from __future__ import annotations import importlib import json import pytest from apps.web.vue3 import main as WEB_MAIN from gfa_core.mas.agent_state_repository import SQLiteAgentStateRepository from gfa_core.mas.persistent_conversation import PersistentConversationStore @pytest.fixture def web_main(monkeypatch): monkeypatch.setattr(WEB_MAIN, "llm_manager", object()) return WEB_MAIN def _install_router(monkeypatch, outcome, captured): router_module = importlib.import_module("gfa_core.mas.agents.router") async def fake_execute(_self, state, message): captured.append(message.payload) state.results["router_outcome"] = outcome return state monkeypatch.setattr(router_module.RouterAgent, "execute", fake_execute) def test_agent_conversation_store_initializes_from_env( monkeypatch, web_main, tmp_path, ): database_path = tmp_path / "state" / "agent-state.sqlite3" monkeypatch.setenv("PLANTAI_AGENT_STATE_DB", str(database_path)) monkeypatch.setattr(web_main, "agent_conversation_store", None, raising=False) web_main.initialize_agent_conversation_store() assert database_path.is_file() assert isinstance(web_main.agent_conversation_store, PersistentConversationStore) @pytest.mark.asyncio @pytest.mark.parametrize( ("action", "task_type", "expected_intent"), [ ("reply", "faq", None), ("clarify", "clarification", None), ("show_result", "result_view", None), ("propose_analysis", "analysis", {"stages": ["motif"], "stage": "motif"}), ( "propose_utility", "utility", { "taskType": "utility", "utility_tool": "fasta_header_clean", "stages": ["sequence_extract_by_id"], "params": {"sequenceExtractFasta": "test.fa"}, }, ), ], ) async def test_agent_chat_maps_router_actions_and_passes_context( monkeypatch, web_main, action, task_type, expected_intent ): captured = [] outcome = { "status": "ok", "action": action, "message": f"message for {action}", "intent": expected_intent, "sources": [{"path": "final/result.tsv"}], "tool_trace": [{"name": "inspect_analysis_results", "result": "large"}] * 12, } _install_router(monkeypatch, outcome, captured) history = [{"role": "user", "content": "上一轮"}] result = await web_main.agent_chat( web_main.AgentChatRequest( message="请接着处理", session_id="session-7", job_id="job-7", history=history, ) ) assert captured == [ { "query": "请接着处理", "history": history, "session_id": "session-7", "job_id": "job-7", } ] assert result["success"] is True assert result["status"] == "ok" assert result["action"] == action assert result["intent"]["taskType"] == task_type assert result["intent"]["message"] == f"message for {action}" assert result["sources"] == [{"path": "final/result.tsv"}] assert len(result["tool_trace"]) <= 5 assert all(len(item["result"]) <= 1000 for item in result["tool_trace"]) if action in {"propose_analysis", "propose_utility"}: assert result["intent"]["stages"] == expected_intent["stages"] else: assert result["intent"]["stages"] == [] @pytest.mark.asyncio async def test_agent_chat_persists_router_proposal_without_analysis_intent( monkeypatch, web_main, tmp_path, ): repository = SQLiteAgentStateRepository(tmp_path / "agent-state.sqlite3") repository.initialize() store = PersistentConversationStore(repository) monkeypatch.setattr(web_main, "agent_conversation_store", store) captured = [] proposal = { "proposal_id": "proposal-123", "request": "执行 Ft 的数据准备", "subject": "执行 Ft 的数据准备", } _install_router( monkeypatch, { "status": "ok", "action": "clarify", "message": "我理解为只执行 Ft 数据准备，是否确认？", "intent": None, "proposal": proposal, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest( message="执行 Ft 的数据准备", session_id="proposal-session", job_id="job-ft", history=[], ) ) assert result["action"] == "clarify" assert result["intent"]["stages"] == [] ledger = repository.get_goal_ledger(result["session_id"]) assert ledger is not None assert ledger.user_terms["pending_proposal"] == { **proposal, "job_id": "job-ft", } @pytest.mark.asyncio async def test_agent_chat_exposes_confirmed_request_for_original_planner( monkeypatch, web_main, ): captured = [] proposal = { "proposal_id": "proposal-next", "request": "我已确认 config.xlsx，接下来应该做什么分析", "subject": "启动苦荞SPL基因家族分析", } _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已交给现有分析系统识别。", "intent": {"stages": ["genome_prepare"]}, "proposal": proposal, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest( message="确认", session_id="session-next", job_id="job-next", history=[], ) ) assert result["action"] == "propose_analysis" assert result["handoff_request"] == proposal["request"] assert "尚未执行" in result["intent"]["message"] @pytest.mark.asyncio async def test_agent_chat_rejects_job_change_before_analysis_handoff( monkeypatch, web_main, tmp_path, ): from fastapi import HTTPException repository = SQLiteAgentStateRepository(tmp_path / "agent-state.sqlite3") repository.initialize() store = PersistentConversationStore(repository) monkeypatch.setattr(web_main, "agent_conversation_store", store) turn = store.begin_turn( session_id="bound-session", job_id="job-ft", query="分析 Ft.SPL", client_history=[], ) with pytest.raises(HTTPException) as exc_info: await web_main.agent_chat( web_main.AgentChatRequest( message="分析 Os.SPL", session_id=turn.conversation_id, job_id="job-os", history=[], ) ) assert exc_info.value.status_code == 409 assert exc_info.value.detail == "conversation job binding conflict" def test_main_agent_normalization_uses_confirmed_config_family_as_authority( monkeypatch, web_main, ): monkeypatch.setattr( web_main, "_confirmed_config_species_scope", lambda _job_id: { "targetSpecies": ["Ft"], "referenceSpecies": ["At"], "comparativeSpecies": [], }, ) monkeypatch.setattr( web_main, "_infer_family_from_config_or_final", lambda _job_id: "SPL", ) normalized = web_main._normalize_main_agent_analysis_intent( { "stages": ["genome_prepare"], "gene_family": "xlsx，接下来应该做什么分析", "prefix": "Ft.xlsx，接下来应该做什么分析", }, job_id="job-ft", ) assert normalized["gene_family"] == "SPL" assert normalized["prefix"] == "Ft.SPL" assert normalized["params"]["geneFamily"] == "SPL" def test_main_agent_normalization_uses_hydrated_job_identity_without_config( monkeypatch, web_main, ): monkeypatch.setattr(web_main, "_confirmed_config_species_scope", lambda _job: {}) normalized = web_main._normalize_main_agent_analysis_intent( {"stages": ["identify"], "params": {}}, job_id="job-hkl", context_terms={ "gene_family": "HKL", "target_species": ["Sv"], "reference_species": ["At"], }, ) assert normalized["gene_family"] == "HKL" assert normalized["prefix"] == "Sv.HKL" assert normalized["target_species"] == ["Sv"] assert normalized["reference_species"] == ["At"] assert normalized["stages"] == ["identify"] def test_rnaseq_family_member_normalization_keeps_species_prefix( monkeypatch, web_main, ): monkeypatch.setattr(web_main, "_confirmed_config_species_scope", lambda _job: {}) normalized = web_main._normalize_main_agent_analysis_intent( { "gene_family": "HKL", "rna_route": "family_member", "stages": ["rnaseq_de"], "params": {"analysisRoute": "family_member"}, }, job_id="job-os", context_terms={"target_species": ["Os"]}, ) assert normalized["prefix"] == "Os" assert normalized["params"]["prefix"] == "Os" assert normalized["params"]["geneFamily"] == "HKL" def test_current_job_hydration_fills_identity_without_planning_stages( monkeypatch, web_main, tmp_path, ): final_dir = tmp_path / "workspace" / "job-hkl" / "final" final_dir.mkdir(parents=True) (final_dir / "Sv.HKL.subfamily_confirm.manifest.json").write_text( json.dumps( { "stage": "subfamily_confirm", "status": "success", "prefix": "Sv.HKL", "target_species": ["Sv"], "reference_species": ["At"], "analysis_route": "confirmed_subfamily", } ), encoding="utf-8", ) (final_dir / "Sv.HKL.ppi.manifest.json").write_text( json.dumps( { "stage": "ppi", "status": "success", "prefix": "Sv.HKL", "target_species": ["Sv"], "reference_species": ["string_taxon_4555"], } ), encoding="utf-8", ) monkeypatch.setattr(web_main, "WORKSPACE_ROOT", tmp_path / "workspace") monkeypatch.setattr(web_main, "_confirmed_config_species_scope", lambda _job: {}) monkeypatch.setattr(web_main, "_infer_family_from_config_or_final", lambda _job: "HKL") hydrated = web_main._hydrate_goal_terms_from_current_job( {"reference_species": ["ExplicitRef"]}, "job-hkl", ) assert hydrated["gene_family"] == "HKL" assert hydrated["target_species"] == ["Sv"] assert hydrated["reference_species"] == ["ExplicitRef"] assert "stages" not in hydrated assert "requested_stages" not in hydrated canonical = web_main._hydrate_goal_terms_from_current_job({}, "job-hkl") assert canonical["reference_species"] == ["At"] def test_whole_genome_rna_job_does_not_inherit_later_gene_family_context( monkeypatch, web_main, tmp_path, ): final_dir = tmp_path / "workspace" / "job-rna" / "final" final_dir.mkdir(parents=True) for name, payload in { "Os.HKL.subfamily_confirm.manifest.json": { "stage": "subfamily_confirm", "status": "success", "prefix": "Os.HKL", "target_species": ["Os"], "reference_species": ["At"], }, "Os.rnaseq_de.manifest.json": { "stage": "rnaseq_de", "status": "success", "prefix": "Os", "foreground_mode": "global", }, }.items(): (final_dir / name).write_text(json.dumps(payload), encoding="utf-8") monkeypatch.setattr(web_main, "WORKSPACE_ROOT", tmp_path / "workspace") monkeypatch.setattr(web_main, "_confirmed_config_species_scope", lambda _job: {}) monkeypatch.setattr(web_main, "_infer_family_from_config_or_final", lambda _job: "HKL") hydrated = web_main._hydrate_goal_terms_from_current_job({}, "job-rna") assert hydrated["rna_route"] == "whole_genome" assert hydrated["target_species"] == ["Os"] assert "gene_family" not in hydrated assert "reference_species" not in hydrated @pytest.mark.asyncio async def test_agent_chat_provider_failure_has_no_faq_fallback(monkeypatch, web_main): captured = [] _install_router( monkeypatch, { "status": "agent_unavailable", "action": None, "message": "The Router model is currently unavailable.", "intent": None, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest(message="请做 motif 分析", job_id="job-7") ) assert result["success"] is False assert result["status"] == "agent_unavailable" assert result["action"] is None assert result["intent"] is None assert "知识库" not in result["message"] @pytest.mark.asyncio async def test_agent_chat_manager_initialization_failure_is_agent_unavailable( monkeypatch, web_main ): from gfa_core.mas.config import config as mas_config monkeypatch.setattr(web_main, "llm_manager", None) monkeypatch.setattr( mas_config, "load_llm_config", lambda: (_ for _ in ()).throw(RuntimeError("missing provider config")), ) result = await web_main.agent_chat( web_main.AgentChatRequest(message="请做 motif 分析", job_id="job-7") ) assert result["success"] is False assert result["status"] == "agent_unavailable" assert result["action"] is None assert result["intent"] is None @pytest.mark.asyncio async def test_only_propose_analysis_can_return_plannable_intent(monkeypatch, web_main): captured = [] _install_router( monkeypatch, { "status": "ok", "action": "reply", "message": "这是解释，不是分析请求。", "intent": {"stages": ["phylogeny"], "stage": "phylogeny"}, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest(message="这个树为什么这样", job_id="job-7") ) assert result["action"] == "reply" assert result["intent"]["taskType"] == "faq" assert result["intent"]["stages"] == [] @pytest.mark.asyncio async def test_proposed_analysis_normalizes_scope_from_the_current_user_action( monkeypatch, web_main, ): captured = [] _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已识别当前数据准备操作。", "intent": { "gene_family": "SPL", "target_species": ["At"], "reference_species": [], "stages": ["genome_prepare"], "primary_stage": "genome_prepare", "params": { "targetSpecies": ["Ft"], "referenceSpecies": [], "comparativeSpecies": [], "scopeMode": "target_only", }, }, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest( message="执行Ft的数据准备", job_id="job-ft", ) ) assert result["intent"]["target_species"] == ["Ft"] assert result["intent"]["reference_species"] == [] assert result["intent"]["params"]["targetSpecies"] == ["Ft"] assert result["intent"]["prefix"] == "Ft.SPL" @pytest.mark.asyncio