Skip to content

ALDC

AL Development Collection · Spec-driven AI for Business Central

Ship Business Central extensions with AI agents that follow your process.

ALDC gives Copilot and Claude Code a working model for AL delivery: specs, plans, tests, review gates and reusable skills. Less vibe coding. More traceable implementation.

4 agents 11 skills 6 workflows MIT Copilot + Claude Code

What's ALDC

Most AI coding tools generate a file and hope for the best. ALDC is different.

Every feature starts with a spec contract that is functional, technical, and testable, kept in .github/plans/{req_name}/. Architecture and test plans live next to it.

A conductor agent orchestrates a TDD cycle: the Implementation Subagent writes tests first, code second, then refactors. A Review Subagent validates against the spec. You approve every phase.

Underneath it all, 11 composable skills covering API, events, performance, testing, and more load on demand so agents only know what they need for the task in front of them.

The result: AL code that passes review the first time, with traceable decisions from requirement to merge.

flowchart TD
    R([Requirement]) --> S[Spec Contract<br/><small>functional + technical</small>]
    S --> A[Architecture<br/><small>decisions + diagrams</small>]
    A --> T[Test Plan<br/><small>RED before GREEN</small>]
    T --> I[TDD Implementation<br/><small>subagent loop</small>]
    I --> V[Review<br/><small>vs spec</small>]
    V --> M([Merged])

    style R fill:#5E35B1,color:#fff,stroke:none
    style M fill:#00C853,color:#fff,stroke:none
    style S fill:#7E57C2,color:#fff,stroke:none
    style A fill:#7E57C2,color:#fff,stroke:none
    style T fill:#FF6D00,color:#fff,stroke:none
    style I fill:#FF6D00,color:#fff,stroke:none
    style V fill:#FF6D00,color:#fff,stroke:none

Why it matters

  •   Spec-driven, not prompt-and-pray


    No more "please re-explain the requirement for the 5th time". Every feature has a written contract that agents consult.

  •   TDD enforced, not suggested


    The Implementation Subagent refuses to write code before tests. RED → GREEN → REFACTOR is hardcoded in the agent.

  •   You approve every phase


    Agents stop at architecture, plan, implementation, review, deploy. Nothing ships without a human saying yes.

  •   Extension-only discipline


    Never touches base app objects. Always tableextensions, pageextensions, event subscribers. Least-privilege permissions by default.

  •   One toolkit, two runtimes


    Same primitives work in GitHub Copilot and Claude Code. Pick the tool your team already uses.

  •   Skills load on demand


    11 composable skills replace 300kb of prompt soup. Agents only load what's relevant to the task.


Quick start

git clone https://github.com/javiarmesto/ALDC-AL-Development-Collection.git
cd ALDC-AL-Development-Collection
npm install
npx aldc init

Then in VS Code with Copilot enabled:

@workspace use al-initialize
/plugin install aldc

Then:

/aldc:al-initialize
git clone https://github.com/javiarmesto/ALDC-AL-Development-Collection.git
cd ALDC-AL-Development-Collection
npm install && npm run validate

Resources

Everything ALDC-related lives here. Pick your path.


Events & talks

ALDC also exists outside the repo: conference talks, Dev Days and hands-on sessions where the collection, workflow and delivery model are the topic. Here you only see confirmed upcoming talks.


How to collaborate

Four ways to make ALDC better. No contribution is too small.

01

Try it and tell me

Install, build something real, and open an issue with what broke, what felt awkward, or what you wish existed. Bug reports are gold.

Open an issue →
02

Contribute a primitive

Have a skill, workflow or agent you'd pay for? Propose it. Fork, follow the contribution guide, open a PR.

Contribution guide →
03

Shape v1.2

Priorities are discussed openly. Vote on trade-offs, challenge assumptions, and propose alternatives with the community.

Join the discussion →
04

Share the workflow

Star the repo. Mention ALDC in your BC community, internal enablement docs or session notes. Useful references beat generic hype.

Star on GitHub →