Implemented the next parity slice.

New runtime/code:

  - src/ask_user_runtime.py
  - src/team_runtime.py

  New real tools in src/agent_tools.py:

  - ask_user_question
  - team_create
  - team_delete
  - team_list
  - team_get
  - send_message
  - team_messages
  - notebook_edit
This commit is contained in:
Abdelrahman Abdallah
2026-04-07 02:51:30 +02:00
parent a54c90b18f
commit a5629295ac
21 changed files with 1886 additions and 24 deletions
+52
View File
@@ -7,6 +7,7 @@ from pathlib import Path
from benchmarks.run_terminal_bench_local import (
TerminalBenchTask,
build_host_agent_command,
build_verifier_exec_command,
discover_tasks,
filter_tasks,
parse_dockerfile_workdir,
@@ -129,6 +130,57 @@ docker_image = "example/other:latest"
self.assertIn("instruction=$(cat", cmd)
self.assertNotIn("claw-code-agent agent", cmd)
def test_build_verifier_command_fakeroot_adds_flags(self) -> None:
with tempfile.TemporaryDirectory() as tmp_dir:
root = Path(tmp_dir)
workspace_dir = root / "workspace"
verifier_logs_dir = root / "verifier"
task_dir = root / "task"
tests_dir = task_dir / "tests"
workspace_dir.mkdir()
verifier_logs_dir.mkdir()
tests_dir.mkdir(parents=True)
image_path = root / "image.sif"
image_path.write_text("fake", encoding="utf-8")
task = TerminalBenchTask(
task_dir=task_dir,
name="terminal-bench/demo",
short_name="demo",
instruction="solve it",
docker_image="example/demo:latest",
agent_timeout_sec=30.0,
verifier_timeout_sec=30.0,
workdir="/workspace",
has_docker_compose=False,
)
cmd_no_fakeroot = build_verifier_exec_command(
task=task,
image_path=image_path,
workspace_dir=workspace_dir,
task_dir=task_dir,
verifier_logs_dir=verifier_logs_dir,
env={},
fakeroot=False,
)
self.assertNotIn("--fakeroot", cmd_no_fakeroot)
self.assertNotIn("--writable-tmpfs", cmd_no_fakeroot)
cmd_fakeroot = build_verifier_exec_command(
task=task,
image_path=image_path,
workspace_dir=workspace_dir,
task_dir=task_dir,
verifier_logs_dir=verifier_logs_dir,
env={},
fakeroot=True,
)
self.assertIn("--fakeroot", cmd_fakeroot)
self.assertIn("--writable-tmpfs", cmd_fakeroot)
self.assertIn("--contain", cmd_fakeroot)
self.assertIn("TMPDIR", cmd_fakeroot)
self.assertIn("CURL_CA_BUNDLE", cmd_fakeroot)
if __name__ == "__main__":
unittest.main()