async def test_proposed_analysis_uses_intent_params_as_the_scope_authority( monkeypatch, web_main, ): captured = [] _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已识别目标与参考物种。", "intent": { "gene_family": "SPL", "species": "At", "target_species": ["At"], "reference_species": [], "stages": ["identify"], "primary_stage": "identify", "params": { "targetSpecies": ["Ft"], "referenceSpecies": ["At"], "comparativeSpecies": [], "scopeMode": "target_plus_reference", }, }, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest( message=( "我想开展苦荞的SPL基因家族分析，" "Ft是目标物种，At是参考物种" ), job_id="job-ft", ) ) assert result["intent"]["target_species"] == ["Ft"] assert result["intent"]["reference_species"] == ["At"] assert result["intent"]["params"]["targetSpecies"] == ["Ft"] assert result["intent"]["params"]["referenceSpecies"] == ["At"] assert result["intent"]["prefix"] == "Ft.SPL" @pytest.mark.asyncio async def test_confirmed_config_species_roles_override_polluted_intent( monkeypatch, web_main, tmp_path, ): import pandas as pd from gfa_core.mas import confirmation_snapshot input_root = tmp_path / "input" job_dir = input_root / "job-ft" job_dir.mkdir(parents=True) pd.DataFrame( [ {"Species_Code": "Ft", "Species_Role": "target", "Ref": False}, {"Species_Code": "At", "Species_Role": "reference", "Ref": True}, ] ).to_excel(job_dir / "config.xlsx", index=False) monkeypatch.setattr(web_main, "INPUT_ROOT", input_root) monkeypatch.setattr( confirmation_snapshot, "build_confirmation_snapshot", lambda *_args, **_kwargs: { "config_confirmed": True, "samples_confirmed": False, "subfamily_confirmed": False, "config": {"artifact_path": str(job_dir / "config.xlsx")}, "samples": None, "subfamily": None, "errors": (), }, ) captured = [] _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已识别目标与参考物种。", "intent": { "gene_family": "SPL", "species": "unknown", "target_species": [], "reference_species": ["Ft", "At"], "stages": ["identify"], "primary_stage": "identify", "params": { "targetSpecies": [], "referenceSpecies": ["Ft", "At"], "scopeMode": "target_plus_reference", }, }, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest( message=( "苦荞数据在 /data/ref_data/Tartary_genome，" "拟南芥数据在 /data/ref_data/ath" ), job_id="job-ft", ) ) assert result["intent"]["target_species"] == ["Ft"] assert result["intent"]["reference_species"] == ["At"] assert result["intent"]["params"]["targetSpecies"] == ["Ft"] assert result["intent"]["params"]["referenceSpecies"] == ["At"] assert result["intent"]["prefix"] == "Ft.SPL" @pytest.mark.asyncio async def test_explicit_species_role_conflict_returns_actionable_clarification( monkeypatch, web_main, tmp_path, ): import pandas as pd from gfa_core.mas import confirmation_snapshot input_root = tmp_path / "input" job_dir = input_root / "job-os" job_dir.mkdir(parents=True) pd.DataFrame( [ {"Species_Code": "Os", "Species_Role": "target", "Ref": False}, {"Species_Code": "At", "Species_Role": "reference", "Ref": True}, ] ).to_excel(job_dir / "config.xlsx", index=False) monkeypatch.setattr(web_main, "INPUT_ROOT", input_root) monkeypatch.setattr( confirmation_snapshot, "build_confirmation_snapshot", lambda *_args, **_kwargs: { "config_confirmed": True, "samples_confirmed": False, "subfamily_confirmed": False, "config": {"artifact_path": str(job_dir / "config.xlsx")}, "samples": None, "subfamily": None, "errors": (), }, ) captured = [] _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已识别分析需求。", "intent": { "gene_family": "PEBP", "target_species": ["At"], "reference_species": ["Os"], "stages": ["identify"], "primary_stage": "identify", "params": { "targetSpecies": ["At"], "referenceSpecies": ["Os"], }, }, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest( message="At是目标物种，Os是参考物种，请执行PEBP基因家族分析", job_id="job-os", ) ) assert result["action"] == "clarify" assert result["intent"]["taskType"] == "clarification" assert "At" in result["intent"]["message"] assert "Os" in result["intent"]["message"] assert "打开 config.xlsx" in result["intent"]["message"] assert result["intent"]["stages"] == [] @pytest.mark.asyncio @pytest.mark.parametrize( "candidate", [ None, {}, [], "not-an-intent", {"foo": "bar"}, {"stages": []}, {"stages": [""]}, {"stages": "motif"}, ], ) async def test_propose_analysis_without_verified_intent_fails_closed( monkeypatch, web_main, candidate ): captured = [] _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "分析建议", "intent": candidate, "sources": [], "tool_trace": [], }, captured, ) result = await web_main.agent_chat( web_main.AgentChatRequest(message="继续分析", job_id="job-7") ) assert result["success"] is False assert result["status"] == "invalid_response" assert result["action"] is None assert result["intent"] is None @pytest.mark.asyncio @pytest.mark.parametrize("query", ["那两个", "继续后面的", "刚才的结果"]) async def test_coreference_journeys_reach_router_with_prior_history( monkeypatch, web_main, query ): captured = [] _install_router( monkeypatch, { "status": "ok", "action": "clarify", "message": "请确认你指的是哪一项。", "intent": None, "sources": [], "tool_trace": [], }, captured, ) history = [ {"role": "user", "content": "比较 motif 和 domain"}, {"role": "assistant", "content": "我找到了两个结果。"}, ] result = await web_main.agent_chat( web_main.AgentChatRequest(message=query, history=history, job_id="job-7") ) assert captured[0]["query"] == query assert captured[0]["history"] == history assert result["action"] == "clarify" @pytest.mark.asyncio async def test_agent_chat_uses_persisted_history_and_goal_ledger( monkeypatch, web_main, tmp_path, ): repository = SQLiteAgentStateRepository(tmp_path / "agent-state.sqlite3") repository.initialize() store = PersistentConversationStore(repository) monkeypatch.setattr(web_main, "agent_conversation_store", store, raising=False) captured = [] _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已记录初始目标。", "intent": {"stages": ["motif"], "stage": "motif"}, "sources": [], "tool_trace": [], "interaction_kind": "independent", }, captured, ) first = await web_main.agent_chat( web_main.AgentChatRequest( message="进行 motif 分析", session_id="legacy-session", job_id="job-ft", ) ) _install_router( monkeypatch, { "status": "ok", "action": "propose_analysis", "message": "已按修正后的目标处理。", "intent": {"stages": ["domain"], "stage": "domain"}, "sources": [], "tool_trace": [], "interaction_kind": "correction", }, captured, ) second = await web_main.agent_chat( web_main.AgentChatRequest( message="不是 motif，改成 domain", session_id=first["session_id"], job_id="", history=[{"role": "user", "content": "客户端伪造历史"}], ) ) assert captured[1]["job_id"] == "job-ft" assert captured[1]["history"] == [ {"role": "user", "content": "进行 motif 分析"}, {"role": "assistant", "content": "已记录初始目标。"}, ] assert captured[1]["goal_ledger"].active_goal == "进行 motif 分析" ledger = repository.get_goal_ledger(second["session_id"]) assert ledger is not None assert ledger.active_goal == "不是 motif，改成 domain" assert ledger.corrections == ("不是 motif，改成 domain",) assert [item.content for item in repository.list_messages(second["session_id"])] == [ "进行 motif 分析", "已记录初始目标。", "不是 motif，改成 domain", "已按修正后的目标处理。", ] @pytest.mark.asyncio async def test_deterministic_utility_turn_uses_same_persistent_conversation( monkeypatch, web_main, tmp_path, ): repository = SQLiteAgentStateRepository(tmp_path / "agent-state.sqlite3") repository.initialize() store = PersistentConversationStore(repository) monkeypatch.setattr(web_main, "agent_conversation_store", store, raising=False) monkeypatch.setattr( web_main.extension_stage_agent, "list_extensions", lambda include_drafts=False: [], ) message = "请从 proteins.fa 按 ids.txt 提取蛋白序列" result = await web_main.agent_chat( web_main.AgentChatRequest( message=message, session_id="browser-session", job_id="job-ft", ) ) assert result["action"] == "propose_utility" assert result["status"] == "ok" assert [item.content for item in repository.list_messages(result["session_id"])] == [ message, result["intent"]["message"], ] ledger = repository.get_goal_ledger(result["session_id"]) assert ledger is not None assert ledger.active_goal == message