"""Tests for the smart skill-list search matcher. ``GET /api/skills?search=`` used to require the whole query as a contiguous substring of the English name / Chinese name only — searching by what a skill *does* (description, detail, tags) or with several keywords always came back empty. These tests pin the intended behaviour: - whitespace-separated keywords, AND across tokens but each token may hit a different field; - fields: name, name_zh, description, detail, tag names; - case-insensitive, and still works as a plain single-keyword substring. """ from __future__ import annotations import pytest from app.gateway.routers.skills import ( SkillResponse, _attach_and_filter_tags, _skill_matches_tokens, _skill_search_tokens, ) from deerflow.persistence.tags import MemoryTagStore from deerflow.skills.types import SkillCategory def _resp( name: str, *, name_zh: str | None = None, description: str = "", detail: str | None = None, ) -> SkillResponse: return SkillResponse( name=name, name_zh=name_zh, description=description, detail=detail, category=SkillCategory.CUSTOM, ) def test_tokens_split_and_lowercase() -> None: assert _skill_search_tokens(None) == [] assert _skill_search_tokens(" ") == [] assert _skill_search_tokens(" Report 生成 ") == ["report", "生成"] def test_match_requires_every_token_somewhere() -> None: skill = _resp("web-search", name_zh="网页搜索", description="Search the public web for reports") # One token hits the name, the other the description — old contiguous # substring semantics would have rejected this. assert _skill_matches_tokens(skill, ["web", "report"]) is True # A token nowhere present rejects the skill even though the other matches. assert _skill_matches_tokens(skill, ["web", "missing"]) is False def test_match_hits_name_zh_description_and_detail() -> None: assert _skill_matches_tokens(_resp("ws", name_zh="舆情分析"), ["舆情"]) is True assert _skill_matches_tokens(_resp("ws", description="生成任务研判报告"), ["研判"]) is True assert _skill_matches_tokens(_resp("ws", detail="本技能用于军事情报整编"), ["整编"]) is True def test_match_is_case_insensitive_substring() -> None: skill = _resp("Task-Report-Builder", description="builds reports") # Tokens reaching the matcher are pre-lowercased by _skill_search_tokens. tokens = _skill_search_tokens("Task-Report BUILDER") assert tokens == ["task-report", "builder"] assert _skill_matches_tokens(skill, tokens) is True def test_empty_tokens_match_everything() -> None: assert _skill_matches_tokens(_resp("anything"), []) is True @pytest.mark.asyncio async def test_attach_and_filter_searches_tags() -> None: store = MemoryTagStore() await store.create_tag({"id": "t1", "name": "情报整编", "tag_type": "skill"}) await store.set_target_tags("skill", "alpha", ["t1"]) responses = [ _resp("alpha", description="unrelated text"), _resp("beta", description="unrelated text"), ] kept = await _attach_and_filter_tags(store, responses, "整编", None) assert [r.name for r in kept] == ["alpha"] # The attached tags ride along on the response. assert [t["name"] for t in kept[0].tags] == ["情报整编"] @pytest.mark.asyncio async def test_attach_and_filter_keeps_legacy_single_keyword_behavior() -> None: store = MemoryTagStore() responses = [ _resp("web-search", name_zh="网页搜索", description="search the web"), _resp("code-lint", name_zh="代码检查", description="lint python files"), ] kept = await _attach_and_filter_tags(store, responses, "web", None) assert [r.name for r in kept] == ["web-search"] kept = await _attach_and_filter_tags(store, responses, "代码", None) assert [r.name for r in kept] == ["code-lint"]