Benchmark: the seed is a git repository's first commit

The agent works in a clone, but benchmark workspaces were plain
directories, so a model that starts with `git log` failed two commands
in a row and was stopped as stuck within seconds (Flash-Next, 10 of 13
failures on 2026-10-24). The seed is now committed on main, locally and
on the build VM, and how a workspace is set up is part of each task's
hash, so earlier runs are reported as not comparable.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J5Utyfzw7gF7UhjRDEtjTk

nisse-dev/tests/test_bench.py

@@ -151,6 +151,33 @@ def test_a_changed_task_gets_a_new_hash(tmp_path: Path) -> None:
     assert copy.content_hash != task.content_hash
 
 
+def test_the_seed_is_a_git_repository_with_one_clean_commit(tmp_path: Path) -> None:
+    # The agent works in a clone; a model that starts with `git log` must not
+    # fail here where it would succeed in production.
+    import subprocess
+
+    [task] = load_tasks(["write-function"])
+    task.seed_into(tmp_path)
+
+    def git(*args: str) -> str:
+        return subprocess.run(
+            ["git", *args], cwd=tmp_path, capture_output=True, text=True, check=True
+        ).stdout
+
+    assert git("log", "--oneline").count("\n") == 1
+    assert git("status", "--short") == ""
+
+
+def test_how_the_seed_is_set_up_is_part_of_the_hash(monkeypatch) -> None:
+    # Runs from before the seed was committed are not comparable.
+    import nisse_dev.bench.tasks as tasks_module
+
+    [task] = load_tasks(["write-function"])
+    before = task.content_hash
+    monkeypatch.setattr(tasks_module, "SEED_SETUP", "seed copied")
+    assert task.content_hash != before
+
+
 def test_an_unknown_task_is_refused() -> None:
     with pytest.raises(ValueError, match="unknown task"):
         load_tasks(["nope"])
@@ -211,6 +238,17 @@ def test_a_solved_task_is_a_success(tmp_path: Path) -> None:
     assert record.tuning["driver"]["max_turns"] == 40
 
 
+def test_the_model_can_read_the_history_on_the_build_vm(tmp_path: Path) -> None:
+    [task] = load_tasks(["write-function"])
+    turns = [
+        [call("run_command", "g", command="git log --oneline && git status --short")],
+        *SOLVE_SLUGIFY,
+    ]
+    record = run_task(task, FACTS, Scripted(turns), Tuning(), host=HOST, note="n")
+    assert record.success
+    assert record.failed_commands == 0
+
+
 def test_saying_finished_is_not_success(tmp_path: Path) -> None:
     # Success is the hidden check passing, never the model's own claim.
     [task] = load_tasks(["write-function"])