# Conflicts: # apps/desktop/e2e/archived-hidden-session-recoverable.spec.ts # apps/desktop/e2e/bot-chat-message-agent-friendly-name.spec.ts # apps/desktop/e2e/bot-mailbox-unreadable-ticket.spec.ts # apps/desktop/e2e/bot-mode-roster-localized.spec.ts # apps/desktop/e2e/bot-mode-row-click-mirrors-registry.spec.ts # apps/desktop/e2e/bot-mode-tab-shows-bot-name.spec.ts # apps/desktop/e2e/bot-roster-group-row-organisation.spec.ts # apps/desktop/e2e/bot-roster-ignores-infra-dirs.spec.ts # apps/desktop/e2e/bot-roster-timestamp-meta.spec.ts # apps/desktop/e2e/bot-roster-user-sections.spec.ts # apps/desktop/e2e/bot-routines-pane-narrow.spec.ts # apps/desktop/e2e/bot-row-open-recent-session.spec.ts # apps/desktop/e2e/bot-tile-ignores-ambient-composer-model.spec.ts # apps/desktop/e2e/group-composer-auto-grow.spec.ts # apps/desktop/e2e/group-create-gate-remote-roster.spec.ts # apps/desktop/e2e/group-prompt-renamed-primary-handle.spec.ts # apps/desktop/e2e/hosted-room-backend-continuity.spec.ts # apps/desktop/e2e/hosted-room-legacy-store-migration.spec.ts # apps/desktop/e2e/settings-scope-chips-bot-title.spec.ts # apps/desktop/e2e/worktree-branch-status.spec.ts # apps/desktop/electron/backend-probes.test.ts # apps/desktop/electron/connection-apply.test.ts # apps/desktop/electron/desktop-electron-pin.test.ts # apps/desktop/electron/desktop-uninstall.test.ts # apps/desktop/electron/gateway-file-download-transport.test.ts # apps/desktop/electron/gateway-stop-before-update.test.ts # apps/desktop/electron/github-api-auth.test.ts # apps/desktop/electron/registry-primary-profile-scope.test.ts # apps/desktop/electron/update-api-check.test.ts # apps/desktop/electron/update-handoff-marker.test.ts # apps/desktop/electron/venv-blocker-scan.test.ts # apps/desktop/scripts/after-extract.test.mjs # apps/desktop/scripts/local-pack-publish.test.mjs # apps/desktop/scripts/tasks-scroll.test.mjs # apps/desktop/src/app/settings/model-settings.test.tsx # apps/desktop/src/app/updates-overlay.blockers.test.tsx # apps/desktop/src/components/desktop-install-overlay.test.tsx # apps/desktop/src/lib/update-copy.test.ts # scripts/ci/check_os_marker_fakes.py # tests-js/desktop-mac-usage-descriptions.test.ts # tests-js/node-engine-alignment.test.ts # tests/agent/lsp/test_install_and_lint_fixes.py # tests/agent/test_command_token_source.py # tests/agent/test_compression_boundary_hook.py # tests/agent/test_create_openai_client_ssl_verify.py # tests/agent/test_custom_provider_ca_probes.py # tests/agent/test_endpoint_blackhole.py # tests/agent/test_estimator_parity.py # tests/agent/test_in_place_compaction.py # tests/agent/test_moa_loop_mode.py # tests/agent/test_model_metadata.py # tests/agent/test_skill_session_platform_gate.py # tests/agent/test_skill_utils.py # tests/agent/test_ssl_ca_guard.py # tests/computer_use/test_doctor.py # tests/cron/test_codex_execution_paths.py # tests/cron/test_cron_bot_chat_delivery.py # tests/cron/test_cron_script.py # tests/cron/test_media_delivery_parity.py # tests/cron/test_misfire_catchup.py # tests/cron/test_parallel_pool.py # tests/cron/test_recurring_eagain_redispatch.py # tests/gateway/test_choice_picker.py # tests/gateway/test_control_socket_windows_live.py # tests/gateway/test_dingtalk.py # tests/gateway/test_feishu.py # tests/gateway/test_feishu_onboard.py # tests/gateway/test_gateway_shutdown.py # tests/gateway/test_matrix.py # tests/gateway/test_model_command_custom_providers.py # tests/gateway/test_reasoning_command.py # tests/gateway/test_runtime_footer.py # tests/gateway/test_session.py # tests/gateway/test_session_hygiene.py # tests/gateway/test_status.py # tests/gateway/test_teams.py # tests/gateway/test_turn_lease.py # tests/gateway/test_whatsapp_connect.py # tests/hermes_cli/test_approvals_command.py # tests/hermes_cli/test_auth_store_lock_concurrent.py # tests/hermes_cli/test_backup.py # tests/hermes_cli/test_banner_git_state.py # tests/hermes_cli/test_certifi_repair.py # tests/hermes_cli/test_cmd_update.py # tests/hermes_cli/test_compat_manifest_targets.py # tests/hermes_cli/test_computer_use_cli.py # tests/hermes_cli/test_cpr_local_leak.py # tests/hermes_cli/test_dashboard_auth_gate.py # tests/hermes_cli/test_dashboard_procs_kill_grace.py # tests/hermes_cli/test_desktop_lifecycle_windows_live.py # tests/hermes_cli/test_doctor.py # tests/hermes_cli/test_doctor_command_install.py # tests/hermes_cli/test_fleet_config_migration_windows_live.py # tests/hermes_cli/test_gateway.py # tests/hermes_cli/test_gateway_platform_gating.py # tests/hermes_cli/test_gateway_restart_loop.py # tests/hermes_cli/test_gateway_task_probe.py # tests/hermes_cli/test_gateway_wsl.py # tests/hermes_cli/test_gui_command.py # tests/hermes_cli/test_install_cua_driver.py # tests/hermes_cli/test_kanban_db.py # tests/hermes_cli/test_lazy_command_exports.py # tests/hermes_cli/test_lazy_refresh_venv_repair.py # tests/hermes_cli/test_linux_desktop_entry.py # tests/hermes_cli/test_local_runtime.py # tests/hermes_cli/test_local_runtime_updates.py # tests/hermes_cli/test_managed_uv.py # tests/hermes_cli/test_mcp_reload_confirm_gate.py # tests/hermes_cli/test_nous_subscription.py # tests/hermes_cli/test_npm_engine.py # tests/hermes_cli/test_personality_none.py # tests/hermes_cli/test_pet_toggle.py # tests/hermes_cli/test_plan_reconciliation_windows_live.py # tests/hermes_cli/test_plugin_event_bus.py # tests/hermes_cli/test_plugin_manifest_v2.py # tests/hermes_cli/test_plugin_packs.py # tests/hermes_cli/test_plugins_cmd.py # tests/hermes_cli/test_plugins_cmd_enable_disable_nested.py # tests/hermes_cli/test_process_identity.py # tests/hermes_cli/test_profiles.py # tests/hermes_cli/test_profiles_sidebar_cache.py # tests/hermes_cli/test_pty_bridge.py # tests/hermes_cli/test_resolve_turn_limit.py # tests/hermes_cli/test_serve_runtime_inventory.py # tests/hermes_cli/test_session_vacuum_config.py # tests/hermes_cli/test_set_config_value.py # tests/hermes_cli/test_signal_handler_kanban_worker.py # tests/hermes_cli/test_slash_confirm_windows.py # tests/hermes_cli/test_stale_pid_guard.py # tests/hermes_cli/test_startup_fast_guards.py # tests/hermes_cli/test_status.py # tests/hermes_cli/test_telegram_managed_bot.py # tests/hermes_cli/test_tools_config.py # tests/hermes_cli/test_update_apply_shallow_count.py # tests/hermes_cli/test_update_autostash.py # tests/hermes_cli/test_update_concurrent_quarantine.py # tests/hermes_cli/test_update_fetch_failure_classifier.py # tests/hermes_cli/test_update_fleet_probe_resume_token.py # tests/hermes_cli/test_update_handoff_backend_reap.py # tests/hermes_cli/test_update_handoff_desktop_rebuild.py # tests/hermes_cli/test_update_head_moved_gate.py # tests/hermes_cli/test_update_host_obligation.py # tests/hermes_cli/test_update_import_guard.py # tests/hermes_cli/test_update_interrupted_recovery.py # tests/hermes_cli/test_update_inventory.py # tests/hermes_cli/test_update_launchd_unloaded_gateway.py # tests/hermes_cli/test_update_missing_configured_deps.py # tests/hermes_cli/test_update_modified_notice.py # tests/hermes_cli/test_update_multiplex_migration_hook.py # tests/hermes_cli/test_update_no_gateway_restart.py # tests/hermes_cli/test_update_orphan_backend_reap.py # tests/hermes_cli/test_update_parked_branch_guard.py # tests/hermes_cli/test_update_post_pull_syntax_guard.py # tests/hermes_cli/test_update_receipt.py # tests/hermes_cli/test_update_self_lock.py # tests/hermes_cli/test_update_shim_fail_closed.py # tests/hermes_cli/test_update_shim_self_lock.py # tests/hermes_cli/test_update_sqlite_remediation.py # tests/hermes_cli/test_update_stale_dashboard.py # tests/hermes_cli/test_update_stale_virtualenv.py # tests/hermes_cli/test_update_venv_health.py # tests/hermes_cli/test_update_venv_ownership_preflight.py # tests/hermes_cli/test_update_wedged_gateway.py # tests/hermes_cli/test_update_yes_flag.py # tests/hermes_cli/test_update_zip_two_phase.py # tests/hermes_cli/test_urllib_security.py # tests/hermes_cli/test_ux_messages_auth_config.py # tests/hermes_cli/test_ux_messages_startup.py # tests/hermes_cli/test_venv_holder_classifier.py # tests/hermes_cli/test_verify_console_scripts.py # tests/hermes_cli/test_verify_core_dependencies.py # tests/hermes_cli/test_web_server.py # tests/hermes_cli/test_web_server_console_ws.py # tests/hermes_cli/test_web_server_ws_ping.py # tests/hermes_cli/test_web_ui_build.py # tests/hermes_state/test_fts_rebuild_admission.py # tests/hermes_state/test_hermes_state.py # tests/plugins/memory/test_memory_lazy_install.py # tests/plugins/test_google_meet_plugin.py # tests/plugins/test_langfuse_plugin.py # tests/plugins/test_security_guidance_plugin.py # tests/plugins/test_transform_llm_output_hook.py # tests/scripts/desktop_update/test_desktop_update_windows_gateway_flag.py # tests/scripts/desktop_update/test_desktop_update_windows_python_handoff.py # tests/scripts/desktop_update/test_desktop_update_windows_timestamp.py # tests/scripts/install/test_install_clone_throttle_fallback.py # tests/scripts/install/test_install_lockfile_churn.py # tests/scripts/install/test_install_no_initial_commit.py # tests/scripts/install/test_install_sh_browser_install.py # tests/scripts/install/test_install_sh_node_prerelease.py # tests/scripts/install/test_install_sh_symlink_stomp.py # tests/scripts/install/test_install_sh_uv_lock_config.py # tests/scripts/install/test_install_unmerged_index.py # tests/scripts/test_contributor_map.py # tests/scripts/test_run_tests_parallel.py # tests/skills/test_competitor_news_monitor_skill.py # tests/skills/test_document_to_action_items_skill.py # tests/skills/test_google_workspace_setup.py # tests/skills/test_google_workspace_setup_deps.py # tests/skills/test_grounded_citations_skill.py # tests/skills/test_ip_as_logo_skill.py # tests/skills/test_live_dashboard_skill.py # tests/skills/test_mcp_oauth_remote_gateway_skill.py # tests/skills/test_office_document_skills.py # tests/skills/test_openclaw_migration.py # tests/skills/test_product_price_monitor_skill.py # tests/skills/test_scrollcraft_skill.py # tests/skills/test_setup_wizard_generator_skill.py # tests/skills/test_weekly_review_planning_skill.py # tests/test_engines_satisfiable.py # tests/test_fast_safe_load.py # tests/test_hermes_bootstrap.py # tests/test_hermes_constants.py # tests/test_hermes_logging.py # tests/test_managed_runtime_resolution.py # tests/test_model_tools_async_bridge.py # tests/test_packaging_build_guard.py # tests/test_packaging_metadata.py # tests/test_yaml_indent_consistency.py # tests/tools/test_approval_timeout_overflow.py # tests/tools/test_base_environment.py # tests/tools/test_bot_mode_dm.py # tests/tools/test_browser_chromium_check.py # tests/tools/test_browser_hardening.py # tests/tools/test_browser_homebrew_paths.py # tests/tools/test_browser_npx_warmup.py # tests/tools/test_browser_orphan_reaper.py # tests/tools/test_browser_real_profile.py # tests/tools/test_browser_use_cli.py # tests/tools/test_clipboard.py # tests/tools/test_code_execution.py # tests/tools/test_code_execution_modes.py # tests/tools/test_code_execution_windows_env.py # tests/tools/test_computer_use.py # tests/tools/test_delegate_liveness_timeout.py # tests/tools/test_execute_code_approval_cluster.py # tests/tools/test_execution_flag_detection.py # tests/tools/test_fal_common.py # tests/tools/test_file_operations.py # tests/tools/test_file_tools.py # tests/tools/test_file_tools_cwd_resolution.py # tests/tools/test_file_tools_live.py # tests/tools/test_lazy_deps.py # tests/tools/test_lazy_deps_durable_target.py # tests/tools/test_lazy_deps_managed.py # tests/tools/test_local_env_blocklist.py # tests/tools/test_local_tempdir.py # tests/tools/test_macos_protected_search.py # tests/tools/test_mcp_npx_cached_bin.py # tests/tools/test_oneshot_completion_linger.py # tests/tools/test_process_registry.py # tests/tools/test_read_file_schema_gating.py # tests/tools/test_skill_improvements.py # tests/tools/test_skills_sync.py # tests/tools/test_termux_api_detection.py # tests/tools/test_tirith_security.py # tests/tools/test_transcription_tools.py # tests/tools/test_tts_streaming.py # tests/tools/test_wake_word.py # tests/tui_gateway/test_compute_host_borrowed_lease.py # tests/tui_gateway/test_compute_host_turn_protocol.py # tests/tui_gateway/test_isolated_orphan_activity.py # tests/tui_gateway/test_protocol.py # tests/tui_gateway/test_slash_worker_profile_home.py # tests/tui_gateway/test_subprocess_encoding.py # tests/tui_gateway/test_tui_gateway_server.py # ui-tui/src/__tests__/terminalParity.test.ts # ui-tui/src/__tests__/termuxComposerLayout.test.ts # ui-tui/src/__tests__/textInputFastEcho.test.ts
127 lines
5.4 KiB
Python
127 lines
5.4 KiB
Python
"""Tests for issue #66267 — multimodal (list) tool-result content must not
|
|
crash interim-assistant-message handling or non-streaming message building.
|
|
|
|
Root cause (from the issue discussion): after a vision/tool turn the
|
|
``tool`` message's ``content`` can be a list of parts (e.g. an image plus
|
|
text). Several code paths fed that raw ``content`` straight into regex /
|
|
string helpers that assumed ``str``, raising ``TypeError`` on the
|
|
list. Because the failure happened inside the per-turn loop *before* the
|
|
``role == "assistant"`` guard, the error was retried repeatedly, producing
|
|
the "retry loop" symptom.
|
|
|
|
These tests pin the two fixed surfaces:
|
|
|
|
* ``chat_completion_helpers.build_assistant_message`` (non-streaming /
|
|
gateway path) — now normalizes ``content`` with ``flatten_message_text``
|
|
before the inline ``<think>`` regex and the surrogate sanitizer.
|
|
* ``run_agent.AIAgent._interim_assistant_visible_text`` (used by the
|
|
duplicate-previous-interim dedup in the tool-call path) — now safe for
|
|
any role/content shape because ``flatten_message_text`` handles lists.
