Code Review with AI

This guide walks through setting up a code review agent that runs automatically when a ticket receives a review label or an @mention. The agent reads the ticket, evaluates the code changes, and posts a structured review comment — including scores and a final verdict.

Create a Code Review Agent

A custom agent dedicated to code review is created from the Agents page. Clicking Create Agent opens the creation form with four fields:

Name (required) — a descriptive name like “Code Reviewer”
Description (optional) — a summary of what the agent does, such as “Reviews code changes for correctness, test coverage, and consistency”
Avatar — an icon to represent the agent in the interface (the shield icon is a natural fit for a reviewer)
System Prompt (required) — the instructions that define the review criteria and output format

The following screenshot shows the form filled out with code review agent details. Clicking Create Agent saves the agent and navigates to its detail page.

The Create Agent form filled with code review agent details

Write the System Prompt

The system prompt defines the review logic. Template variables like {{ ticket_number }} are automatically replaced with real data when the agent runs. A code review prompt typically instructs the agent to:

Read the code changes associated with the ticket
Evaluate correctness and potential bugs
Check test coverage for new or modified code
Identify edge cases and consistency issues
Post a comment with scores and a final verdict (LGTM or remaining issues)
Remove the “Requires review” label when the review passes

The key template variables for a code review agent are:

Variable	Description
`{{ ticket_number }}`	The ticket number, used to reference the ticket in review comments
`{{ ticket.title }}`	The ticket title
`{{ ticket.body }}`	The ticket description
`{{ comment.body }}`	The triggering comment text (available on mention triggers)
`{{ kombuse_session_id }}`	Required for the agent to post comments and update tickets

The Available variables panel below the prompt editor lists all variables with descriptions and notes on when each is available.

The system prompt editor showing review instructions with template variables

Add Triggers

Triggers determine when the agent runs automatically. The code review agent is configured with two triggers on the Configuration tab of the agent detail page, in the Triggers section (zap icon).

Label trigger

The first trigger fires on Label Added events with a condition matching the “Requires review” label. When any ticket in the project receives this label, the agent starts a review session. This is the primary automation path — other agents or team members add the “Requires review” label when code is ready for review.

The trigger form with Label Added event type and Requires review label condition

Mention trigger

The second trigger fires on Mention Created events with the mention type set to “profile”. This allows anyone to start a review by @mentioning the Code Reviewer in a ticket comment. Setting the mention type to “profile” ensures the trigger fires only on direct @mentions of the agent, not on ticket references.

After both triggers are saved, they appear in the list with individual enable/disable toggles, making it easy to pause either trigger without deleting it.

The Triggers section showing both configured triggers with enable/disable toggles

Set Permissions

Permissions define what the agent is allowed to access at runtime. For a code review agent, four resource permissions are configured:

Resource	Actions	Purpose
Tickets	Read, Update	Read ticket content and update the ticket when the review is complete
Ticket Status	Update	Transition the ticket status (e.g. close it after an approved review)
Ticket Labels	Update, Delete	Manage the “Requires review” label after a review
Comments	Read, Create	Read existing discussion for context and post the review comment

Permissions are added one at a time using the Add Permission button in the Permissions section (shield icon) on the Configuration tab. Without these permissions, the agent’s tool calls are denied at runtime and the denial details are logged in the agent session history.

The Permissions section with code review permissions configured for tickets, labels, and comments

How It Works

With the code review agent configured, the full flow runs automatically:

A user or another agent adds the “Requires review” label to a ticket, or @mentions the Code Reviewer in a comment
The matching trigger fires (label.added or mention.created) and the agent is invoked
The agent receives the ticket context through template variables and begins reviewing
The agent evaluates the code changes, then posts a comment with its findings — including scores for Code Quality, Test Coverage, and Readiness on a 1-10 scale
If the review passes (LGTM), the agent removes the “Requires review” label to signal completion
The session completes — the full review is visible as a comment on the ticket

The agent runs in the background. If it encounters an error or lacks permission for an action, the details are logged in the agent session history, accessible from the agent detail page. The code review agent pairs well with other agents in a pipeline — for example, a coding agent can add the “Requires review” label after completing its work, automatically triggering the review.