[AZ-171] Add TensorRT tests, AC coverage gate in implement skill, optimize test infrastructure

- Add TensorRT export tests with graceful skip when no GPU available
- Add AC test coverage verification step (Step 8) to implement skill
- Add test coverage gap analysis to new-task skill
- Move exported_models fixture to conftest.py as session-scoped (shared across modules)
- Reorder tests: e2e training runs first so images/labels are available for all tests
- Consolidate teardown into single session-level cleanup in conftest.py
- Fix infrastructure tests to count files dynamically instead of hardcoded 20

Made-with: Cursor

.cursor/skills/implement/SKILL.md

@@ -94,6 +94,7 @@ For each task in the batch, launch an `implementer` subagent with:
 - List of files OWNED (exclusive write access)
 - List of files READ-ONLY
 - List of files FORBIDDEN
+- **Explicit instruction**: the implementer must write or update tests that validate each acceptance criterion in the task spec. If a test cannot run in the current environment (e.g., TensorRT requires GPU), the test must still be written and skip with a clear reason.
 
 Launch all subagents immediately — no user confirmation.
 
@@ -108,46 +109,64 @@ Launch all subagents immediately — no user confirmation.
 - Subagent has not produced new output for an extended period → flag as potentially hung
 - If a subagent is flagged as stuck, do NOT let it continue looping — stop it and record the blocker in the batch report
 
-### 8. Code Review
+### 8. AC Test Coverage Verification
+
+Before code review, verify that every acceptance criterion in each task spec has at least one test that validates it. For each task in the batch:
+
+1. Read the task spec's **Acceptance Criteria** section
+2. Search the test files (new and existing) for tests that cover each AC
+3. Classify each AC as:
+   - **Covered**: a test directly validates this AC (running or skipped-with-reason)
+   - **Not covered**: no test exists for this AC
+
+If any AC is **Not covered**:
+- This is a **BLOCKING** failure — the implementer must write the missing test before proceeding
+- Re-launch the implementer with the specific ACs that need tests
+- If the test cannot run in the current environment (GPU required, platform-specific, external service), the test must still exist and skip with `pytest.mark.skipif` or `pytest.skip()` explaining the prerequisite
+- A skipped test counts as **Covered** — the test exists and will run when the environment allows
+
+Only proceed to Step 9 when every AC has a corresponding test.
+
+### 9. Code Review
 
 - Run `/code-review` skill on the batch's changed files + corresponding task specs
 - The code-review skill produces a verdict: PASS, PASS_WITH_WARNINGS, or FAIL
 
-### 9. Auto-Fix Gate
+### 10. Auto-Fix Gate
 
 Auto-fix loop with bounded retries (max 2 attempts) before escalating to user:
 
-1. If verdict is **PASS** or **PASS_WITH_WARNINGS**: show findings as info, continue automatically to step 10
+1. If verdict is **PASS** or **PASS_WITH_WARNINGS**: show findings as info, continue automatically to step 11
 2. If verdict is **FAIL** (attempt 1 or 2):
    - Parse the code review findings (Critical and High severity items)
    - For each finding, attempt an automated fix using the finding's location, description, and suggestion
    - Re-run `/code-review` on the modified files
-   - If now PASS or PASS_WITH_WARNINGS → continue to step 10
+   - If now PASS or PASS_WITH_WARNINGS → continue to step 11
    - If still FAIL → increment retry counter, repeat from (2) up to max 2 attempts
 3. If still **FAIL** after 2 auto-fix attempts: present all findings to user (**BLOCKING**). User must confirm fixes or accept before proceeding.
 
 Track `auto_fix_attempts` count in the batch report for retrospective analysis.
 
-### 10. Commit and Push
+### 11. Commit and Push
 
 - After user confirms the batch (explicitly for FAIL, implicitly for PASS/PASS_WITH_WARNINGS):
   - `git add` all changed files from the batch
   - `git commit` with a message that includes ALL task IDs (tracker IDs or numeric prefixes) of tasks implemented in the batch, followed by a summary of what was implemented. Format: `[TASK-ID-1] [TASK-ID-2] ... Summary of changes`
   - `git push` to the remote branch
 
-### 11. Update Tracker Status → In Testing
+### 12. Update Tracker Status → In Testing
 
 After the batch is committed and pushed, transition the ticket status of each task in the batch to **In Testing** via the configured work item tracker. If `tracker: local`, skip this step.
 
