feat: add official Codex skill and harden task execution

Install and update artemis-use with task diagnostics and plugin guidance. Preserve cross-device file receipts, review canonical reports, guard stage transitions and reduce small history summaries. Improve exploration lifecycle, request recovery and browser live connections.

tests/unit/agents/test_final_resource_evidence.py

@@ -64,6 +64,20 @@ def test_phone_and_unchanged_pc_captures_are_both_sent_with_origin(tmp_path):
     assert "Step 27" in text and "Step 28" in text
     assert "identical to pre-capture" in text
     assert "not proof of success" in text
+    assert "saved screenshot file=" in text
+    assert "Distinct capture events may share one file" in text
+
+
+def test_unchanged_android_boundary_capture_is_saved_evidence_not_missing_export(tmp_path):
+    footer = image(tmp_path, b"same-footer")
+    blocks = final_evidence(
+        engine(tmp_path, [step(26, post=footer), step(28, pre=footer, post=footer)])
+    )
+    text = str(blocks)
+    assert "Step 26" in text and "Step 28" in text
+    assert str(tmp_path / f"{footer}.jpg") in "\n".join(b.get("text", "") for b in blocks)
+    assert "automatic trace screenshot is recorded evidence" in text
+    assert sum(b["type"] == "image_url" for b in blocks) == 1
 
 
 def test_native_resolution_change_does_not_hide_previous_same_window_capture(tmp_path):
@@ -111,3 +125,13 @@ def test_later_call_without_observation_marks_previous_picture_historical(tmp_pa
     assert "Step 2" in str(blocks) and "no post-action image" in str(blocks)
     assert "Earlier captures do not prove" in str(blocks)
     assert sum(x["type"] == "image_url" for x in blocks) == 1
+
+
+def test_file_receipt_is_linked_to_its_resource_image_without_an_extra_image(tmp_path):
+    name = image(tmp_path, b"pc")
+    action = pc(name)
+    action["attachments"] = [{"path": "artifacts/plugins/cua-driver/pc.png", "sha256": name}]
+    blocks = final_evidence(engine(tmp_path, [step(35, action=action)]))
+    assert "runtime-verified saved attachments" in str(blocks)
+    assert "artifacts/plugins/cua-driver/pc.png" in str(blocks)
+    assert sum(x["type"] == "image_url" for x in blocks) == 1