An AL-aware code review bot for Business Central (GitHub Action)

I built a small GitHub Action for my employer @AidentErfurt that reviews AL pull requests using OpenAI, Azure OpenAI or OpenRouter.ai and some Business Central context out of the box.

Yes, there are plenty of “AI reviewers”. This one is opinionated for BC:

• Detects app structure and pulls nearby files (like app.json and docs) as context.
• Surfaces permissions/entitlements changes and missing coverage.
• Adds hints from the AL Guidelines and links the relevant sections.
• Keeps noise down (inline comment cap) and runs incrementally on new commits.

It’s open source (Apache-2.0), and it’s not our core product, just something useful we wanted to share.

• Marketplace:https://github.com/marketplace/actions/business-central-ai-code-reviewer
• Repo: https://github.com/AidentErfurt/bc-ai-reviewer

Update - 2025-08-19

• GPT-5-ready. When you use AI_MODEL: gpt-5 (OpenAI or an Azure deployment with that name), the action switches to a GPT-5-optimized prompt and accepts REASONING_EFFORT: low|medium|high. It will try the Responses API when available and fall back to Chat Completions automatically.
• Cheat-sheet aligned. The reviewer prompt/behavior now follows OpenAI’s GPT-5 for coding cheat sheet to keep outputs structured (strict JSON), precise, and focused: https://cdn.openai.com/API/docs/gpt-5-for-coding-cheatsheet.pdf.

Minimal config example:

AI_PROVIDER: azure
AI_MODEL:    gpt-5          # deployment name on Azure; model name on OpenAI
PROMPT_STYLE: auto          # auto-switches to GPT-5 prompt
REASONING_EFFORT: medium

Why another reviewer?

Generic code reviewers don’t know AL (yet), Business Central terminology, or the usual pitfalls. This action brings a bit of BC context so the feedback is more relevant and less noisy. Also, and this is probably the actual reason, it’s my vibe coding project 😇.

What it does in a pull request

1. Reads the PR’s unified diff and (optionally) linked issues.
2. Automatically discovers app roots and adds app.json, nearby docs, and (optionally) permissions/entitlements to the context.
3. Pulls AL Guideline docs for patterns it detects in the diff.
4. Asks the model for a short summary + a limited number of inline comments.
5. Posts a PR review; subsequent runs focus on new commits only.

Quick start (Azure OpenAI, o3-mini)

Create .github/workflows/bc-ai-review.yml:

name: AI Code Review (BC)

on:
  pull_request:
    branches: [ main ]
    types: [ opened, synchronize ]

permissions:
  contents: read
  pull-requests: write
  issues: read

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Run BC AI Reviewer (Azure)
        uses: AidentErfurt/bc-ai-reviewer@main
        with:
          GITHUB_TOKEN:      $
          AI_PROVIDER:       azure
          AZURE_ENDPOINT:    https://<your-resource>.openai.azure.com
          AZURE_API_KEY:     $
          AZURE_API_VERSION: 2025-01-01-preview
          AI_MODEL:          o3-mini

          # Optional tuning
          MAX_COMMENTS:      10
          INCLUDE_PATTERNS:  "**/*.al,**/*.xlf,**/*.json"
          AUTO_DETECT_APPS:  true

What’s different (BC-specific bits)

• App structure awareness Finds the nearest app.json for changed files and adds relevant context automatically.

• Permissions & entitlements Highlights permission/entitlement changes and can nudge when coverage is missing.

• AL Guideline hints Pulls guideline snippets for patterns it detects in the diff (performance/locking/naming/etc.).

• Incremental and low-noise Hard cap on inline comments; focuses on new commits so you don’t re-litigate old feedback.

Configuration you’ll probably care about

• AI_PROVIDER - azure (default), openai, or openrouter
• AI_MODEL - defaults to o3-mini
• INCLUDE_PATTERNS / EXCLUDE_PATTERNS - scope the diff so you don’t review pipelines/docs, unless you want to
• AUTO_DETECT_APPS, INCLUDE_APP_PERMISSIONS, INCLUDE_APP_MARKDOWN - context controls
• MAX_COMMENTS - 0 for “as many as needed”, or set a cap (we default to 10)

Full list is in the README.

Privacy & data flow

The action sends PR diffs and selected context files to your chosen provider (Azure OpenAI by default). Keep secrets and sensitive data out of diffs/context. See provider privacy docs before using it in regulated environments.

• Marketplace: https://github.com/marketplace/actions/business-central-ai-code-reviewer
• Repo: https://github.com/AidentErfurt/bc-ai-reviewer