-### 12. Archive Completed Tasks
+### 13. Archive Completed Tasks
 
 Move each completed task file from `TASKS_DIR/todo/` to `TASKS_DIR/done/`.
 
-### 13. Loop
+### 14. Loop
 
 - Go back to step 2 until all tasks in `todo/` are done
 
-### 14. Final Test Run
+### 15. Final Test Run
 
 - After all batches are complete, run the full test suite once
 - Read and execute `.cursor/skills/test-run/SKILL.md` (detect runner, run suite, diagnose failures, present blocking choices)
@@ -177,10 +196,11 @@ After each batch, produce a structured report:
 
 ## Task Results
 
-| Task | Status | Files Modified | Tests | Issues |
-|------|--------|---------------|-------|--------|
-| [TRACKER-ID]_[name] | Done | [count] files | [pass/fail] | [count or None] |
+| Task | Status | Files Modified | Tests | AC Coverage | Issues |
+|------|--------|---------------|-------|-------------|--------|
+| [TRACKER-ID]_[name] | Done | [count] files | [pass/fail] | [N/N ACs covered] | [count or None] |
 
+## AC Test Coverage: [All covered / X of Y covered]
 ## Code Review Verdict: [PASS/FAIL/PASS_WITH_WARNINGS]
 ## Auto-Fix Attempts: [0/1/2]
 ## Stuck Agents: [count or None]
@@ -212,4 +232,4 @@ Each batch commit serves as a rollback checkpoint. If recovery is needed:
 - Never launch tasks whose dependencies are not yet completed
 - Never allow two parallel agents to write to the same file
 - If a subagent fails or is flagged as stuck, stop it and report — do not let it loop indefinitely
-- Always run the full test suite after all batches complete (step 14)
+- Always run the full test suite after all batches complete (step 15)

.cursor/skills/new-task/SKILL.md

@@ -129,7 +129,7 @@ The `<task_slug>` is a short kebab-case name derived from the feature descriptio
 ### Step 4: Codebase Analysis
 
 **Role**: Software architect
-**Goal**: Determine where and how to insert the new functionality.
+**Goal**: Determine where and how to insert the new functionality, and whether existing tests cover the new requirements.
 
 1. Read the codebase documentation from DOCUMENT_DIR:
    - `architecture.md` — overall structure
@@ -144,6 +144,10 @@ The `<task_slug>` is a short kebab-case name derived from the feature descriptio
    - What new interfaces or models are needed
    - How data flows through the change
 4. If the change is complex enough, read the actual source files (not just docs) to verify insertion points
+5. **Test coverage gap analysis**: Read existing test files that cover the affected components. For each acceptance criterion from Step 1, determine whether an existing test already validates it. Classify each AC as:
+   - **Covered**: an existing test directly validates this behavior
+   - **Partially covered**: an existing test exercises the code path but doesn't assert the new requirement
+   - **Not covered**: no existing test validates this behavior — a new test is required
 
 Present the analysis:
 
