aldc-al-development-collection

ALDC — AL Agentic Engineering System

Skills-based, spec-driven, TDD-orchestrated development framework for Microsoft Dynamics 365 Business Central.

Engineering systems, visibly reasoned.

ALDC combines specialist agents, reusable domain skills and human approval to support Business Central extension development from requirements to review.

ALDC 5.2.0 brings improvements to Copilot Chat, Copilot CLI and Codex. It adds explicit security review, BCQuality article dispositions, stronger planning and delegation contracts, and configurable Conductor autonomy and test execution. The Marketplace badge reports the currently published extension version. See what changed and the changelog.

Channel	Current delivery	Scope
VS Code Marketplace	See Marketplace badge	Copilot Chat extension and packaged toolkit; existing projects adopt new contracts through Update Toolkit.
Latest GitHub release	v5.0.3	Tagged source and toolkit archive; a separate delivery from the VSIX.
5.2.0	Root package, Copilot catalogs and generated CLI/Codex distributions 5.2.0	Prepared source; publication to Marketplace, GitHub releases and npm is separate.
Claude Code edition	5.1.1	Maintained independently, with its own agents, hooks, contracts and release evolution.

Claude Code remains part of ALDC, with a shared engineering foundation and its own maintained agents, hooks, contracts and release evolution. Its surface changes are documented separately in the changelog and claude-plugin/SURFACE.md. A version in one channel does not establish an equivalent update in another, or publication to npm.

For an installed Claude Code plugin, refresh the official source and plugin:

claude plugin marketplace update aldc-marketplace
claude plugin update aldc@aldc-marketplace

Open a new Claude Code session and check the version against its plugin catalog. This refreshes the plugin's contracts; project-local toolkit copies need their own reviewed update. See installation and recovery for each surface.

Start here

Choose the installation guide for your host, then follow the Quick Start. Use your own AL project and BC sandbox; check its app.json, symbols and permissions before any build or publication. ALDC supplies development context and workflows, not a Business Central environment.

Start with the toolkit installed in your project and a bounded requirement. Architect and Spec prepare the design and technical specification for your review before implementation.

Architecture and distribution map · Installation/recovery · Doctor · Public BC/AL/MCP repository catalogue.

What changed in 5.2.0

Reviews now include five mandatory security checks and a disposition for every selected BCQuality article: apply it or explain why it does not apply. Agent SDK reviews check completeness when the app extends Agent Metadata Provider. Architect, Spec Agent and Planning Subagent cite the knowledge informing their decisions. Planning records the standard publisher flow and the chosen event; subscribers explain the choice above their attribute.

Conductor delegates research, each implementation phase and review to their owners. It records a Mini-design for eligible LOW work without an architecture document; the Core v1.2 architecture gate for MEDIUM/HIGH remains. Final reports distinguish verification performed, checks pending, decisions and open risks.

aldc.yaml adds autonomy.gates: human | pre-approved and testing.runner. Human is the default. With the requester's prior approval, gates become recorded checkpoints in decisions.md; hard stops still produce blocked.md. autonomy.commit remains false by default. A configured test runner supplies actual per-test results after compilation; without one, execution is not-run. ALDC_GATES and ALDC_TEST_RUNNER can override the corresponding settings where the host exposes environment variables. These contracts add no host hooks.

The companion VS Code Project Manager offers an explicit BCQuality mode selector for existing projects and native VS Code installation help. Fresh extension installations select BCQuality plugin mode. The canonical aldc.yaml keeps its external-multiroot default. Project Manager's Human/default and Automatic options map to autonomy.gates: human and pre-approved; Automatic requires prior approval and grants no additional commit or publication permission.

Fresh installations in both Chat profiles select BCQuality plugin mode with automatic discovery and no fixed plugin version pin. Updates preserve the project's selected mode, availability preference and provider identity; customized configuration follows the existing collision policy. A newly seeded workspace includes an external knowledge folder only for an enabled external mode and uses its configured path. Existing workspace files remain owned by the project.

Both Chat profiles retain 5.0.4's dependency guidance: resolve only libraries used by the planned tests from approved dependencies, declared versions and symbols. Claude Code stays on its independently maintained 5.1.1 edition.

Use Project Manager in VS Code

1. Update the extension to 5.2.0 and run AL Collection: Open Project Manager.
2. Select your solution, preview its toolkit installation or update, review customizations and confirm. Updating the extension does not replace the toolkit already installed in a project.
3. In BCQuality, choose Plugin — recommended, External knowledge folder or Disabled — native review, then select Change mode. Install BCQuality… copies the official repository URL for the native source prompt; paste it there. Hosts without that command open the installation guide. Manage plugins opens the host's inventory.
4. In Workflow autonomy, choose Human — approve each checkpoint or Automatic — prior approval, then select Apply mode. Start a new chat with AL Development Conductor.

Human waits at approval checkpoints. Automatic records those decisions and continues within the approved task, retaining hard stops. Changes preserve commit and publication permissions and keep a recoverable configuration copy. A retained older Conductor must be updated before using the autonomy control. Plugin installation and loading are verified by the host.

These Project Manager controls apply to VS Code/Copilot Chat. The other hosts keep their own configuration and permissions. Canvas remains a separate experimental companion.

Earlier Copilot Chat improvements

5.0.3 — knowledge before implementation. Architect and the independent Spec Agent consult BCQuality through al-knowledge, read complete articles and record the evidence used in their decisions. Hosts without a skill invocation tool follow the instructions inline. BCQuality is installed separately; unavailable knowledge and manual fallbacks remain explicit. Windows validation and generated-path handling also improve.

5.0.4 — native AL tools for your project. Copilot Chat's bc29-native Developer and Implementation Subagent can query object IDs within approved project ranges. IDs are not reserved, and the local scan excludes installed extensions. Architect/Spec research stays bounded to relevant symbols and dependencies; native diagnostics use the compatible AL18.1 selector. Developer can compile the selected AL project and consult its diagnostics as part of the implementation workflow.

Conductor keeps coordination and Reviewer keeps query permissions. Publication requires explicit authorization. Graph, LSP, profiling and snapshot operations remain optional and depend on the installed host and tools. The native tooling guide explains the available capabilities, project selection and permissions. Claude Code's later surface releases follow their own contracts.

To update an existing VS Code project, update the extension, run AL Collection: Update Toolkit, review the preview and managed-file collisions, then run AL Collection: Verify Installed Toolkit and reload the window. Each terminal plugin has its own update route. Earlier improvements, including Copilot Chat's Sonnet 5.5 selection, remain documented under their original changelog versions.

Capability	What it provides
Architect and Spec Agent	Approved design, bounded Specs, explicit dependencies and joint consistency review before implementation.
Conductor	Planning, implementation and review coordinated through existing human gates.
Doctor	Read-only diagnosis of configuration and operation-specific runtime observations.
Recommended BCQuality	Plugin or external multiroot reviews, with discovery, loading, execution and index evidence distinguished.
BC28 / BC29-native	Explicit Chat profile selection with role-specific tools and AL18 guidance.
Host adapters	Dedicated content for Copilot Chat, Copilot CLI, Claude Code and Codex.
Recoverable installation	Preview, visible collisions, receipts, verification and restoration.

See Spec Agent, profiles, plugins and recovery and Doctor.

---

Why ALDC?

AI accelerates AL development — but raw code generation is unpredictable, hard to review, and easy to get wrong against Business Central's extension rules. ALDC adds the engineering discipline around it.

Without ALDC	With ALDC
Ad-hoc, one-shot code generation	Contract-driven development (architecture → spec → human approval → test-plan → code)