FAQ
Getting started
Q1. What is prr-kit and how is it different from tools like CodeRabbit?
prr-kit is a framework that turns your existing AI IDE (Claude Code, Cursor, Windsurf, etc.) into a structured code reviewer. It installs a set of Markdown/YAML workflow files into your project — your IDE's AI reads and executes them.
Unlike SaaS tools like CodeRabbit or Graphite, prr-kit:
- Runs entirely inside your IDE — no external server, no background process
- Requires no API key beyond what your IDE already uses
- Uses 5 specialized reviewers (General, Security, Performance, Architecture, Business) instead of a single generic pass
- Can read your project's own rules (
CLAUDE.md,CONTRIBUTING.md,.eslintrc) and apply them during review - Works offline on local branches — no PR required
Q2. Does prr-kit need an API key or a separate account?
No. prr-kit uses whatever AI is already running in your IDE session. If your IDE is connected to Claude, GPT-4, Gemini, or any other model — prr-kit uses that. No extra keys, no signup, no monthly fee for the framework itself.
Q3. Which IDEs are supported?
Any AI IDE that can read Markdown and YAML workflow files. Tested with: Antigravity, Auggie, Claude Code, Cline, Codex, Crush, Cursor, Gemini CLI, GitHub Copilot, iFlow, Kilo, Kiro, OpenCode, QwenCoder, Roo Cline, Rovo Dev, Trae, Windsurf.
Q4. How do I install prr-kit?
Run the interactive installer inside your project repo:
npx prr-kit install
It will prompt you for your name, language, output folder, and Git platform.
Everything is written automatically to _prr/prr/config.yaml.
Then open your IDE in that project and run /prr-master to start.
Q5. What's the difference between /prr-quick and /prr-master?
/prr-quick | /prr-master | |
|---|---|---|
| What it does | Full pipeline in one command: select PR → describe → collect context → 5 reviews → report | Full menu of all commands — run any step individually |
| Pauses | Once — to ask which PR/branch to review | At every step — you control the sequence |
| Best for | Everyday reviews — fastest path from PR to report | Focused reviews — run only Security, or only Architecture, etc. |
Review workflow
Q6. Can I review a local branch without opening a GitHub/GitLab PR?
Yes. When selecting a PR, you can type any branch name directly instead of choosing from the
platform PR list. prr-kit will diff it against the base branch using git diff
— no platform PR required.
In this mode, review findings are saved to _prr-output/reviews/ only.
Posting inline comments to the platform is not available without a real PR number.
Q7. What is Party Mode?
Party Mode ([PM]) runs all 5 reviewer agents in a single collaborative session —
General, Security, Performance, Architecture, and Business — back to back, sharing context.
Each reviewer sees the findings of the previous ones, so they can build on each other
instead of repeating the same observations.
Use it when you want a deep, comprehensive review in one go.
Q8. What happens if I skip the [DP] Describe PR step?
Context collection runs automatically after [DP]. If you skip it and go straight
to a review, the reviewer agents won't have the PR knowledge base — no ESLint rules, no
standards docs, no external context. Reviews will still run but will be less accurate.
Always run [DP] first, or use /prr-quick which handles this automatically.
Q9. How do I post review findings as inline comments on the PR?
Run [PC] Post Comments from the agent menu after generating a report.
You need platform_repo set in your config and the platform CLI installed:
| Platform | Required CLI | Auth command |
|---|---|---|
| GitHub | gh | gh auth login |
| GitLab | glab | glab auth login |
| Azure DevOps | az + DevOps extension | az login |
| Bitbucket | bb or curl | — |
Q10. Can I run only one reviewer — e.g. only Security Review?
Yes. From /prr-master, select [SR] Security Review directly.
You still need to have run [SP] and [DP] first so the agent
has the diff and context loaded.
Platforms & CLI
Q11. Why does prr-kit need gh / glab CLI?
The platform CLI is only needed for two things: listing open PRs and posting inline comments.
If you don't need either of those, prr-kit works entirely via git diff — no CLI required.
Set platform: none in your config to explicitly run in local-only mode.
Q12. Does prr-kit work with private repos?
Yes. prr-kit runs locally in your IDE — it never sends your code to a prr-kit server. It uses your IDE's AI (which you've already authorized) and your platform CLI (which uses your existing auth token). Private repos work exactly the same as public ones.
Q13. I use Azure DevOps. Is anything different?
Set platform: azure and platform_repo: org/project/repo
(note the three-part format for Azure, vs two-part for GitHub/GitLab).
The az CLI with the Azure DevOps extension is required for PR listing and comment posting.
platform: azure
platform_repo: "my-org/my-project/my-repo"Stack support
Q14. What languages and frameworks does prr-kit understand?
prr-kit ships with built-in rules for 47 stacks. A few examples:
| Category | Stacks |
|---|---|
| Frontend | Vue 3, React, Angular, Svelte, Next.js, Nuxt.js, Astro, SolidJS, HTMX |
| Styling | TypeScript, Tailwind CSS, CSS |
| Backend — Node | NestJS, Express |
| Backend — Python | FastAPI, Django, Flask |
| Backend — JVM | Spring Boot, Java, Kotlin |
| Backend — Systems | C#, C++, Rust, Go |
| Database / ORM | SQL, Prisma, TypeORM, MongoDB, Redis |
| Infrastructure | Docker, Kubernetes, Terraform, GitHub Actions |
| Mobile | React Native, Flutter, Swift |
| Testing | Jest/Vitest, Playwright |
Each stack file contains rules across Security, Performance, Architecture, Code Quality, and Common Bugs — with severity labels [CRITICAL] / [HIGH] / [MEDIUM] / [LOW].
Q15. How does prr-kit know which stack my project uses?
Automatically — no configuration needed. During context collection, the agent analyzes the
changed files in the PR: file extensions, import statements, and config files (e.g.
package.json, pyproject.toml) are used to detect the active stacks.
The relevant rule files are then loaded and injected into the review knowledge base.
Q16. My stack isn't in the list. Can I add custom rules?
Yes. Add a @context: or @rule: annotation anywhere in your source code —
the agent extracts it during context collection and applies it during review:
// @rule: Always use the Result type for error handling in this module
// @pattern: Use repository pattern for all database access
For project-wide rules, add them to CLAUDE.md, AGENTS.md,
or CONTRIBUTING.md — the agent reads these automatically.
Maintenance
Q17. How do I update prr-kit to the latest version?
Run the installer again inside your project — it updates the installed files in _prr/:
npx prr-kit@latest installYour config.yaml is preserved — the installer only updates workflow and agent files.
Q18. How do I reset review state and start fresh?
Run [CL] Clear from the agent menu. It removes context files and/or review reports
from your review_output folder. You can choose to clear just the context,
just the reports, or both.
Alternatively, delete the files manually from _prr-output/reviews/.
Q19. Should I commit the _prr/ folder to Git?
Yes — commit _prr/ so the whole team shares the same workflow files and config.
Add _prr-output/ to .gitignore to keep generated review reports
out of version control.
# .gitignore
_prr-output/Q20. Can multiple developers on the same team use prr-kit with different languages?
Yes. Each developer has their own config.yaml with their own communication_language.
Since _prr/prr/config.yaml is committed to Git, you can either: set a team default
language in the committed config, or have each developer maintain a local override.
Review output files are gitignored, so each developer's reports stay local.