test: clear the audit debt in the test suite

.pre-commit-config.yaml

@@ -22,7 +22,7 @@ repos:
         args: [--maxkb=1024]
       - id: check-merge-conflict
       - id: detect-private-key
-        exclude: ^tests/test_runtime_checkpoint_bug2_deep\.py$
+        exclude: ^tests/test_runtime_checkpoint_deep\.py$
 
   - repo: local
     hooks:

docs/everos-memory-e2e-test-plan.md

@@ -42,7 +42,7 @@ expectations.
 
 | Layer | What | Where | LLM? |
 |---|---|---|---|
-| L1 | Backend translation (owner routing, message conversion, result flatten, degradation) | `tests/test_em2_backend.py`, `tests/test_em3_http.py` | no (fakes) — default CI |
+| L1 | Backend translation (owner routing, message conversion, result flatten, degradation) | `tests/test_everos_backend.py`, `tests/test_everos_http_adapter.py` | no (fakes) — default CI |
 | L2 | everos extraction quality — direct service + `is_final` flush | `tests/integration/test_everos_extraction_real_llm.py` | yes (`real_llm`) |
 | L3 | backend ↔ everos e2e — embedded mode `store`/`recall` | `tests/integration/test_everos_backend_e2e.py` | yes (`real_llm`) |
 

docs/memory-plugin-architecture.md

@@ -323,8 +323,8 @@ place (raven's `pyproject.toml`). The upgrade surface is one line.
    written in adapter comments.
 3. **Test (all three layers)**:
    ```bash
-   uv run pytest tests/test_em1_skeleton.py tests/test_em2_backend.py \
-     tests/test_em3_http.py tests/test_memory_backend_protocol.py \
+   uv run pytest tests/test_everos_plugin_discovery.py tests/test_everos_backend.py \
+     tests/test_everos_http_adapter.py tests/test_memory_backend_protocol.py \
      tests/test_memory_backend_contract.py -q          # unit (mock adapter)
    uv run pytest tests/integration/test_everos_backend_e2e.py -m real_llm  # real
    python scripts/everos_memory_roundtrip.py            # native shell smoke
@@ -342,7 +342,7 @@ place (raven's `pyproject.toml`). The upgrade surface is one line.
 
 | Layer | Result |
 |---|---|
-| Unit (em1/em2/em3, protocol, contract, plugin discovery/command/tools, cl1, context, config, ag1/fb1, agent-loop pipeline) | 240 passed |
+| Unit (everos plugin discovery / backend / http adapter, protocol, contract, plugin command/tools, cli plugin stack, context, config, agent-loop backend dispatch + feedback, agent-loop pipeline) | 240 passed |
 | `raven plugins` | everos-memory · Source=`bundled` · Status=`activated` |
 | `real_llm` e2e (`test_everos_backend_e2e.py`) | 2 passed, 1 xfailed (best-effort skill-cluster check) — store→extract→recall + dual-track isolation |
 | roundtrip script (new import path) | OK; `users/user-raven/user.md` generated |

docs/sandbox/usage.md

@@ -632,7 +632,7 @@ On Linux without `/dev/kvm` the entire file is **automatically skipped** — no
 
 **First run — pre-pull OCI images:**
 
-A session-scoped fixture in `test_sandbox_integration.py` pre-pulls all required images
+A session-scoped fixture in `test_sandbox_real_vm.py` pre-pulls all required images
 (`ubuntu:22.04` and `node:20-slim`) before the first test. On a fast connection this takes
 ~30–60 s on first run and is instant on subsequent runs (images are cached by boxlite).
 
@@ -646,18 +646,18 @@ SKIPPED  OCI image pull failed for 'ubuntu:22.04' — likely a network issue, no
 **Run all integration tests:**
 
 ```bash
-uv run python -m pytest tests/test_sandbox_integration.py -v
+uv run python -m pytest tests/integration/test_sandbox_real_vm.py -v
 ```
 
