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

nisse-dev/src/nisse_dev/bench/runner.py

@@ -1,11 +1,13 @@
 """Running one task once with one driver model, and recording it."""
 
+import contextlib
 import io
 import shlex
 import subprocess
 import tempfile
 import time
 import uuid
+from collections.abc import Iterator
 from dataclasses import asdict, dataclass, field
 from datetime import UTC, datetime
 from pathlib import Path
@@ -21,6 +23,7 @@ from nisse_core.log.canonical import (
     UserMessage,
 )
 from nisse_core.loop.driver import Driver, RunOutcome
+from nisse_core.models import calls
 from nisse_core.models.catalog import ModelFacts
 from nisse_core.models.compact import Compactor
 from nisse_core.oplog import OpLog
@@ -184,6 +187,37 @@ def _count(events: list[StoredEvent]) -> tuple[int, int]:
     return failed, invented
 
 
+@contextlib.contextmanager
+def _run_dir(keep: Path | None, workdir: Path | None, name: str) -> Iterator[Path]:
+    """Where a run's state store and call record go: kept under ``keep``
+    for the status page (X6), or a temporary directory."""
+    if keep is not None:
+        root = keep / f"{name}-{uuid.uuid4().hex[:8]}"
+        root.mkdir(parents=True)
+        yield root
+        return
+    with tempfile.TemporaryDirectory(dir=workdir) as tmp:
+        yield Path(tmp)
+
+
+@contextlib.contextmanager
+def _recording(client: Any, root: Path) -> Iterator[None]:
+    """The run's calls recorded beside its own state store (L7), as the
+    agent records its own, so the run measures what the agent would do."""
+    recorder = calls.CallRecorder(calls.path_for(root / "state.db"))
+    had = hasattr(client, "call_recorder")
+    previous = getattr(client, "call_recorder", None)
+    client.call_recorder = recorder
+    try:
+        yield
+    finally:
+        if had:
+            client.call_recorder = previous
+        else:
+            del client.call_recorder
+        recorder.close()
+
+
 def run_task(
     task: Task,
     facts: ModelFacts,
@@ -194,6 +228,7 @@ def run_task(
     note: str = "",
     commit: str = "",
     workdir: Path | None = None,
+    keep: Path | None = None,
 ) -> RunRecord:
     """Run one task once: seed a workspace, hand the driver an agreed item
     whose restatement is the task's plan, then judge the result with the
@@ -202,8 +237,7 @@ def run_task(
     It is the loop's own driver, system prompt and tools, so what is measured
     is what the agent would do, without the harness around it: the milestone
     is started but no branch is cut, and finishing it ends the run."""
-    with tempfile.TemporaryDirectory(dir=workdir) as tmp:
-        root = Path(tmp)
+    with _run_dir(keep, workdir, task.name) as root, _recording(client, root):
         # The workspace is on the build VM, like the agent's own (L13).
         workspace = Workspace(
             host,