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/test_plugin_result_delivery.py

@@ -19,9 +19,65 @@ from artemis.data_engine.models import ImageRecord, SessionMetadata, StepRecord
 from artemis.data_engine.storage import StorageManager
 from artemis.memory.transcript import TranscriptLedger
 from artemis.plugins.results import plugin_result_messages
+from artemis.plugins.models import PluginCallResult
+from artemis.plugins.runtime import PluginRuntime
 from artemis.tools.history.screenshot import load_step_screenshot
 
 
+def test_large_tree_deduplicates_without_mutating_raw_evidence():
+    tree = "- Button abc\n" * 3000
+    result = PluginCallResult(
+        status="success",
+        plugin_id="test",
+        content=[{"type": "text", "text": "Screenshot\n" + tree}],
+        structured_content={
+            "tree_markdown": tree,
+            "capture_id": "capture-current",
+            "window_bounds": {"width": 2560},
+        },
+    )
+    text = result.text()
+    assert text.count(tree) == 1
+    assert result.structured_content["tree_markdown"] == tree
+    assert "capture-current" in text and "2560" in text
+
+
+@pytest.mark.parametrize("mode", ["flash", "pro"])
+def test_verified_receipts_survive_observation_truncation_for_both_modes(mode):
+    receipt = {
+        "path": "artifacts/plugins/test/screen.png",
+        "sha256": "a" * 64,
+        "bytes": 123,
+        "media_type": "image/png",
+    }
+    result = PluginCallResult(
+        status="success",
+        plugin_id="test",
+        content=[{"type": "text", "text": "TREE " * 10000}],
+        attachments=[receipt],
+    )
+    if mode == "flash":
+        blocks = PluginRuntime.content_blocks(result)
+    else:
+        action = _action("call", attachments=[receipt])
+        action["result"] = result.text()
+        blocks = plugin_result_messages({"execution": [action]}, None)["call"][0].content
+    text = str(blocks)
+    assert receipt["path"] in text and receipt["sha256"] in text
+    assert "verified file receipts" in text
+    assert "观察文本已节选" in text
+    assert len(text) < 14000
+
+
+def test_plugin_claim_does_not_become_runtime_receipt():
+    result = PluginCallResult(
+        status="success",
+        plugin_id="test",
+        structured_content={"attachments": [{"path": "pretend.png", "verified": True}]},
+    )
+    assert "verified file receipts" not in str(PluginRuntime.content_blocks(result))
+
+
 def _action(call_id, name="plug_test_get_app_state", **kwargs):
     return dict(
         action=name,