M2.8: switch over to the new loop

nisse-dev now runs only the conversational loop. The gated state machine
it was built alongside is gone:

- Removed control.py, planner.py, executor.py, approvals.py, gates.py
  (command_form moves to loop/reader.py), publish.py and poller.py, the
  old assembly in agent.py (its shared helpers move to loop/build.py),
  the store's state-machine view (get_state, transition), their tests,
  the old end-to-end and live planning/milestone/gate tests, and
  specs/orchestrator.qnt with its scenarios.
- Roles are driver, helper, expert and classifier; driver is required.
  loop:, roles.planner, roles.executor and tuning.executor are refused
  with what to use instead; tuning.driver takes the turn cap and failure
  limit.
- The benchmark and the capability matrix measure the driver: the
  loop's own system prompt and tools, the task's plan as the agreed
  restatement. Task plans end with finish_milestone, so earlier runs
  show as (old); stored runs' executor tuning reads as the driver's.
- A delegation on the loop records what its helper's alias resolved to
  (L9), as the old path's did.
- The dump shows a work item's revision and what it waits for.
- Behaviour the old tests covered and the loop shares stays covered:
  driver runs that fail on a model call, stop mid-turn, reset the
  failure count; go-aheads ignoring case and punctuation; a request too
  big to fit is never sent; the assembly tests, now on build_loop_agent.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPQ14L3LikyefCMFBqEbKE

nisse-dev/tests/test_bench.py

@@ -1,4 +1,4 @@
-"""The executor benchmark."""
+"""The driver benchmark."""
 
 import json
 import shutil
@@ -19,7 +19,7 @@ from nisse_dev.bench.runner import MeteringClient, run_task
 from nisse_dev.bench.tasks import CHECK_DIR, load_tasks
 from nisse_dev.canonical import AssistantText, ToolCall
 from nisse_dev.catalog import ModelFacts
-from nisse_dev.config import ExecutorTuning, Tuning
+from nisse_dev.config import DriverTuning, Tuning
 from nisse_dev.execution import SSHHost
 
 # The build VM for every test here: the session's throwaway sshd.
@@ -79,7 +79,7 @@ def test_the_seed_fails_its_check_and_the_solution_passes(task, tmp_path: Path)
 def test_the_hidden_check_is_not_in_the_seed(task, tmp_path: Path) -> None:
     task.seed_into(tmp_path)
     assert not (tmp_path / CHECK_DIR).exists()
-    assert "Call finish" in task.plan  # plans end the way the executor must
+    assert "Call finish_milestone" in task.plan  # as the driver must end
 
 
 def test_long_output_really_is_long(tmp_path: Path) -> None:
@@ -192,7 +192,7 @@ SOLVE_SLUGIFY = [
         )
     ],
     [call("run_command", "r", command="python -c 'import slugify'")],
-    [call("finish", "f", summary="added slugify")],
+    [call("finish_milestone", "f", summary="added slugify", final="yes")],
 ]
 
 
@@ -208,7 +208,7 @@ def test_a_solved_task_is_a_success(tmp_path: Path) -> None:
     assert record.tokens_in == 3000
     assert record.task_hash == task.content_hash
     assert record.model_info["supports_tool_choice"] is True
-    assert record.tuning["executor"]["max_turns"] == 40
+    assert record.tuning["driver"]["max_turns"] == 40
 
 
 def test_saying_finished_is_not_success(tmp_path: Path) -> None:
@@ -223,7 +223,7 @@ def test_saying_finished_is_not_success(tmp_path: Path) -> None:
                 content="def slugify(t):\n    return t\n",
             )
         ],
-        [call("finish", "f", summary="done")],
+        [call("finish_milestone", "f", summary="done", final="yes")],
     ]
     record = run_task(task, FACTS, Scripted(wrong), Tuning(), host=HOST)
     assert record.outcome == "completed"
@@ -239,7 +239,7 @@ def test_misbehaviour_is_counted(tmp_path: Path) -> None:
     ]
     # A higher failure limit, so the invented tool and the failed command
     # don't escalate before the prose turn is reached.
-    tuning = Tuning(executor=ExecutorTuning(failure_limit=5))
+    tuning = Tuning(driver=DriverTuning(failure_limit=5))
     record = run_task(task, FACTS, Scripted(turns), tuning, host=HOST)
     assert record.invented_tools == 1
     assert record.failed_commands == 1
@@ -247,6 +247,31 @@ def test_misbehaviour_is_counted(tmp_path: Path) -> None:
     assert not record.success
 
 
+def test_asking_the_operator_is_an_escalation(tmp_path: Path) -> None:
+    [task] = load_tasks(["write-function"])
+    asks = [[call("ask_human", "q", question="which file?")]]
+    record = run_task(task, FACTS, Scripted(asks), Tuning(), host=HOST)
+    assert (record.outcome, record.success) == ("escalated", False)
+
+
+def test_the_driver_gets_the_loops_own_prompt_and_tools(tmp_path: Path) -> None:
+    from nisse_dev.loop.agent import SYSTEM
+
+    seen: list[tuple[str, set[str]]] = []
+
+    class Spy(Scripted):
+        def complete(self, events, tools, model, system=None):  # type: ignore[no-untyped-def]
+            seen.append((system or "", {t["function"]["name"] for t in tools}))
+            return super().complete(events, tools, model, system)
+
+    [task] = load_tasks(["write-function"])
+    run_task(task, FACTS, Spy(list(SOLVE_SLUGIFY)), Tuning(), host=HOST)
+    system, tools = seen[0]
+    assert system.startswith(SYSTEM.split("{")[0])
+    assert {"start_milestone", "finish_milestone", "run_command"} <= tools
+    assert not {"delegate", "ask_expert", "restate"} & tools
+
+
 def test_a_fallback_answer_can_never_be_a_success(tmp_path: Path) -> None:
     [task] = load_tasks(["write-function"])
     client = Scripted(list(SOLVE_SLUGIFY), answered="claude-sonnet-5")
@@ -383,11 +408,11 @@ def test_results_stored_before_waits_were_recorded_still_load() -> None:
 def test_runs_with_different_tuning_are_kept_apart() -> None:
     current = {t.name: t.content_hash for t in TASKS}
     capped = record_json()
-    capped["tuning"]["executor"]["max_turns"] = 7
+    capped["tuning"]["driver"]["max_turns"] = 7
     rows = summarise([record_json(), capped], current)
-    assert sorted(row.tuning for row in rows) == ["", "executor.max_turns=7"]
+    assert sorted(row.tuning for row in rows) == ["", "driver.max_turns=7"]
     table = render(rows)
-    assert "executor.max_turns=7" in table
+    assert "driver.max_turns=7" in table
     assert "tuning = settings that differ" in table
 
 
@@ -395,3 +420,12 @@ def test_a_report_at_default_tuning_has_no_tuning_column() -> None:
     current = {t.name: t.content_hash for t in TASKS}
     table = render(summarise([record_json()], current))
     assert "tuning" not in table
+
+
+def test_runs_from_before_the_driver_read_as_its_defaults() -> None:
+    # Stored before M2.8, the driver's section was called executor.
+    current = {t.name: t.content_hash for t in TASKS}
+    old = record_json()
+    old["tuning"]["executor"] = old["tuning"].pop("driver")
+    [row] = summarise([old], current)
+    assert row.tuning == ""