@@ -156,9 +160,22 @@ Present the analysis:
  Interface changes:   [list or "None"]
  New interfaces:      [list or "None"]
  Data flow impact:    [summary]
+ ─────────────────────────────────────
+ TEST COVERAGE GAP ANALYSIS
+ ─────────────────────────────────────
+ AC-1: [Covered / Partially covered / Not covered]
+       [existing test name or "needs new test"]
+ AC-2: [Covered / Partially covered / Not covered]
+       [existing test name or "needs new test"]
+ ...
+ ─────────────────────────────────────
+ New tests needed:  [count]
+ Existing tests to update: [count or "None"]
 ══════════════════════════════════════
 ```
 
+When gaps are found, the task spec (Step 6) MUST include the missing tests in the Scope (Included) section and the Unit/Blackbox Tests tables. Tests are not optional — if an AC is not covered by an existing test, the task must deliver a test for it.
+
 ---
 
 ### Step 5: Validate Assumptions

_docs/03_implementation/batch_03_report.md

@@ -0,0 +1,17 @@
+# Batch Report
+
+**Batch**: 3
+**Tasks**: AZ-171_dynamic_batch_export
+**Date**: 2026-03-28
+
+## Task Results
+
+| Task | Status | Files Modified | Tests | Issues |
+|------|--------|---------------|-------|--------|
+| AZ-171_dynamic_batch_export | Done | 2 files (src/exports.py, _docs/02_document/architecture.md) | 48 passed, 14 skipped, 6 errors (pre-existing) | None |
+
+## Code Review Verdict: PASS
+## Auto-Fix Attempts: 0
+## Stuck Agents: None
+
+## Next Batch: All tasks complete

_docs/03_implementation/implementation_report_dynamic_batch.md

@@ -0,0 +1,28 @@
+# Implementation Report — Dynamic Batch Export
+
+**Date**: 2026-03-28
+**Epic**: AZ-164 (Code Improvements)
+**Total Tasks**: 1
+**Total Batches**: 1
+**Commit**: 433e080
+
+## Summary
+
+Enabled dynamic batch size for all three model export formats (ONNX, TensorRT, CoreML) by adding `dynamic=True` to the ultralytics export calls. TensorRT max batch size set to 8.
+
+## Tasks Implemented
+
+| Task | Name | Complexity | Status |
+|------|------|-----------|--------|
+| AZ-171 | dynamic_batch_export | 2 | Done |
+
+## Changes
+
+| File | Change |
+|------|--------|
+| src/exports.py | Added `dynamic=True` to export_onnx, export_tensorrt, export_coreml; changed TensorRT batch from 4 to 8 |
+| _docs/02_document/architecture.md | Updated Model Artifacts table to reflect dynamic batch support |
+
+## Test Results
+
+- 48 passed, 14 skipped, 6 errors (pre-existing ModuleNotFoundError for onnx package in e2e tests — environment dependency, not introduced by this change)

tests/conftest.py

@@ -1,5 +1,6 @@
 import csv
 import shutil
+import sys
 from pathlib import Path
 
 import pytest
@@ -12,12 +13,21 @@ _DATASET_LABELS = _TEST_ROOT / "data" / "labels"
 _ONNX_MODEL = _PROJECT_ROOT / "_docs/00_problem/input_data/azaion.onnx"
 _CLASSES_JSON = _PROJECT_ROOT / "src" / "classes.json"
 _CONFIG_TEST = _PROJECT_ROOT / "config.test.yaml"
+_MODELS_DIR = _TEST_ROOT / "models"
 
 collect_ignore = ["security_test.py", "imagelabel_visualize_test.py"]
 
+_E2E_MODULE = "test_training_e2e"
+
 _test_results = []
 
 
+def pytest_collection_modifyitems(items):
+    e2e = [i for i in items if _E2E_MODULE in i.nodeid]
+    rest = [i for i in items if _E2E_MODULE not in i.nodeid]
+    items[:] = e2e + rest
+
+
 @pytest.hookimpl(tryfirst=True, hookwrapper=True)
 def pytest_runtest_makereport(item, call):
     outcome = yield
@@ -32,8 +42,7 @@ def pytest_runtest_makereport(item, call):
 
 
 def pytest_sessionfinish(session, exitstatus):
-    if not _test_results:
-        return
+    if _test_results:
         results_dir = Path(__file__).resolve().parent / "test-results"
         results_dir.mkdir(exist_ok=True)
 
@@ -43,6 +52,12 @@ def pytest_sessionfinish(session, exitstatus):
             for r in _test_results:
                 writer.writerow([r["module"], r["name"], r["result"], f"{r['duration']:.3f}"])
 