|
|
|
|
They also assert the *behavior* the fix preserves: a tool message never
|
|
produces interim text, and list content with an inline ``<think>`` block is
|
|
flattened + stripped correctly.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from types import SimpleNamespace
|
|
from unittest.mock import MagicMock
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Agent fixture — real methods bound where the fix depends on them
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def _make_agent():
|
|
"""Minimal AIAgent with the real text helpers the fix relies on."""
|
|
from agent.message_sanitization import _sanitize_surrogates
|
|
from run_agent import AIAgent
|
|
|
|
agent = MagicMock(spec=AIAgent)
|
|
agent._extract_reasoning = lambda msg: AIAgent._extract_reasoning(agent, msg)
|
|
agent._extract_codex_interim_visible_text = (
|
|
lambda msg: AIAgent._extract_codex_interim_visible_text(agent, msg)
|
|
)
|
|
agent._strip_think_blocks = lambda text: AIAgent._strip_think_blocks(agent, text)
|
|
agent._sanitize_surrogates = lambda text: _sanitize_surrogates(text)
|
|
agent.show_commentary = True
|
|
agent.verbose_logging = False
|
|
agent.reasoning_callback = None
|
|
agent.stream_delta_callback = None
|
|
agent._stream_callback = None
|
|
return agent
|
|
|
|
def _vision_tool_result(text="The image shows a red stop sign."):
|
|
"""A realistic tool message whose content is a list (vision result)."""
|
|
return {
|
|
"role": "tool",
|
|
"tool_call_id": "call_1",
|
|
"content": [
|
|
{"type": "text", "text": text},
|
|
{"type": "image_url", "image_url": {"url": "data:image/png;base64,AAAA"}},
|
|
],
|
|
}
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# build_assistant_message — non-streaming / gateway path (site 2)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
class TestBuildAssistantMessageMultimodal:
|
|
def test_list_content_does_not_crash(self):
|
|
"""A list content (vision result) must build without TypeError."""
|
|
from agent.chat_completion_helpers import build_assistant_message
|
|
|
|
agent = _make_agent()
|
|
sdk_msg = SimpleNamespace(
|
|
content=[
|
|
{"type": "text", "text": "answer after seeing the screenshot"},
|
|
{"type": "image_url", "image_url": {"url": "data:image/png;base64,AAAA"}},
|
|
],
|
|
tool_calls=None,
|
|
reasoning_content=None,
|
|
reasoning_details=None,
|
|
codex_reasoning_items=None,
|
|
codex_message_items=None,
|
|
)
|
|
|
|
msg = build_assistant_message(agent, sdk_msg, "stop")
|
|
|
|
assert msg["role"] == "assistant"
|
|
assert isinstance(msg["content"], str)
|
|
assert "answer after seeing the screenshot" in msg["content"]
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# _interim_assistant_visible_text — dedup path (site 1)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
class TestInterimVisibleTextMultimodal:
|
|
def test_tool_list_content_does_not_crash(self):
|
|
"""A tool message with list content must return a string, not raise."""
|
|
from run_agent import AIAgent
|
|
|
|
agent = _make_agent()
|
|
tool_msg = _vision_tool_result()
|
|
# The helper must not raise on a tool message whose content is a list.
|
|
visible = AIAgent._interim_assistant_visible_text(agent, tool_msg)
|
|
assert isinstance(visible, str)
|
|
|
|
def test_assistant_list_content_flattened(self):
|
|
"""An assistant message with list content yields flattened visible text."""
|
|
from run_agent import AIAgent
|
|
|
|
agent = _make_agent()
|
|
assistant_msg = {
|
|
"role": "assistant",
|
|
"content": [
|
|
{"type": "text", "text": "Let me look at the screenshot."},
|
|
{"type": "image_url", "image_url": {"url": "data:image/png;base64,AAAA"}},
|
|
],
|
|
"finish_reason": "incomplete",
|
|
}
|
|
visible = AIAgent._interim_assistant_visible_text(agent, assistant_msg)
|
|
assert "Let me look at the screenshot." in visible
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# duplicate_previous_interim dedup — the exact shape from conversation_loop.py
|
|
# ---------------------------------------------------------------------------
|