Review Mode for VS Code

Review Mode is a powerful VS Code extension designed to bring enhanced collaboration and review tools directly into your editor for any text file.
It can be used in two primary ways:
- Standalone Review Tool: Review and annotate code, documentation, or any text file locally without needing an external platform.
- AI-Assisted Plan Review: Turn your editor into a collaborative planning surface where you and your AI agent iterate on design and implementation plans, powered by the Review Mode MCP server.
Features
- Inline Threaded Annotations: Add, reply to, and resolve comments directly within your text files. Keep all your feedback contextualized.
- Revision Tracking: Automatically track revisions of your documents and see how comments evolve over time.
- Visual Status Tracking: Quickly identify the status of comments using intuitive icons.
- Review History UI: A dedicated Review Notes panel in the Explorer allows you to easily navigate through all annotations.
- Customizable Priorities: Assign priorities to comments (e.g., Low, Medium, High) so you know what needs addressing first.
AI-Assisted Plan Review Workflow
AI writes plan → Opens in Review Mode → You annotate → AI reads feedback → Updates plan → Repeat
- The AI writes a plan. Instead of printing it in chat, the AI saves it as a Markdown file and opens it in Review Mode.
- You review and annotate. In the Review Mode panel, you can:
- Click the + button next to any line to add a comment
- Set comment priority (none, low, medium, high, urgent)
- Set comment status (open, in-progress, resolved, won't fix)
- Reply to existing comment threads
- You ask the AI to iterate. Tell the AI something like:
- "Refine the plan with my comments" — the AI will update the plan and reopen Review Mode for another round
- "Implement my comments and proceed" — the AI will finalize the plan and ask for confirmation before starting implementation
- The AI processes your annotations:
- Comments it can fully address → implemented and marked resolved
- Comments needing clarification → the AI replies in the thread (prefixed with
[AGENT]) and marks in-progress
- Comments it disagrees with → the AI explains why and marks won't fix
- Repeat until satisfied. When all comments are resolved and you're happy with the plan, tell the AI to proceed with implementation.
One-Click Install
The extension includes a unified Install Skills command that sets up everything you need — the MCP server, agent workflows, and editor-specific configuration — in a single step.
- Open the Command Palette (
Ctrl+Shift+P) and run Review Mode: Install Skills, or click the ⬇ icon in the Review Mode sidebar panel.
- Pick your editor: Cline, Cursor, or VS Code (Copilot).
- The extension will:
- Install (or upgrade) the
review-mode-mcp server via uv tool
- Copy the appropriate agent workflows and skills into the correct locations
- Register the MCP server in your editor's configuration
Prerequisite: uv must be installed and available on your PATH.
When you update the extension, you'll be prompted to re-run Install Skills so your agent files stay in sync.
What Gets Installed
| Editor |
MCP server |
Workflows & skills |
MCP config |
| Cursor |
uv tool install review-mode-mcp |
Copied to ~/.cursor/plugins/local/review-mode/ (rules + skills + mcp.json) |
Bundled in the plugin directory |
| Cline |
uv tool install review-mode-mcp |
Copied into the workspace (.clinerules/workflows/, .cline/skills/) |
Auto-added to Cline's cline_mcp_settings.json |
| VS Code (Copilot) |
uv tool install review-mode-mcp |
Provided by the extension's built-in MCP definition provider |
Registered automatically via VS Code's MCP API |
Agent Workflows
After installation, your AI agent gains two slash-command workflows:
| Command |
What it does |
/review-mode |
Opens the current plan (or a specified file) in Review Mode so you can annotate it with inline comments. |
/update-plan |
Reads your annotations, implements the requested changes, resolves comments, and re-opens the file for another review round. |
These workflows are installed as Cursor rules (.mdc files) or Cline workflow rules (.md files), and they teach the agent the full review loop automatically.
Manual MCP Configuration
If you prefer to configure the MCP server manually instead of using Install Skills, install it with uv and add the JSON block to your MCP settings:
uv tool install review-mode-mcp
{
"mcpServers": {
"review-mode-mcp": {
"command": "review-mode-mcp"
}
}
}
Review Mode Controls
- Open any text file.
- Open a file in Review Mode (click the 💬 icon in the editor title bar, or use the command palette: Review Mode: Open File)
- In the Review Mode panel, click the + button next to the line you want to comment on
- Type your comment and press Enter
- Priority: Click the priority dropdown on any comment to set urgency
- Status: Click the status badge to change it:
open — New comment, not yet addressed
in-progress — Being worked on or awaiting clarification
resolved — Fully addressed
won't fix — Acknowledged but not implementing
- Reply: Click the reply button to add a message to the thread
- Delete: Click the trash icon to remove a comment or message
The Review Mode activity bar icon (left sidebar) shows:
- Under Review — All files currently being reviewed
- Review Notes — Annotations for the active review (when a file is open in Review Mode)
Extension Details
- Icon / Codicon Support: Uses built-in VS Code codicons for visually consistent feedback and interaction.
- Data Storage: Annotations and threads are stored efficiently in local JSON metadata files alongside your documents.
Configuration
| Setting |
Default |
Description |
reviewMode.revisionsDirectory |
.revisions |
Directory name for storing revision snapshots and annotation data |
Change this in VS Code settings (Ctrl+,) → search for "Review Mode".
File Structure
When you review a file, Review Mode creates a revision directory:
.revisions/
└── my_plan_md/ # Flattened path of the reviewed file
├── revisions.json # Index of all revisions
├── myplan.rev0.md # Snapshot of the file at revision 0
├── rev0.json # Annotations for revision 0
├── myplan.rev1.md # Snapshot at revision 1
└── rev1.json # Annotations for revision 1
- Snapshots are read-only copies of the file at each revision point
- Annotations are JSON files containing all comments, priorities, and statuses
- The original file is never modified by Review Mode — only the AI agent edits it when iterating
Tips
- Be specific in your comments. Instead of "this needs work", say "add error handling for the case where the API returns 404".
- Use priorities to signal what matters most. The AI will see them.
- Use threads for discussion. If the AI asks a clarifying question (prefixed with
[AGENT]), reply in the same thread to keep context together.
- Don't rush to implement. Take as many review rounds as you need — the goal is a solid plan before any code is written.