 Expected output:
 
 ```
-tests/test_sandbox_integration.py::TestBoxliteExecutorIntegration::test_exec_echo              PASSED
-tests/test_sandbox_integration.py::TestBoxliteExecutorIntegration::test_exec_timeout          PASSED
-tests/test_sandbox_integration.py::TestBoxliteExecutorIntegration::test_exec_cwd              PASSED
-tests/test_sandbox_integration.py::TestBoxliteExecutorIntegration::test_volume_mount_file_visible_in_vm  PASSED
-tests/test_sandbox_integration.py::TestBoxliteExecutorIntegration::test_lifecycle_context_manager        PASSED
-tests/test_sandbox_integration.py::TestBoxliteStdioMCPRoundtrip::test_npx_mcp_server_everything          PASSED
+tests/integration/test_sandbox_real_vm.py::TestBoxliteExecutorIntegration::test_exec_echo              PASSED
+tests/integration/test_sandbox_real_vm.py::TestBoxliteExecutorIntegration::test_exec_timeout          PASSED
+tests/integration/test_sandbox_real_vm.py::TestBoxliteExecutorIntegration::test_exec_cwd              PASSED
+tests/integration/test_sandbox_real_vm.py::TestBoxliteExecutorIntegration::test_volume_mount_file_visible_in_vm  PASSED
+tests/integration/test_sandbox_real_vm.py::TestBoxliteExecutorIntegration::test_lifecycle_context_manager        PASSED
+tests/integration/test_sandbox_real_vm.py::TestBoxliteStdioMCPRoundtrip::test_npx_mcp_server_everything          PASSED
 
 6 passed in ~55s
 ```
@@ -670,13 +670,13 @@ on each run (~15 s), then starts the MCP server and validates the full `initiali
 **Run unit and integration tests together:**
 
 ```bash
-uv run python -m pytest tests/test_sandbox_unit.py tests/test_sandbox_integration.py -v
+uv run python -m pytest tests/test_sandbox_unit.py tests/integration/test_sandbox_real_vm.py -v
 ```
 
 **Run the full project test suite** (all test files, excluding integration):
 
 ```bash
-uv run python -m pytest tests/ --ignore=tests/test_sandbox_integration.py -q
+uv run python -m pytest tests/ --ignore=tests/integration/test_sandbox_real_vm.py -q
 ```
 
 ---
@@ -688,7 +688,7 @@ uv run python -m pytest tests/ --ignore=tests/test_sandbox_integration.py -q
 uv run python -m pytest "tests/test_sandbox_unit.py::TestBoxliteTranslateCwd::test_subdir_translates_correctly" -v
 
 # A single integration test
-uv run python -m pytest "tests/test_sandbox_integration.py::TestBoxliteStdioMCPRoundtrip::test_npx_mcp_server_everything" -v -s
+uv run python -m pytest "tests/integration/test_sandbox_real_vm.py::TestBoxliteStdioMCPRoundtrip::test_npx_mcp_server_everything" -v -s
 ```
 
 ---
@@ -768,5 +768,5 @@ python -c "from raven.sandbox import build_executor, SandboxConfig; print('sandb
 To verify end-to-end (requires KVM / Apple Silicon):
 
