ci: run the full make test target so the Kaggle suite executes in CI

PR #66 added the skills/72-kaggle-research runtime suite to `make test`
and widened the pre-commit `files:` triggers to include that directory,
but left both the CI step and the pre-commit `entry:` as a bare
`unittest discover -s tests`. The 50 new tests therefore never ran in
CI, and editing Kaggle code fired a hook that did not cover it.

Point both at `make test` (still stdlib-only, no pip) and add
assertions in tests/test_repo_tools.py so the gap cannot reopen.

Verified: make check-fast green (268 + 50 tests); the Kaggle suite also
passes standalone under CPython 3.9.23, matching the CI matrix floor.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Author: brycewang-stanford

.github/workflows/quality-evals.yml

@@ -42,7 +42,10 @@ jobs:
           --expect-categories causal-identification,reproducibility,citation-hygiene,runtime-safety,research-integrity,writing-compliance,writing-style
 
       - name: Unit tests (stdlib unittest)
-        run: python3 -m unittest discover -s tests -p "test_*.py"
+        # `make test` rather than a bare `unittest discover -s tests`: the
+        # Makefile target also runs the skills/72-kaggle-research runtime
+        # suite, which would otherwise never execute in CI. Still stdlib-only.
+        run: make test
 
       - name: Benchmark (reference pipeline + strict grading)
         run: |

.pre-commit-config.yaml

@@ -15,7 +15,11 @@ repos:
         files: ^eval-harness/
       - id: aers-unit-tests
         name: AERS unit tests
-        entry: python3 -m unittest discover -s tests -p "test_*.py"
+        # `make test` (not a bare `unittest discover -s tests`) so the hook
+        # covers every suite the Makefile knows about — including the
+        # skills/72-kaggle-research runtime tests that this hook's `files:`
+        # pattern already triggers on.
+        entry: make test
         language: system
         pass_filenames: false
         files: ^(tests/|eval-harness/|benchmark/|scripts/|skills/72-kaggle-research/)

tests/test_repo_tools.py

@@ -309,6 +309,10 @@ def test_quality_workflow_uses_non_writing_gates(self):
         self.assertIn("python3 benchmark/reference_pipeline.py --check", text)
         self.assertNotIn("python3 benchmark/reference_pipeline.py\n", text)
         self.assertIn("--no-write", text)
+        # CI must invoke the Makefile target, not a bare `discover -s tests`:
+        # the latter silently skips every suite outside tests/.
+        self.assertIn("run: make test", text)
+        self.assertNotIn("run: python3 -m unittest discover -s tests", text)
 
     def test_pre_commit_uses_non_writing_benchmark_gate(self):
         text = (ROOT / ".pre-commit-config.yaml").read_text(encoding="utf-8")
@@ -320,6 +324,10 @@ def test_pre_commit_uses_non_writing_benchmark_gate(self):
         self.assertIn("aers-tracked-file-hygiene", text)
         self.assertIn("python3 scripts/check-repo-hygiene.py", text)
         self.assertIn("skills/72-kaggle-research/", text)
+        # Same reason as the CI workflow: the unit-test hook runs `make test`
+        # so it covers every suite its `files:` pattern can trigger on.
+        self.assertIn("entry: make test", text)
+        self.assertNotIn("entry: python3 -m unittest discover -s tests", text)
 
     def test_make_check_includes_python_compatibility_compile(self):
         text = (ROOT / "Makefile").read_text(encoding="utf-8")