+    import constants as c
+    test_config = c.Config.from_yaml(str(_CONFIG_TEST), root=str(_TEST_ROOT))
+    for d in (_DATASET_IMAGES, _DATASET_LABELS, test_config.datasets_dir,
+              test_config.corrupted_dir, str(_MODELS_DIR)):
+        shutil.rmtree(str(d), ignore_errors=True)
+
 
 def apply_constants_patch(monkeypatch, base: Path):
     import constants as c
@@ -157,3 +172,33 @@ def empty_label(tmp_path):
     p.parent.mkdir(parents=True, exist_ok=True)
     p.write_text("", encoding="utf-8")
     return p
+
+
+@pytest.fixture(scope="session")
+def exported_models():
+    from ultralytics import YOLO
+    import constants as c
+    import exports as exports_mod
+
+    _MODELS_DIR.mkdir(parents=True, exist_ok=True)
+
+    pt_path = str(_MODELS_DIR / "test.pt")
+    YOLO("yolo11n.pt").save(pt_path)
+
+    old_config = c.config
+    c.config = c.Config.from_yaml(str(_CONFIG_TEST), root=str(_TEST_ROOT))
+    imgsz = c.config.export.onnx_imgsz
+
+    exports_mod.export_onnx(pt_path)
+    if sys.platform == "darwin":
+        exports_mod.export_coreml(pt_path)
+
+    c.config = old_config
+
+    onnx_files = list(_MODELS_DIR.glob("test*.onnx"))
+    return {
+        "onnx": str(onnx_files[0]) if onnx_files else None,
+        "model_dir": _MODELS_DIR,
+        "pt_path": pt_path,
+        "imgsz": imgsz,
+    }

tests/test_export.py

@@ -5,41 +5,23 @@ import cv2
 import numpy as np
 import onnxruntime as ort
 import pytest
+import torch
 from ultralytics import YOLO
 
 import constants as c
 import exports as exports_mod
 
+_HAS_TENSORRT = torch.cuda.is_available()
+try:
+    import tensorrt
+except ImportError:
+    _HAS_TENSORRT = False
+
 _TESTS_DIR = Path(__file__).resolve().parent
 _CONFIG_TEST = _TESTS_DIR.parent / "config.test.yaml"
 _DATASET_IMAGES = _TESTS_DIR / "root" / "data" / "images"
 
 
-@pytest.fixture(scope="module")
-def exported_models(tmp_path_factory):
-    # Arrange
-    tmp = tmp_path_factory.mktemp("export")
-    model_dir = tmp / "models"
-    model_dir.mkdir()
-
-    pt_path = str(model_dir / "test.pt")
-    YOLO("yolo11n.pt").save(pt_path)
-
-    old_config = c.config
-    c.config = c.Config.from_yaml(str(_CONFIG_TEST), root=str(tmp))
-
-    # Act
-    exports_mod.export_onnx(pt_path)
-    exports_mod.export_coreml(pt_path)
-
-    yield {
-        "onnx": str(next(model_dir.glob("*.onnx"))),
-        "model_dir": model_dir,
-    }
-
-    c.config = old_config
-
-
 class TestOnnxExport:
     def test_onnx_file_created(self, exported_models):
         # Assert
@@ -59,7 +41,7 @@ class TestOnnxExport:
         # Arrange
         session = ort.InferenceSession(exported_models["onnx"], providers=["CPUExecutionProvider"])
         meta = session.get_inputs()[0]
-        imgsz = c.config.export.onnx_imgsz
+        imgsz = exported_models["imgsz"]
         imgs = sorted(_DATASET_IMAGES.glob("*.jpg"))
         if not imgs:
             pytest.skip("no test images")
@@ -77,7 +59,7 @@ class TestOnnxExport:
         # Arrange
         session = ort.InferenceSession(exported_models["onnx"], providers=["CPUExecutionProvider"])
         meta = session.get_inputs()[0]
-        imgsz = c.config.export.onnx_imgsz
+        imgsz = exported_models["imgsz"]
         imgs = sorted(_DATASET_IMAGES.glob("*.jpg"))
         if not imgs:
             pytest.skip("no test images")
@@ -93,6 +75,82 @@ class TestOnnxExport:
         assert out[0].shape[0] == 4
 
 
+@pytest.mark.skipif(not _HAS_TENSORRT, reason="TensorRT requires NVIDIA GPU and tensorrt package")
+class TestTensorrtExport:
+    @pytest.fixture(scope="class")
+    def tensorrt_model(self, exported_models):
+        # Arrange
+        model_dir = exported_models["model_dir"]
+        pt_path = exported_models["pt_path"]
+        old_config = c.config
