fix: isolate sandbox tests and clarify evaluator pipe input

manifests/context-packs/skill-triggers@1.json

@@ -10,7 +10,7 @@
     "id": "hand-seeded",
     "source": "manual-curation-pending-regeneration"
   },
-  "registryDigest": "6373302fba7684a29f4287625420fc4d7d2b3a989b23373e0f9a6f14e5c9b55e",
+  "registryDigest": "80fbd629779375546bc17b0d5a49c08cb4e22491e03e80f0590ba75aedf9315c",
   "schemaVersion": 1,
   "triggers": {
     "skill:agent-runtime-gateway-smoke-test": [

pi/core/skills/agent-self-evaluation/scripts/evaluate.py

@@ -398,7 +398,13 @@ def main() -> None:
     parser = argparse.ArgumentParser(
         description="Evaluate agent output against the 5-axis rubric"
     )
-    parser.add_argument("--task", help="Task description (file path or inline text)")
+    parser.add_argument(
+        "--task",
+        help=(
+            "Task description (file path or inline text); when omitted with --output and non-TTY "
+            "stdin, read task text until EOF and wait if the input remains open"
+        ),
+    )
     parser.add_argument("--output", help="Agent output to evaluate (file path)")
     parser.add_argument("--interactive", action="store_true", help="Prompt for task and read output from stdin")
     args = parser.parse_args()

skills/agent-self-evaluation/scripts/evaluate.py

@@ -398,7 +398,13 @@ def main() -> None:
     parser = argparse.ArgumentParser(
         description="Evaluate agent output against the 5-axis rubric"
     )
-    parser.add_argument("--task", help="Task description (file path or inline text)")
+    parser.add_argument(
+        "--task",
+        help=(
+            "Task description (file path or inline text); when omitted with --output and non-TTY "
+            "stdin, read task text until EOF and wait if the input remains open"
+        ),
+    )
     parser.add_argument("--output", help="Agent output to evaluate (file path)")
     parser.add_argument("--interactive", action="store_true", help="Prompt for task and read output from stdin")
     args = parser.parse_args()

skills/skill-comply/tests/test_runner.py

@@ -12,7 +12,6 @@ from unittest.mock import patch
 
 import pytest
 from scripts.runner import (
-    SANDBOX_BASE,
     _parse_stream_json,
     _safe_sandbox_dir,
     _setup_sandbox,
@@ -30,8 +29,18 @@ class _FakeScenario:
     setup_commands: tuple[str, ...] = ()
 
 
+@pytest.fixture(autouse=True)
+def isolated_sandbox_base(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    monkeypatch.setattr("scripts.runner.SANDBOX_BASE", tmp_path / "sandbox")
+
+
 class TestSafeSandboxDirectory:
-    def test_empty_id_cannot_delete_the_shared_base(self) -> None:
+    def test_empty_id_cannot_delete_the_shared_base(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        monkeypatch.setattr("scripts.runner.SANDBOX_BASE", tmp_path / "sandbox")
         with (
             patch("scripts.runner.Path.exists", return_value=True),
             patch("scripts.runner.shutil.rmtree") as remove,
@@ -45,13 +54,21 @@ class TestSafeSandboxDirectory:
             mkdir.assert_not_called()
             execute.assert_not_called()
 
-    def test_resolved_base_is_not_a_scenario_directory(self) -> None:
-        with patch("scripts.runner.Path.resolve", return_value=SANDBOX_BASE):
+    def test_resolved_base_is_not_a_scenario_directory(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        base = tmp_path / "sandbox"
+        monkeypatch.setattr("scripts.runner.SANDBOX_BASE", base)
+        with patch("scripts.runner.Path.resolve", return_value=base):
             with pytest.raises(ValueError, match="sandbox"):
                 _safe_sandbox_dir("alias")
 
-    def test_valid_id_keeps_existing_sanitization(self) -> None:
-        assert _safe_sandbox_dir("scenario/one") == SANDBOX_BASE / "scenario_one"
+    def test_valid_id_keeps_existing_sanitization(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        base = tmp_path / "sandbox"
+        monkeypatch.setattr("scripts.runner.SANDBOX_BASE", base)
+        assert _safe_sandbox_dir("scenario/one") == base / "scenario_one"
 
     def test_resolved_escape_is_still_rejected(
         self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch