mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-11-16 19:36:54 +00:00
# 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
247 lines
11 KiB
Python
247 lines
11 KiB
Python
"""Tests for reactive multimodal-tool-content recovery.
|
|
|
|
Covers the full chain for providers that reject list-type content in
|
|
``role: "tool"`` messages (Xiaomi MiMo's 400 "text is not set", etc.):
|
|
|
|
1. agent/error_classifier.py: 400 with the right wording classifies as
|
|
``FailoverReason.multimodal_tool_content_unsupported``.
|
|
2. run_agent._try_strip_image_parts_from_tool_messages downgrades tool
|
|
messages whose ``content`` is a list-with-image to a string text
|
|
summary, in-place, and records the active (provider, model) in
|
|
``self._no_list_tool_content_models`` so future tool results in this
|
|
session preemptively downgrade.
|
|
3. run_agent._tool_result_content_for_active_model short-circuits to a
|
|
text summary when the (provider, model) is in the cache, even though
|
|
``_model_supports_vision`` returns True — avoiding a wasted round
|
|
trip on every subsequent screenshot in the session.
|
|
|
|
The end-to-end retry loop wiring (`conversation_loop.py`) is exercised by
|
|
the classifier signal + helper-mutation tests; the integration only adds
|
|
a trivial flag-and-continue around the existing pattern used for
|
|
``image_too_large`` recovery.
|
|
|
|
See: https://github.com/NousResearch/hermes-agent/issues/27344
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
|
|
from agent.error_classifier import FailoverReason, classify_api_error
|
|
|
|
|
|
class _FakeApiError(Exception):
|
|
"""Stand-in for an openai.BadRequestError with status_code + body."""
|
|
|
|
def __init__(self, status_code: int, message: str, body: dict | None = None):
|
|
super().__init__(message)
|
|
self.status_code = status_code
|
|
self.body = body or {"error": {"message": message}}
|
|
self.response = None
|
|
|
|
|
|
def _make_agent(provider: str = "xiaomi", model: str = "mimo-v2.5"):
|
|
"""Build a bare AIAgent for method-level testing, no provider setup."""
|
|
from run_agent import AIAgent
|
|
agent = object.__new__(AIAgent)
|
|
agent.provider = provider
|
|
agent.model = model
|
|
return agent
|
|
|
|
|
|
# ─── Strip helper ────────────────────────────────────────────────────────────
|
|
|
|
|
|
class TestStripImagePartsHelper:
|
|
|
|
|
|
def test_tool_message_with_string_content_unchanged(self):
|
|
agent = _make_agent()
|
|
msgs = [
|
|
{"role": "tool", "tool_call_id": "x", "content": "plain string result"},
|
|
]
|
|
assert agent._try_strip_image_parts_from_tool_messages(msgs) is False
|
|
assert msgs[0]["content"] == "plain string result"
|
|
|
|
def test_tool_message_list_without_image_unchanged(self):
|
|
"""List content with only text parts is left alone — caller surfaces
|
|
the original error if this turns out to also be rejected."""
|
|
agent = _make_agent()
|
|
msgs = [
|
|
{"role": "tool", "tool_call_id": "x", "content": [
|
|
{"type": "text", "text": "hello"},
|
|
]},
|
|
]
|
|
assert agent._try_strip_image_parts_from_tool_messages(msgs) is False
|
|
|
|
|
|
|
|
def test_records_provider_model_in_session_cache(self):
|
|
agent = _make_agent(provider="xiaomi", model="mimo-v2.5")
|
|
msgs = [
|
|
{"role": "tool", "tool_call_id": "x", "content": [
|
|
{"type": "text", "text": "summary"},
|
|
{"type": "image_url", "image_url": {"url": "data:image/png;base64,X"}},
|
|
]},
|
|
]
|
|
agent._try_strip_image_parts_from_tool_messages(msgs)
|
|
assert ("xiaomi", "mimo-v2.5") in agent._no_list_tool_content_models
|
|
|
|
def test_only_tool_messages_get_downgraded(self):
|
|
"""User / assistant messages with list-type content are out of
|
|
scope — they're handled by the existing image-routing path."""
|
|
agent = _make_agent()
|
|
msgs = [
|
|
{"role": "user", "content": [
|
|
{"type": "text", "text": "describe"},
|
|
{"type": "image_url", "image_url": {"url": "data:image/png;base64,X"}},
|
|
]},
|
|
{"role": "tool", "tool_call_id": "x", "content": [
|
|
{"type": "text", "text": "summary"},
|
|
{"type": "image_url", "image_url": {"url": "data:image/png;base64,Y"}},
|
|
]},
|
|
]
|
|
agent._try_strip_image_parts_from_tool_messages(msgs)
|
|
# User message untouched.
|
|
assert isinstance(msgs[0]["content"], list)
|
|
assert any(p.get("type") == "image_url" for p in msgs[0]["content"])
|
|
# Tool message downgraded.
|
|
assert isinstance(msgs[1]["content"], str)
|
|
assert "summary" in msgs[1]["content"]
|
|
|
|
|
|
|
|
# ─── Short-circuit on cached models ──────────────────────────────────────────
|
|
|
|
|
|
class TestToolResultContentShortCircuit:
|
|
"""Once the session has learned that (provider, model) rejects list
|
|
content, ``_tool_result_content_for_active_model`` returns a text
|
|
summary even though ``_model_supports_vision`` reports True.
|
|
"""
|
|
|
|
def _multimodal_result(self, png_b64: str = "iVBORw0KGgoAAAA"):
|
|
return {
|
|
"_multimodal": True,
|
|
"content": [
|
|
{"type": "text", "text": "capture mode=som 800x600 app=Safari"},
|
|
{"type": "image_url",
|
|
"image_url": {"url": f"data:image/png;base64,{png_b64}"}},
|
|
],
|
|
"text_summary": "capture mode=som 800x600 app=Safari",
|
|
"meta": {"mode": "som", "width": 800, "height": 600, "elements": 5,
|
|
"png_bytes": 1024},
|
|
}
|
|
|
|
def test_returns_text_summary_for_xiaomi_proactively(self, monkeypatch):
|
|
"""Xiaomi MiMo rejects list-type tool content, so even with an
|
|
empty cache, _tool_result_content_for_active_model should
|
|
proactively downgrade to a text summary."""
|
|
agent = _make_agent(provider="xiaomi", model="mimo-v2.5")
|
|
agent._no_list_tool_content_models = set() # explicit empty
|
|
monkeypatch.setattr(agent, "_model_supports_vision", lambda: True)
|
|
out = agent._tool_result_content_for_active_model(
|
|
"computer_use", self._multimodal_result()
|
|
)
|
|
# Proactive downgrade: text summary instead of list with images.
|
|
assert isinstance(out, str)
|
|
assert "data:image" not in out
|
|
assert "image_url" not in out
|
|
|
|
|
|
|
|
|
|
|
|
# ─── Classifier ──────────────────────────────────────────────────────────────
|
|
|
|
|
|
class TestRecoveryEndToEndClassification:
|
|
"""Lock in that the patterns used by the recovery path classify to
|
|
the right ``FailoverReason``. (The recovery hook in
|
|
``agent.conversation_loop`` consumes this reason directly.)
|
|
"""
|
|
|
|
def test_xiaomi_mimo_classifies(self):
|
|
err = _FakeApiError(
|
|
status_code=400,
|
|
message=(
|
|
"Error code: 400 - {'error': {'code': '400', 'message': "
|
|
"'Param Incorrect', 'param': 'text is not set', 'type': ''}}"
|
|
),
|
|
)
|
|
result = classify_api_error(err, provider="xiaomi", model="mimo-v2.5")
|
|
assert result.reason == FailoverReason.multimodal_tool_content_unsupported
|
|
assert result.retryable is True
|
|
|
|
def test_alibaba_variant_classifies(self):
|
|
err = _FakeApiError(
|
|
status_code=400,
|
|
message="tool_call.content must be string",
|
|
)
|
|
result = classify_api_error(err, provider="alibaba", model="qwen3.5-plus")
|
|
assert result.reason == FailoverReason.multimodal_tool_content_unsupported
|
|
|
|
def test_opencode_go_422_console_go_classifies(self):
|
|
"""Regression test for #104731: Console Go relay HTTP 422 on list-type tool content."""
