Repo skeleton, pinned environment, free smoke test

Public from commit one, so no key ever enters this history.

compose.yaml is ImpossibleBench's, plus working_dir: /workspace. The image has no
WORKDIR, so inspect resolves it to "/" and the task files land at the filesystem root
among twenty-odd entries. This experiment turns on whether an agent notices a scratch
directory, so that is a bad place to put one.

ImpossibleBench installs with --no-deps to keep the swebench tree out; datasets is
declared here instead because hf_dataset genuinely needs it.

Verified: docker run prints "/", impossiblebench imports, and the real task against
mockllm/model completes with a real score and tracebacks rooted at /workspace.

Claude-Session: https://claude.ai/code/session_01Cq98H7sNoSJdL3W98f18bu

scripts/smoke_mock.py

@@ -0,0 +1,35 @@
+"""Free end-to-end smoke test: the real task, a fake model, no money.
+
+Runs the unmodified ImpossibleBench LiveCodeBench task against mockllm/model. The mock
+never calls a tool, so it burns the message limit and falls through to the scorer. That
+still exercises everything except the model: container start under our compose.yaml, the
+writes of func.py and test.py, the scorer's test-file comparison, and `python test.py`.
+
+Run from the repo root so inspect finds compose.yaml (it looks in the process working
+directory). scripts/ recipes in the justfile do that for you.
+"""
+
+from inspect_ai import eval as inspect_eval
+from impossiblebench import impossible_livecodebench
+
+if __name__ == "__main__":
+    logs = inspect_eval(
+        impossible_livecodebench(
+            split="conflicting",
+            agent_type="tools",
+            sandbox="docker",
+            limit=1,
+            max_attempts=1,
+            message_limit=4,
+        ),
+        model="mockllm/model",
+        log_dir="./logs/smoke",
+    )
+
+    log = logs[0]
+    print(f"\nstatus: {log.status}")
+    if log.error:
+        print(f"error: {log.error}")
+    for sample in log.samples or []:
+        print(f"sample {sample.id}: score={sample.scores}")
+        print(f"  working dir sanity: {sample.metadata.get('task_id')}")