diff --git a/pyproject.toml b/pyproject.toml
index 04e371f..c09681c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -22,5 +22,14 @@ dependencies = [
 [project.scripts]
 anvil = "anvil.cli:app"
 
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+]
+
 [tool.uv]
 package = true
+
+[tool.pytest.ini_options]
+pythonpath = ["src"]
+testpaths = ["tests"]
diff --git a/src/anvil/wizard/models.py b/src/anvil/wizard/models.py
index 476b178..a6c516b 100644
--- a/src/anvil/wizard/models.py
+++ b/src/anvil/wizard/models.py
@@ -15,11 +15,15 @@ class TestSpec:
 
     def to_fail_to_pass_str(self) -> str:
         """Format fail_to_pass as a string list for instance_info.txt."""
-        return "[" + ", ".join(f"'{t}'" for t in self.fail_to_pass) + "]"
+        # Use repr so test IDs containing quotes (e.g. parametrized pytest ids
+        # like "test_x[can't]") stay valid Python literals that round-trip
+        # through the instance_info.txt parser. Manual quoting produced invalid
+        # syntax and silently dropped the whole list on parse.
+        return repr(list(self.fail_to_pass))
 
     def to_pass_to_pass_str(self) -> str:
         """Format pass_to_pass as a string list for instance_info.txt."""
-        return "[" + ", ".join(f"'{t}'" for t in self.pass_to_pass) + "]"
+        return repr(list(self.pass_to_pass))
 
 
 @dataclass
diff --git a/tests/test_wizard_models.py b/tests/test_wizard_models.py
new file mode 100644
index 0000000..82a77a1
--- /dev/null
+++ b/tests/test_wizard_models.py
@@ -0,0 +1,34 @@
+"""Tests for anvil.wizard.models."""
+
+from __future__ import annotations
+
+import ast
+
+from anvil.wizard import models
+
+
+def test_test_spec_str_roundtrip_common_case():
+    spec = models.TestSpec(
+        fail_to_pass=["tests/test_a.py::test_one"],
+        pass_to_pass=["tests/test_a.py::test_two", "tests/test_b.py::test_three"],
+    )
+    assert ast.literal_eval(spec.to_fail_to_pass_str()) == spec.fail_to_pass
+    assert ast.literal_eval(spec.to_pass_to_pass_str()) == spec.pass_to_pass
+
+
+def test_test_spec_str_roundtrip_with_quotes_in_test_id():
+    # Parametrized pytest ids can contain apostrophes, e.g. test_x[can't].
+    # These used to produce invalid Python and get silently dropped when the
+    # instance_info.txt parser evaluated the string.
+    fail = ["tests/test_x.py::test_quote[can't]"]
+    pass_ = ['tests/test_y.py::test_dquote[say "hi"]']
+    spec = models.TestSpec(fail_to_pass=fail, pass_to_pass=pass_)
+
+    assert ast.literal_eval(spec.to_fail_to_pass_str()) == fail
+    assert ast.literal_eval(spec.to_pass_to_pass_str()) == pass_
+
+
+def test_test_spec_str_empty():
+    spec = models.TestSpec()
+    assert ast.literal_eval(spec.to_fail_to_pass_str()) == []
+    assert ast.literal_eval(spec.to_pass_to_pass_str()) == []