|
|
err = _FakeApiError(
|
|
status_code=422,
|
|
message=(
|
|
"Error code: 422 - {'error': {'param': 'messages.67.tool.content.str', "
|
|
"'type': 'invalid_request_error', 'message': 'Error from provider (Console Go): "
|
|
"Upstream request failed: [invalid_request_error] Input should be a valid string'}}"
|
|
),
|
|
)
|
|
result = classify_api_error(err, provider="opencode-go", model="deepseek-v4-flash-vision-exp")
|
|
assert result.reason == FailoverReason.multimodal_tool_content_unsupported
|
|
assert result.retryable is True
|
|
|
|
def test_nvidia_nim_serde_tool_content_classifies(self):
|
|
"""Regression test for #111231 (bug 2): NVIDIA NIM's Rust gateway rejects
|
|
list-type tool content without naming the field — the serde text names the
|
|
*enum* instead. Verbatim provider wording from the report."""
|
|
err = _FakeApiError(
|
|
status_code=400,
|
|
message=(
|
|
"HTTP 400: Failed to deserialize the JSON body into the target type: "
|
|
"data did not match any variant of untagged enum "
|
|
"ChatCompletionRequestToolMessageContent at line 1 column 1974809"
|
|
),
|
|
)
|
|
result = classify_api_error(err, provider="nvidia", model="moonshotai/kimi-k3")
|
|
assert result.reason == FailoverReason.multimodal_tool_content_unsupported
|
|
assert result.retryable is True
|
|
|
|
def test_nvidia_nim_serde_other_enum_not_misclassified(self):
|
|
"""A NIM serde rejection naming a different enum must stay a format error —
|
|
only the tool-message-content enum means list-type tool content."""
|
|
err = _FakeApiError(
|
|
status_code=400,
|
|
message=(
|
|
"HTTP 400: Failed to deserialize the JSON body into the target type: "
|
|
"data did not match any variant of untagged enum ChatCompletionRequestMessage"
|
|
),
|
|
)
|
|
result = classify_api_error(err, provider="nvidia", model="moonshotai/kimi-k3")
|
|
assert result.reason != FailoverReason.multimodal_tool_content_unsupported
|
|
|
|
class TestOpenCodeGoProactiveToolResultDowngrade:
|
|
def _multimodal_result(self, png_b64: str = "iVBORw0KGgoAAAA"):
|
|
return {
|
|
"_multimodal": True,
|
|
"content": [
|
|
{"type": "text", "text": "detected 2 objects"},
|
|
{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{png_b64}"}},
|
|
],
|
|
"text_summary": "detected 2 objects",
|
|
"meta": {"png_bytes": 1024},
|
|
}
|
|
|
|
def test_returns_text_summary_for_opencode_go_proactively(self, monkeypatch):
|
|
"""OpenCode Go rejects list-type tool content, so _tool_result_content_for_active_model
|
|
should proactively downgrade to a text summary (#104731)."""
|
|
agent = _make_agent(provider="opencode-go", model="deepseek-v4-flash-vision-exp")
|
|
agent._no_list_tool_content_models = set()
|
|
monkeypatch.setattr(agent, "_model_supports_vision", lambda: True)
|
|
out = agent._tool_result_content_for_active_model("vision_analyze", self._multimodal_result())
|
|
assert isinstance(out, str)
|
|
assert "data:image" not in out
|
|
assert "image_url" not in out
|