Merge pull request 'The call record: every model call, for the operator's study (M6.6a)' (#179) from call-record into main

nisse-dev/tests/test_bench.py

@@ -467,3 +467,26 @@ def test_runs_from_before_the_driver_read_as_its_defaults() -> None:
     old["tuning"]["executor"] = old["tuning"].pop("driver")
     [row] = summarise([old], current)
     assert row.tuning == ""
+
+
+def test_a_kept_run_leaves_its_state_and_call_record(tmp_path: Path) -> None:
+    # As the agent records itself (L7), so the status page can open a
+    # benchmark run (X6); each run gets its own record, and the client is
+    # left as it was.
+    from nisse_core.models.calls import CallReader
+
+    [task] = load_tasks(["write-function"])
+    client = Scripted(list(SOLVE_SLUGIFY))
+    record = run_task(task, FACTS, client, Tuning(), host=HOST, keep=tmp_path / "runs")
+
+    [kept] = list((tmp_path / "runs").iterdir())
+    assert kept.name.startswith("write-function-")
+    assert (kept / "state.db").exists()
+    [run] = CallReader(kept / "state.calls.db").runs()
+    assert (run["work_item"], run["role"], run["outcome"]) == (
+        "bench",
+        "driver",
+        "wait",
+    )
+    assert record.success
+    assert not hasattr(client, "call_recorder")

nisse-dev/tests/test_config.py

@@ -573,3 +573,28 @@ def test_the_vault_is_read() -> None:
 def test_a_bad_vault_is_refused(vault: dict[str, Any], said: str) -> None:
     with pytest.raises(ConfigError, match=re.escape(said)):
         build(vault=vault)
+
+
+def test_the_call_record_is_on_unless_turned_off() -> None:
+    assert build().call_record is True
+    assert build(call_record=False).call_record is False
+    with pytest.raises(ConfigError, match="call_record"):
+        build(call_record="no")
+
+
+def test_both_routes_record_into_one_file_beside_the_state_store(
+    tmp_path: Path,
+) -> None:
+    from nisse_core.models.catalog import ModelCatalog
+    from nisse_dev.loop.build import make_llm_clients
+
+    normalized, passthrough = make_llm_clients(
+        build(state_db=str(tmp_path / "state.db")), ModelCatalog({})
+    )
+    assert normalized.call_recorder is passthrough.call_recorder
+    assert normalized.call_recorder.path == tmp_path / "state.calls.db"
+    off, _ = make_llm_clients(
+        build(state_db=str(tmp_path / "other.db"), call_record=False), ModelCatalog({})
+    )
+    assert off.call_recorder is None
+    assert not (tmp_path / "other.calls.db").exists()

nisse-dev/tests/test_delegation.py

@@ -702,3 +702,25 @@ def test_a_bad_request_is_not_given_the_benefit_of_the_doubt(
     result = build(BadRequestClient([]), store, ws, repo).run(brief="b", branch="d")
     assert result.status == "failed"
     assert not result.interrupted
+
+
+def test_a_delegation_is_recorded_as_a_run_of_its_own(
+    store: StateStore, ws: Workspace, repo: GitRepo
+) -> None:
+    # Its calls are the helper's turns (L7), inside the run that delegated.
+    from nisse_core.models import calls
+
+    seen: list[calls.Scope] = []
+
+    class Watching(ScriptedClient):
+        def complete(self, *args: Any, **kwargs: Any) -> list[CanonicalEvent]:
+            seen.append(calls.current())
+            return super().complete(*args, **kwargs)
+
+    client = Watching([[hand_back(succeeded=False, changed="nothing")]])
+    with calls.run(None, work_item=ITEM, role="driver"):
+        outer = calls.current().run
+        build(client, store, ws, repo).run(brief="look into it", branch="d1")
+    [scope] = seen
+    assert (scope.role, scope.work_item, scope.parent) == ("helper", ITEM, outer)
+    assert scope.run not in (None, outer)

nisse-dev/tests/test_loop_classifier.py

@@ -346,3 +346,27 @@ def test_the_export_script_reads_a_store(tmp_path: Path) -> None:
     lines = [json.loads(line) for line in run.stdout.splitlines()]
     assert [line["verdict"] for line in lines] == ["needs_discussion", "safe"]
     assert lines[0]["input"]["action"]["arguments"] == {"command": "x"}
+
+
+def test_the_classifiers_calls_are_recorded_as_its_own() -> None:
+    # In the call record (L7), a verdict is the classifier's call, not a
+    # turn of the run that asked for it.
+    from nisse_core.models import calls
+
+    seen: list[calls.Scope] = []
+
+    class Watching(Judge):
+        def complete(self, events, tools, model, system=None) -> list[Any]:
+            seen.append(calls.current())
+            return super().complete(events, tools, model, system)
+
+    classifier = ActionClassifier(Watching(verdict("contained")), "local")  # type: ignore[arg-type]
+    with calls.run(None, work_item="w", role="driver"):
+        classifier.judge(Action("run_command", {"command": "make"}), SCOPE, "/w")
+    [scope] = seen
+    assert (scope.role, scope.purpose, scope.work_item) == (
+        "classifier",
+        "classification",
+        "w",
+    )
+    assert scope.run is not None

nisse-dev/tests/test_loop_labels.py

@@ -316,3 +316,19 @@ def test_a_restatement_that_asks_is_sent_back(tmp_path: Path) -> None:
         if hasattr(e.event, "result")
     ]
     assert any("a restatement asks no questions" in r for r in results)
+
+
+def test_the_go_ahead_readers_calls_are_recorded_as_the_classifiers() -> None:
+    from nisse_core.models import calls
+
+    seen: list[calls.Scope] = []
+
+    class Watching(_Model):
+        def complete(self, *args: Any, **kwargs: Any) -> Any:
+            seen.append(calls.current())
+            return super().complete(*args, **kwargs)
+
+    yes = [ToolCall(call_id="v", tool="verdict", arguments={"answer": "yes"})]
+    GoAheadReader(Watching(yes), "m").agrees("add count_words", "ok")  # type: ignore[arg-type]
+    [scope] = seen
+    assert (scope.role, scope.purpose) == ("classifier", "go-ahead reading")