skill

Limited, non-deterministic code review — highly dependent on the model chosen. You don't really know what is actually checked or verified during the review.

CodeRabbit

Cloud-based (your code is sent to their servers), closed source, expensive, and you cannot use a model of your choice or a local LLM.

Runs locally or in CI, open source, you control the review process, and you can use any LLM provider — including local models.

Get started in minutes

Prerequisite: Node.js 22+

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/Network-Perspective/OpenShrike/main/install | bash

Windows PowerShell

irm https://raw.githubusercontent.com/Network-Perspective/OpenShrike/main/install.ps1 | iex

Setup your project

Interactive setup, configures your AI provider, lets you choose default policies:
```
shrike init
```
Runs the checks:
```
shrike scan
```
Scan and fix failing checks:
```
shrike fix
```

See it in action

A recorded scan run, with two violations autofixed with a single command.