Skill Sdd Run Al Tests

Skill: Lean SDD, Run AL Tests

Purpose

Execute the AL test codeunits for a spec folder, map results to acceptance criteria in tasks.md, and report which criteria pass, fail, or are missing coverage. This is the test & validate step of the lean feature loop.

When to Load

Load when:

• The user says "run tests", /speckit.analyze, "validate", or "check acceptance criteria"
• Implementation tasks in tasks.md are all checked off but AC rows are not yet validated
• A test regression needs to be diagnosed

Prerequisites

1. specs/SDD/YYYY-MM-DD-/tasks.md, AC checklist to validate
2. Test codeunit exists in test/{FeatureName}/ for the spec
3. al_get_diagnostics returns 0 errors (run implement-feature first if not)

Step 1, Read Context

Read specs//tasks.md   → acceptance criteria and test procedure names
Read specs//spec.md    → requirement IDs and AC descriptions

> ` is the dated subfolder under specs/SDD/, e.g. specs/SDD/2026-07-08-fleet-registration/`.

Step 2, Build Test Execution Map

From tasks.md, extract the AC rows:

AC1: {criterion}, test: {TestProcedureName}
AC2: {criterion}, test: {TestProcedureName}

If any AC row has no mapped test procedure, flag it as missing coverage before running.

Step 3, Run Diagnostics First

al_get_diagnostics

If there are compile errors, stop and ask the user to fix them. Do not run tests against broken code.

Step 4, Report Results

For each AC row in tasks.md, determine status:

| Status | Meaning | |--------|---------| | ✅ Pass | Test ran and passed | | ❌ Fail | Test ran and failed, include error message | | ⚠️ Missing | No test exists for this AC, flag as gap | | ⏭️ Skipped | Test is marked [Ignore] |

Step 5, Analyse Failures

For each ❌ failure:

1. Identify the failing assertion and line
2. Compare expected vs actual values
3. Root-cause: business logic bug? Setup missing? Event not fired?
4. Propose a minimal fix, do not refactor beyond the failing assertion

Use bclsp_goToDefinition and bclsp_findReferences to navigate the call stack without re-reading the entire file.

Step 6, Update tasks.md

After all tests pass, check off the AC rows:

- [x] AC1: {criterion}, ✅ validated by `{TestProcedureName}`
- [x] AC2: {criterion}, ✅ validated by `{TestProcedureName}`

Check the Quality Checklist items:

- [x] All new fields have `DataClassification`
- [x] All user-visible strings use AL Labels
- [x] No database calls inside loops
- [x] Permission set updated
- [x] `al_get_diagnostics` returns 0 errors, 0 warnings
- [x] BCQuality instructions honored (auto-applied via extension)

Step 7, Report Summary

Emit a concise summary:

## Test Results, specs/SDD/

| AC | Criterion | Test | Result |
|----|-----------|------|--------|
| AC1 | {criterion} | {TestProc} | ✅ Pass |
| AC2 | {criterion} | {TestProc} | ✅ Pass |

**Coverage**: {N}/{Total} AC validated
**Diagnostics**: 0 errors, 0 warnings
**Status**: Ready for generate-docs → finalise-feature

If any failures remain, list them with proposed fixes and ask the user to confirm before applying.

Common Failure Patterns in AL Tests

Missing Initialize() side-effects

Symptom: Test passes in isolation, fails when run as part of a suite. Fix: Ensure Initialize() resets all shared state and table data.

Event not firing in test

Symptom: Subscriber code is never reached. Fix: The event requires the correct BindSubscription call; in integration tests, the codeunit implementing the subscriber must be bound explicitly if it uses [EventSubscriber] with a non-automatic binding.

Wrong LibrarySales / LibraryPurchase version

Symptom: Library - Sales procedure signature mismatch. Fix: Use al_symbolsearch to find the exact overload available in the current BC version.

Hardcoded string comparison failure

Symptom: Assert.AreEqual fails on text that looks identical. Fix: Never compare user-facing strings directly, compare record IDs or enum values instead. Per constitution rules: no hardcoded strings.

Skills Evidencing

> **Skills loaded**: skill-sdd-run-al-tests (test & validate phase)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Louage
• Source: Louage/frw-agentic-coding
• License: MIT

Install and usage instructions live in the source repository linked above.