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/memory/test_history_chunking.py

@@ -72,6 +72,49 @@ def _step(i: int, subgoal: str = "hash-a", **overrides) -> dict:
     return step
 
 
+@pytest.mark.parametrize("minimal", [False, True])
+def test_runtime_file_paths_survive_even_minimal_history_folding(minimal):
+    from artemis.memory.chunking import build_action_ledger
+
+    action = {
+        "action": "plug_pc_save",
+        "plugin_id": "cua-driver",
+        "attachments": [
+            {"path": "artifacts/plugins/cua-driver/footer.png", "sha256": "a" * 64, "bytes": 123}
+        ],
+    }
+    text = build_action_ledger([_step(35, action_taken=action)], SESSION_START, minimal=minimal)
+    assert "Step 35" in text and "footer.png" in text and "a" * 64 in text
+
+
+def test_default_larger_batches_reduce_capsule_calls_without_losing_step_index():
+    from artemis.config import MemoryChunkingConfig
+
+    current = MemoryChunkingConfig()
+    steps = [_step(i) for i in range(1, 48)]
+    counts = []
+    for minimum, target in [(3, 2000), (current.min_steps, current.target_source_tokens)]:
+        ledger, chunker, _, capsule = _make(
+            steps,
+            min_active=5,
+            min_steps=minimum,
+            target_tokens=target,
+            auto_capsule=True,
+            meter=lambda: 75_000,
+        )
+        _run_turns(ledger, chunker, 1, 47, lambda _: "same-stage")
+        counts.append(len(capsule.submitted))
+        consumed = {sid for c in chunker.chunks for sid in c.source_step_ids}
+        consumed.update(
+            sid for era in chunker.eras for c in era.chunks for sid in c.source_step_ids
+        )
+        remaining = {sid for turn in ledger.unchunked_turns() for sid in turn.get("step_keys", [])}
+        assert consumed | remaining  # remaining live/frozen records are not dropped
+        for number in range(1, 42):
+            assert f"s{number}" in consumed | remaining
+    assert counts[1] < counts[0]
+
+
 class FakeEngine:
     def __init__(self, steps: list[dict], base_dir=None):
         self.steps = steps