Context-Aware Multi-VCS AI Code Reviewer

Automated Code Reviews with AST Scope Precision

The Code Reviewer AI Agent emulates the precision of CodeRabbit. Built with the Divmora LocalHarness SDK, it analyzes pull requests across GitLab, GitHub, Bitbucket, and local workspaces—producing structured scorecards, file walkthroughs, and 1-click committable suggestions directly on changed lines.

Commit-Scoped Net Diff

Isolates author feature commits, completely filtering out merge-drift noise from master/pre-prod.

Anti-Hallucination AST

Native Go AST & multi-language scope analyzer trims context to enclosing definitions with line decoration.

1-Click Inline Suggestions

Publishes GitHub & GitLab committable replacement blocks with scoped quality badges.

Engine Architecture

graph TD
    A[VCS Ingress: GitLab / GitHub / Bitbucket] --> B[Smart Diff & Commit Isolator]
    B --> C[Noise Filter & Secrets Pre-Scanner]
    C --> D[AST Scope Slicer: Go / Python / TS / PHP / Java]
    D --> E[Divmora LocalHarness SDK]
    E --> F[Review Evaluator & Comment Budgeter]
    F --> G[VCS Publisher: Inline Suggestions, Scorecards, Labels]

GitLab CI/CD Integration

Automatically review GitLab Merge Requests (both GitLab.com SaaS and Self-Hosted instances) on every commit.

.gitlab-ci.yml

stages:
  - test
  - review

# AI Code Review runs in parallel with test, non-blocking
ai-code-review:
  stage: review
  image: ghcr.io/divmora/code-reviewer-ai-agent:latest
  rules:
    # Run on Merge Requests (excluding draft / WIP MRs)
    - if: '$CI_PIPELINE_SOURCE == "merge_request_event" && $CI_MERGE_REQUEST_TITLE !~ /^(Draft|WIP):/i'
  allow_failure: true
  variables:
    # Masked CI/CD Variables configured in Settings > CI/CD > Variables
    CODE_REVIEWER_LITELLM_BASE_URL: "$LITELLM_BASE_URL"
    CODE_REVIEWER_LITELLM_API_KEY: "$LITELLM_API_KEY"
    CODE_REVIEWER_LITELLM_MODEL: "workers-ai/@cf/zai-org/glm-5.2"
  script:
    - code-reviewer \
        --url "$CI_MERGE_REQUEST_PROJECT_URL/-/merge_requests/$CI_MERGE_REQUEST_IID" \
        --token "$GITLAB_TOKEN" \
        --skip-tls-verify \
        --post \
        --update-description \
        --update-labels

GitLab CI Best Practices:

Allow Failure: Setting allow_failure: true ensures the AI review provides rich insights without blocking deployments.
Self-Hosted SSL: Use --skip-tls-verify for internal enterprise GitLab servers with private CA certificates.
Watermark Dedup: Automatically skips review if the target commit SHA has already been evaluated.

GitHub Actions Integration

Run automated reviews on Pull Request creation and code synchronization using GitHub Actions.

.github/workflows/ai-review.yml

name: AI Code Review

on:
  pull_request:
    types: [opened, synchronize, reopened, ready_for_review]

jobs:
  review:
    if: github.event.pull_request.draft == false
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
      statuses: write

    steps:
      - name: Run Code Reviewer Agent
        uses: docker://ghcr.io/divmora/code-reviewer-ai-agent:latest
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          CODE_REVIEWER_LITELLM_BASE_URL: ${{ secrets.LITELLM_BASE_URL }}
          CODE_REVIEWER_LITELLM_API_KEY: ${{ secrets.LITELLM_API_KEY }}
          CODE_REVIEWER_LITELLM_MODEL: "workers-ai/@cf/zai-org/glm-5.2"
        with:
          args: >-
            --url ${{ github.event.pull_request.html_url }}
            --token ${{ secrets.GITHUB_TOKEN }}
            --post

Bitbucket Pipelines Integration

Automate code review feedback on pull requests inside Bitbucket Cloud or Data Center.

bitbucket-pipelines.yml

pipelines:
  pull-requests:
    '**':
      - step:
          name: AI Code Review
          image: ghcr.io/divmora/code-reviewer-ai-agent:latest
          script:
            - code-reviewer \
                --url "$BITBUCKET_GIT_HTTP_ORIGIN/pull-requests/$BITBUCKET_PR_ID" \
                --token "$BITBUCKET_TOKEN" \
                --post

Model & LiteLLM Resolution Hierarchy

The agent resolves LiteLLM credentials, models, and endpoints following a 4-tier resolution chain:

1. CLI Flags (Highest)

--litellm-base-url
--litellm-api-key
--litellm-model
--litellm-endpoint

2. Tool-Specific Env (Scoped)

CODE_REVIEWER_LITELLM_BASE_URL
CODE_REVIEWER_LITELLM_API_KEY
CODE_REVIEWER_LITELLM_MODEL
CODE_REVIEWER_LITELLM_ENDPOINT

3. Generic LiteLLM Env

LITELLM_BASE_URL
LITELLM_API_KEY
LITELLM_MODEL
LITELLM_ENDPOINT

4. Local Config File

~/.divmora/config/litellm.json
Named endpoint or default endpoint resolved automatically.

Review Sensitivity Profiles

Control review tone and strictness via the --profile flag:

chill Light

Focuses strictly on critical errors and security flaws. Ignores styling and minor nits.

balanced (Default) Standard

Thorough and practical. Catches edge cases, security, performance, and maintainability issues.

assertive Strict

Strict audit mode. Flags subtle bugs, design anti-patterns, test gaps, and typing inconsistencies.

CLI Flags Reference

Flag	Default	Description
--url	""	PR / MR URL (GitLab, GitHub, Bitbucket)
--workspace	"."	Local workspace directory to review
--diff	false	Review uncommitted local git changes
--token	$TOKEN	VCS API access token
--post	false	Post inline comments, scorecard description, and labels
--skip-tls-verify	false	Skip TLS verification for internal GitLab self-hosted instances
--profile	"balanced"	Sensitivity profile: chill, balanced, assertive
--max-comments	15	Comment budget cap prioritizing High > Medium > Low
--version, -v	false	Print version metadata and exit