How to Use kimi-cli for Code Review: Complete Workflow Guide

Use the /flow:code-review slash command in the interactive shell to trigger an automated review that collects changed files, renders a diff preview, queries the LLM, and displays a structured report.

MoonshotAI's kimi-cli is a Python-based interactive AI assistant that ships with a built-in code review flow. This feature orchestrates multiple sub-agents to analyze repository changes and generate actionable feedback, accessible via both the terminal UI and programmatic Python APIs.

How the Code Review Flow Works

The architecture defined in src/kimi_cli/soul/kimisoul.py treats code review as a first-class flow that loads its behavior from src/kimi_cli/agents/code-review.yaml. When you invoke /flow:code-review, the system executes four phases:

1. File Collection – Utilities in src/kimi_cli/utils/file_utils.py walk the workspace to identify changed or staged files based on Git status.
2. Diff Rendering – The tool generates a compact visual diff using src/kimi_cli/utils/rich/diff_render.py for display in the approval panel.
3. LLM Evaluation – The diff and system prompts are transmitted via src/kimi_cli/llm.py to the configured model, which returns a structured critique.
4. Report Visualization – Results are rendered in src/kimi_cli/ui/shell/visualize/_approval_panel.py and cached in the session directory (~/.kimi/).

Running Interactive Code Reviews

Start by installing the CLI and launching the interactive shell:

# Requires Python 3.12+

uv pip install kimi-cli

# Launch the TUI

kimi

Inside the shell, trigger the review workflow:

/flow:code-review

This command respects settings in src/kimi_cli/config.py, including the LLM model and max token limits. To export the generated review to a markdown file:

/skill:code-review --output review.md

Automating Reviews in Python Scripts

For CI/CD integration or batch processing, use the KimiCLI class exposed in src/kimi_cli/app.py:

import asyncio
from kimi_cli.app import KimiCLI

async def main() -> None:
    cli = await KimiCLI.create()
    # Target specific paths or use "." for the entire repo

    await cli.run_flow("code-review", {"path": "."})

if __name__ == "__main__":
    asyncio.run(main())

The run_flow method accepts a dictionary of parameters that map to the same file-collection logic used in interactive mode, allowing you to script reviews for specific modules or pull requests.

Core Implementation Files

Understanding these source files enables advanced customization:

• src/kimi_cli/cli/__init__.py – Parses command-line arguments and bootstraps the runtime.
• src/kimi_cli/soul/kimisoul.py – Dispatches the /flow:code-review command to the appropriate skill handler.
• src/kimi_cli/agents/code-review.yaml – Defines the sub-agent prompts and review criteria.
• src/kimi_cli/utils/file_utils.py – Handles workspace scanning and Git status detection.
• src/kimi_cli/utils/rich/diff_render.py – Generates the syntax-highlighted diff shown in the approval interface.
• src/kimi_cli/llm.py – Manages API communication with the underlying language model.
• src/kimi_cli/ui/shell/visualize/_approval_panel.py – Renders the interactive diff preview and final report.
• src/kimi_cli/config.py – Centralizes configuration for models, tokens, and the session directory.

Summary

• Invoke /flow:code-review in the interactive shell to start an automated review of your current workspace.
• The flow leverages src/kimi_cli/utils/file_utils.py and src/kimi_cli/llm.py to collect changes and generate AI feedback.
• Programmatic control is available via KimiCLI.run_flow() for automation pipelines.
• Configuration is governed by src/kimi_cli/config.py and persists sessions to ~/.kimi/.
• Export results to markdown using /skill:code-review --output <file>.

Frequently Asked Questions

How do I install kimi-cli?

Install the package using uv pip install kimi-cli or pip install kimi-cli. The tool requires Python 3.12 or higher and registers the kimi entry point for interactive use.

Can I review specific files instead of the entire repository?

Yes. When using the Python API, pass a specific path to run_flow: await cli.run_flow("code-review", {"path": "src/module.py"}). In interactive mode, the flow pre-selects changed files but allows you to adjust the selection in the approval panel before sending to the LLM.

Where are review outputs stored?

Sessions are cached in ~/.kimi/ as managed by src/kimi_cli/config.py. To persist a review to your working directory, use /skill:code-review --output review.md after the flow completes.

Which LLM model does kimi-cli use for code reviews?

The model is defined in src/kimi_cli/config.py and can be overridden via environment variables. The src/kimi_cli/llm.py module handles all inference, allowing you to specify different models for varying levels of review strictness.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}