Skip to main content

Review Features Overview

Buzzbin does more than post one summary comment. It has several independent capabilities that can be enabled, disabled, or configured at the organization or repository level. The goal is for each team to receive only the output that fits its workflow.

Core capabilities​

  • Code Review: inspect the diff and post inline or summary findings
  • Comment Answers: answer discussions when the bot is mentioned
  • Reviewer Suggestion: suggest reviewers from repository history and activity
  • Title & Description: suggest or apply MR title and description text
  • Related MRs: surface related open merge requests
  • Custom Review Instructions: inject team guidance into the review prompt

A disabled capability does not run, post output, or consume cost for that task.

Output language​

Review, answer, and title/description output can be Persian or English. These settings usually inherit from review configuration, but can be changed per repository when needed.

Diff limits and ignored paths​

Buzzbin prepares the diff before sending it to the model. Ignored paths are removed with glob patterns such as:

  • **/migrations/**
  • **/*.lock
  • **/dist/**

If nothing remains after ignored paths are applied, the review does not run. For very large diffs, Buzzbin may limit the reviewed scope to keep context and cost bounded.

Categories and severity​

Category describes the type of issue, such as Security, Bug, Performance, Maintainability, Testing, or Documentation. Category configuration lets you focus the reviewer on the issue types your team cares about.

Severity describes impact, from Info to Critical. It is used for prioritization, dashboards, and approval gates, but it does not replace team judgment.

Output quality rules​

A good finding should be actionable: it should point to a file and line, explain the risk, and include a fix suggestion when possible. Buzzbin tries to limit noisy, style-only, or out-of-diff findings, but enabled categories and ignored paths are still your best controls for reducing noise.

Execution guardrails​

These controls help keep behavior and spend predictable:

  • daily run cap per repository
  • answer only on mentions
  • auto-approve only when no open blocking issue exists
  • SUGGEST or AUTO_APPLY mode for title/description
  • auto-assign and excluded reviewers for reviewer suggestion

For detailed controls, see Configure Reviews.