 ```bash
-uv run python -m pytest tests/test_sandbox_integration.py -v
+uv run python -m pytest tests/integration/test_sandbox_real_vm.py -v
 ```

tests/conftest.py

@@ -12,11 +12,12 @@
 def pytest_unconfigure(config: pytest.Config) -> None:
     """On CI, hard-exit past interpreter finalization once the run is over.
 
-    Native runtimes pulled in by the suite (lancedb's Rust/tokio thread, asyncio
-    subprocess transports finalized during GC) segfault Py_FinalizeEx on Linux,
-    turning a fully green run into exit 139. raven.cli._exit guards the CLI the
-    same way; the pytest process needs its own guard because it finalizes with
-    those runtimes live.
+    A fully green run still exited 139 on Linux: the suite finalizes with
+    native state live (asyncio subprocess transports collected during GC),
+    and Py_FinalizeEx segfaults on it, masking the recorded status. The CLI
+    routes its exit through the same helper, but on a different trigger --
+    see raven.cli._exit for the lancedb-specific gate it uses, which is not
+    what fires here.
 
     Local runs keep normal semantics so nothing masks an exit-time error, and
     the recorded status is preserved either way -- a failing run still exits

tests/integration/test_everos_backend_e2e.py

@@ -83,7 +83,11 @@ async def test_user_track_recall_through_backend(
             user_id=ids.user_id,
             top_k=5,
         )
-        assert isinstance(hits, list)
+        # Recall returning nothing is a best-effort miss per this module's
+        # strategy, but it must not read as a pass: the per-hit assertions below
+        # are vacuous on an empty list.
+        if not hits:
+            pytest.xfail("recall returned no hits; the per-hit assertions would be vacuous")
         for h in hits:
             assert isinstance(h, Memory)
             assert h.metadata.get("owner_type") == "user"

tests/test_ag1_backend_dispatch.py → tests/test_agent_loop_backend_dispatch.py

@@ -1,4 +1,4 @@
-"""AG-1 — AgentLoop ``backend`` wiring + ``_dispatch_backend_store``.
+"""AgentLoop ``backend`` wiring + ``_dispatch_backend_store``.
 
 The two after-turn callsites (system-message path + REPL path) now call
 :meth:`AgentLoop._dispatch_backend_store` as the third peer step in the
@@ -150,7 +150,7 @@ async def test_backend_exception_swallowed(
 
 
 # ---------------------------------------------------------------------------
-# Legacy compatibility — pre-AG-1 callsites still pass
+# Legacy compatibility -- callsites predating the backend keyword still pass
 # ---------------------------------------------------------------------------
 
 
@@ -159,7 +159,7 @@ def test_construction_without_backend_unchanged(
         self,
         tmp_path: Path,
     ) -> None:
-        """Pre-AG-1 construction (no ``backend=`` keyword) still works
+        """Construction without the ``backend=`` keyword still works
         end-to-end. After Phase B-3 the ``self.memory`` facade is gone;
         we now assert against the direct subsystem fields AgentLoop
         holds (``memory_consolidator`` + ``context.skills``)."""

tests/test_appworld_precheck.py

@@ -12,13 +12,11 @@
 from pathlib import Path
 from types import SimpleNamespace
 
-import pytest
-
 REPO_ROOT = Path(__file__).resolve().parents[1]
 if str(REPO_ROOT) not in sys.path:
     sys.path.insert(0, str(REPO_ROOT))
 
-httpx = pytest.importorskip("httpx")
+import httpx
 
 from benchmarks.appworld.evolve.precheck import (  # noqa: E402
     _endpoint_problem,

tests/test_auth_allowlist.py

@@ -10,7 +10,7 @@
 from __future__ import annotations
 
 import logging
-from dataclasses import dataclass
+from dataclasses import FrozenInstanceError, dataclass
 from typing import Any
 
 import pytest
@@ -215,5 +215,5 @@ def test_locked_field_membership(self):
 
     def test_frozen_dataclass(self):
         s = ManagedSettings()
-        with pytest.raises(Exception):  # FrozenInstanceError subclass
+        with pytest.raises(FrozenInstanceError):
             s.description = "mutated"  # type: ignore[misc]

tests/test_fb1_feedback_dispatch.py → tests/test_backend_feedback_dispatch.py

@@ -1,4 +1,4 @@
-"""FB-1 — qualified_id feedback dispatcher.
+"""qualified_id feedback dispatcher.
 
 Exercises:
 1. ``_filter_qualified_ids`` helper — prefix matching, native id

tests/test_cli_deep_research_commands.py

@@ -179,7 +179,7 @@ def test_configure_validation_fail_then_save(tmp_path: Path, monkeypatch):
 
 def test_configure_validation_fail_then_cancel(tmp_path: Path, monkeypatch):
     fail = {"ok": False, "status": "http_401", "model_ids": None, "error": "bad"}
-    p = _setup_interactive(monkeypatch, tmp_path, ["configure", "cancel"], validate=lambda *a, **k: fail)
+    _setup_interactive(monkeypatch, tmp_path, ["configure", "cancel"], validate=lambda *a, **k: fail)
     assert configure_deep_research(non_interactive=False, warnings=[]) is False  # cancelled, nothing written
 
 
@@ -231,9 +231,7 @@ def _flaky(*a, **k):
         calls["n"] += 1
         return {"ok": calls["n"] > 1, "status": "http_401" if calls["n"] == 1 else "ok", "model_ids": [], "error": None}
 
-    p = _setup_interactive(
-        monkeypatch, tmp_path, ["configure", "retry", "mirothinker-1-7-deepresearch"], validate=_flaky
-    )
+    _setup_interactive(monkeypatch, tmp_path, ["configure", "retry", "mirothinker-1-7-deepresearch"], validate=_flaky)
     assert configure_deep_research(non_interactive=False, warnings=[]) is True
     assert calls["n"] == 2  # first validate failed, retry validated ok