neues modell

2026-09-01 08:12:07 +02:00
parent 611fd0a80c
commit 654339464e
109 changed files with 70505 additions and 27 deletions
@@ -160,6 +160,91 @@ class OpenCodeAdapterTests(unittest.TestCase):
        self.assertEqual(1, len(result["written_files"]))


+class ReadonlyShellTests(unittest.TestCase):
+    """Denylist: alles erlaubt ausser schreibenden und bauenden Kommandos."""
+
+    def test_catch_all_allow_comes_first(self):
+        # OpenCode wertet der Reihe nach aus, die letzte passende Regel gewinnt.
+        # Stuende das Catch-all hinten, waere jede Sperre wirkungslos.
+        perms = ADAPTER.readonly_shell_permissions()
+        self.assertEqual("allow", perms["*"])
+        self.assertEqual("*", next(iter(perms)))
+
+    def test_writing_and_building_commands_are_denied(self):
+        perms = ADAPTER.readonly_shell_permissions()
+        for muster in ("rm *", "mv *", "sed -i*", "git commit*", "git push*",
+                       "dotnet *", "msbuild *", "npm install*",
+                       "Remove-Item *", "Set-Content *", "Out-File *"):
+            self.assertEqual("deny", perms[muster], muster)
+
+    def test_reading_commands_need_no_rule(self):
+        # Der Kern der Umstellung: Lesekommandos - auch als Pipeline - sind
+        # erlaubt, ohne einzeln aufgefuehrt zu sein. Genau daran scheiterte
+        # 'Get-ChildItem ... | Format-Table ...' unter der Allowlist.
+        perms = ADAPTER.readonly_shell_permissions()
+        for kommando in ("ls -la", "cat foo.cs", "rg muster",
+                         "Get-ChildItem -Recurse | Format-Table Name",
+                         "find . -name *.cs"):
+            self.assertNotIn(kommando, perms)
+        self.assertEqual("allow", perms["*"])
+
+    def test_no_reading_command_is_denied(self):
+        verboten_praefixe = ("ls", "cat", "rg", "grep", "find", "head", "tail",
+                             "Get-ChildItem", "Get-Content", "Select-String",
+                             "git status", "git log", "dir", "type")
+        for muster, aktion in ADAPTER.readonly_shell_permissions().items():
+            if aktion != "deny":
+                continue
+            for praefix in verboten_praefixe:
+                self.assertFalse(
+                    muster.lower().startswith(praefix.lower()),
+                    f"Lesekommando gesperrt: {muster}",
+                )
+
+
+class UsageCapturedTests(unittest.TestCase):
+    """Eine Null aus einem laufenden Subagenten ist kein Messwert."""
+
+    def _result(self, session, output):
+        return ADAPTER.normalize_result(
+            session=session, events=[], model_ref="lmstudio/google/gemma-4-e4b",
+            mode="builtin", effort="high", exit_code=1, timed_out=True,
+            interrupted=False, duration_s=3600.0, output_dir=output, errors=[],
+            provider="lmstudio", effort_applied=False,
+        )
+
+    def test_running_subagent_with_zero_tokens_is_not_captured(self):
+        session = {
+            "info": {"model": {"id": "google/gemma-4-e4b"},
+                     "tokens": {"input": 0, "output": 0, "reasoning": 0,
+                                "cache": {"read": 0, "write": 0}}},
+            "messages": [{
+                "info": {"role": "assistant"},
+                "parts": [{"type": "tool", "tool": "task",
+                           "state": {"status": "running",
+                                     "input": {"subagent_type": "explore"}}}],
+            }],
+        }
+        with tempfile.TemporaryDirectory() as temp_dir:
+            result = self._result(session, Path(temp_dir))
+        self.assertFalse(result["usage_captured"])
+        self.assertIn("nicht erfasst", result["usage_note"])
+        self.assertIn("1 Subagent(en)", result["usage_note"])
+
+    def test_real_token_counts_are_marked_captured(self):
+        session = {
+            "info": {"model": {"id": "google/gemma-4-e4b"},
+                     "tokens": {"input": 100, "output": 10, "reasoning": 0,
+                                "cache": {"read": 0, "write": 0}}},
+            "messages": [{"info": {"role": "assistant"},
+                          "parts": [{"type": "text", "text": "ok"}]}],
+        }
+        with tempfile.TemporaryDirectory() as temp_dir:
+            result = self._result(session, Path(temp_dir))
+        self.assertTrue(result["usage_captured"])
+        self.assertNotIn("usage_note", result)
+
+
 class OpenCodeLmStudioTests(unittest.TestCase):
    def test_model_reference_uses_lmstudio_provider(self):
        self.assertEqual(