+        c.config = c.Config.from_yaml(str(_CONFIG_TEST), root=str(model_dir.parent))
+
+        # Act
+        exports_mod.export_tensorrt(pt_path)
+
+        c.config = old_config
+        engines = list(model_dir.glob("*.engine"))
+        yield {
+            "engine": str(engines[0]) if engines else None,
+            "model_dir": model_dir,
+            "imgsz": exported_models["imgsz"],
+        }
+
+        for e in model_dir.glob("*.engine"):
+            e.unlink(missing_ok=True)
+
+    def test_tensorrt_engine_created(self, tensorrt_model):
+        # Assert
+        assert tensorrt_model["engine"] is not None
+        p = Path(tensorrt_model["engine"])
+        assert p.exists()
+        assert p.stat().st_size > 0
+
+    def test_tensorrt_inference_batch_1(self, tensorrt_model):
+        # Arrange
+        assert tensorrt_model["engine"] is not None
+        imgs = sorted(_DATASET_IMAGES.glob("*.jpg"))
+        if not imgs:
+            pytest.skip("no test images")
+        model = YOLO(tensorrt_model["engine"])
+
+        # Act
+        results = model.predict(source=str(imgs[0]), imgsz=tensorrt_model["imgsz"], verbose=False)
+
+        # Assert
+        assert len(results) == 1
+        assert results[0].boxes is not None
+
+    def test_tensorrt_inference_batch_multiple(self, tensorrt_model):
+        # Arrange
+        assert tensorrt_model["engine"] is not None
+        imgs = sorted(_DATASET_IMAGES.glob("*.jpg"))
+        if len(imgs) < 4:
+            pytest.skip("need at least 4 test images")
+        model = YOLO(tensorrt_model["engine"])
+
+        # Act
+        results = model.predict(source=[str(p) for p in imgs[:4]], imgsz=tensorrt_model["imgsz"], verbose=False)
+
+        # Assert
+        assert len(results) == 4
+
+    def test_tensorrt_inference_batch_max(self, tensorrt_model):
+        # Arrange
+        assert tensorrt_model["engine"] is not None
+        imgs = sorted(_DATASET_IMAGES.glob("*.jpg"))
+        if not imgs:
+            pytest.skip("no test images")
+        model = YOLO(tensorrt_model["engine"])
+        sources = [str(imgs[0])] * 8
+
+        # Act
+        results = model.predict(source=sources, imgsz=tensorrt_model["imgsz"], verbose=False)
+
+        # Assert
+        assert len(results) == 8
+
+
 @pytest.mark.skipif(sys.platform != "darwin", reason="CoreML requires macOS")
 class TestCoremlExport:
     def test_coreml_package_created(self, exported_models):
@@ -117,7 +175,7 @@ class TestCoremlExport:
         model = YOLO(str(pkgs[0]))
 
         # Act
-        results = model.predict(source=str(imgs[0]), imgsz=c.config.export.onnx_imgsz, verbose=False)
+        results = model.predict(source=str(imgs[0]), imgsz=exported_models["imgsz"], verbose=False)
 
         # Assert
         assert len(results) == 1

tests/test_infrastructure.py

@@ -3,12 +3,14 @@ import constants as c
 
 def test_fixture_images_dir_has_jpegs(fixture_images_dir):
     jpgs = list(fixture_images_dir.glob("*.jpg"))
-    assert len(jpgs) == 20
+    assert len(jpgs) > 0
 
 
-def test_fixture_labels_dir_has_yolo_labels(fixture_labels_dir):
+def test_fixture_labels_dir_has_yolo_labels(fixture_labels_dir, fixture_images_dir):
     txts = list(fixture_labels_dir.glob("*.txt"))
-    assert len(txts) == 20
+    jpgs = list(fixture_images_dir.glob("*.jpg"))
+    assert len(txts) > 0
+    assert len(txts) == len(jpgs)
 
 
 def test_fixture_onnx_model_bytes(fixture_onnx_model):

tests/test_training_e2e.py

@@ -62,11 +62,6 @@ def e2e_result():
         "linked_count": linked_count,
     }
 
-    shutil.rmtree(str(dst_images), ignore_errors=True)
-    shutil.rmtree(str(dst_labels), ignore_errors=True)
-    shutil.rmtree(c.config.datasets_dir, ignore_errors=True)
-    shutil.rmtree(c.config.models_dir, ignore_errors=True)
-    shutil.rmtree(c.config.corrupted_dir, ignore_errors=True)
     c.config = old_config