Method setup

benchmarks/terminal_bench/mux_agent.py:287–325

Stage the mux payload before installing it in the task environment.

(self, environment: BaseEnvironment)

            )

    async def setup(self, environment: BaseEnvironment) -> None:
        """Stage the mux payload before installing it in the task environment."""
        env = self._env

        # Harbor no longer renders installed-agent templates for custom agents.
        # Stage the rendered script ourselves so scheduled tbench runs are not
        # coupled to Harbor internals that have changed over time.
        await environment.exec(command="mkdir -p /installed-agent", user="root")

        if not self._archive_bytes:
            self._archive_bytes = build_app_archive(
                self._repo_root, self._INCLUDE_PATHS
            )

        archive_path = self.logs_dir / self._ARCHIVE_NAME
        archive_path.write_bytes(self._archive_bytes)
        await environment.upload_file(
            source_path=archive_path,
            target_path=f"/installed-agent/{self._ARCHIVE_NAME}",
        )

        await environment.upload_file(
            source_path=self._runner_path,
            target_path=f"/installed-agent/{self._RUNNER_NAME}",
        )

        await environment.upload_file(
            source_path=self._write_setup_script(),
            target_path=f"/installed-agent/{self._SETUP_SCRIPT_NAME}",
        )

        await self.install(environment)

        # Optionally seed the sandbox with providers.jsonc from the host machine.
        # This is required for OAuth-only configs where env var API keys are absent.
        await self._stage_providers_config(environment, env)

        # Store environment reference for token extraction later.
        self._last_environment = environment

    def create_run_agent_commands(self, instruction: str) -> list[_AgentCommand]:
        escaped = shlex.quote(instruction)