diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md index fcbb3651..a6f364e5 100644 --- a/offline-backend-20260512/backend/CLAUDE.md +++ b/offline-backend-20260512/backend/CLAUDE.md @@ -273,7 +273,7 @@ The AI-writing graph (`packages/harness/deerflow/agents/ai_writing/`) has four s Seeding mirrors the roundtable pattern: verified `config.yaml` + `SOUL.md` ship in `app/gateway/routers/_ai_writing_seed_assets//`; `_ai_writing_seed.py::ensure_ai_writing_functional_agents()` copies them to `.deer-flow/agents/` when missing (never overwrites local edits), called from the startup lifespan (before `_sync_legacy_agents`, so the agents appear in `/api/agents` on first boot) and self-heals from `GET /api/ai-writing/sessions` + `GET /api/ai-writing/article-types`. Tests: `tests/test_ai_writing_seed.py`, `tests/test_ai_writing_agent_runner.py`. -**Skill-driven retrieval (检索类型).** `material_source` has three values: `general` (通用检索, default), `knowledge_base` (知识库检索), and `notebook` (我的空间); legacy `skill` is normalized to `general` by the researcher node. **`knowledge_base` 直连内网 ES 接口** — it bypasses skill orchestration entirely and calls `IntranetSearchProvider` over `ai_writing.intranet_search_url` (concurrent over the generated keywords; `intranet_knowledge_base`/`intranet_search_timeout` apply); an unconfigured url → no materials + a `skill_error` notice. This is the "弱模型识别不了技能" escape hatch — the old no-skill retrieval path restored as an explicit option (weak models like deepseek-chat often fail to invoke configured skills). Tests: `tests/test_ai_writing_knowledge_base.py`. General retrieval is orchestrated from the **skills configured on the `ai-writing-researcher` agent** (`config.yaml → skills`, editable from the agent management page): the built-in `knowledge-base-search` skill (知识库检索) runs the native keyword fan-out web/intranet search (the old default path), while any other configured skill searches with a `description`-targeted query; skills run in configured order with early-stop at `max_materials`, and an empty/unresolvable skills config falls back to the built-in knowledge search so retrieval always works. The skill itself is seeded from `_ai_writing_seed_assets/skills/knowledge-base-search/SKILL.md` into `skills/public/` (never overwrites), and a pre-existing researcher `config.yaml` lacking the `skills` key is patched in place with the default (explicit `skills: []` is respected). **PAUSE-1 input append flow**: when the user types natural language into the material-confirm input (`re_search`/legacy `enable_skill_search` + `user_query`), the node matches the most relevant configured skill via `SkillPicker(candidates=…)` (skipped when only one skill is configured), searches with the user's words, and **appends** the new materials to the existing package (URL-dedup, original order preserved, no LLM re-rank/truncation, 100-item hard cap). **Skill-failure fault tolerance**: a configured skill that times out / raises / has all its keyword searches fail (vs. legitimately returning 0 results) no longer just silently moves on — the researcher node emits a `skill_search_step` of type `skill_error` (carrying the error `reason`), and if the configured skills collected **no** new materials at all it runs a `fallback` pass over other available skills (`_fallback_candidate_skills`: the built-in `knowledge-base-search` native path first, then other enabled catalog skills not yet tried) instead of erroring out. The frontend `SkillStepsCard` renders `skill_error` (rose) and `fallback` (amber) so the timeline shows "技能报错 → 改用其他技能搜集". Tests: `tests/test_ai_writing_material_source.py`. +**Skill-driven retrieval (检索类型).** `material_source` has three values: `general` (通用检索, default), `knowledge_base` (知识库检索), and `notebook` (我的空间); legacy `skill` is normalized to `general` by the researcher node. **`knowledge_base` 直连内网 ES 接口（= 旧版「内网直连检索」）** — it bypasses skill orchestration entirely and calls `IntranetSearchProvider` over `ai_writing.intranet_search_url` (concurrent over the generated keywords; `intranet_knowledge_base`/`intranet_search_timeout` apply). **An empty/unconfigured `intranet_search_url` does NOT disable it** — the empty string is passed straight to `IntranetSearchProvider`, which falls back to its built-in default endpoint (`intranet_stub._DEFAULT_ENDPOINT`), exactly like the legacy `get_search_provider(search_provider=intranet)` path. So 内网直连检索 works out of the box without configuring an address (configure `intranet_search_url` only to point at a different endpoint). This is the "弱模型识别不了技能" escape hatch — the old no-skill retrieval path restored as an explicit option (weak models like deepseek-chat often fail to invoke configured skills). Tests: `tests/test_ai_writing_knowledge_base.py`. General retrieval is orchestrated from the **skills configured on the `ai-writing-researcher` agent** (`config.yaml → skills`, editable from the agent management page): the built-in `knowledge-base-search` skill (知识库检索) runs the native keyword fan-out web/intranet search (the old default path), while any other configured skill searches with a `description`-targeted query; skills run in configured order with early-stop at `max_materials`, and an empty/unresolvable skills config falls back to the built-in knowledge search so retrieval always works. The skill itself is seeded from `_ai_writing_seed_assets/skills/knowledge-base-search/SKILL.md` into `skills/public/` (never overwrites), and a pre-existing researcher `config.yaml` lacking the `skills` key is patched in place with the default (explicit `skills: []` is respected). **PAUSE-1 input append flow**: when the user types natural language into the material-confirm input (`re_search`/legacy `enable_skill_search` + `user_query`), the node matches the most relevant configured skill via `SkillPicker(candidates=…)` (skipped when only one skill is configured), searches with the user's words, and **appends** the new materials to the existing package (URL-dedup, original order preserved, no LLM re-rank/truncation, 100-item hard cap). **Skill-failure fault tolerance**: a configured skill that times out / raises / has all its keyword searches fail (vs. legitimately returning 0 results) no longer just silently moves on — the researcher node emits a `skill_search_step` of type `skill_error` (carrying the error `reason`), and if the configured skills collected **no** new materials at all it runs a `fallback` pass over other available skills (`_fallback_candidate_skills`: the built-in `knowledge-base-search` native path first, then other enabled catalog skills not yet tried) instead of erroring out. The frontend `SkillStepsCard` renders `skill_error` (rose) and `fallback` (amber) so the timeline shows "技能报错 → 改用其他技能搜集". Tests: `tests/test_ai_writing_material_source.py`. **Skill-call timeout** is hardcoded **300s (5 min)** — `researcher.py` `_SKILL_SEARCH_TIMEOUT_SECONDS` (per-skill wrapper) and `_skill_agent.py` `_REAL_AGENT_TIMEOUT_SECONDS` (whole real-agent skill run). The intranet ES fallback (`IntranetSearchProvider`) gets a longer **600s (10 min)** default (`intranet_stub.py`; overridable via `ai_writing.intranet_search_timeout`) — that interface is the slow one, so our own fallback timeout is generous to avoid being blamed for "检不出". diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py index 92ba3692..acfbe067 100644 --- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py +++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py @@ -5,9 +5,10 @@ config.yaml 配置的技能顺序编排检索。内置「知识库检索」技能 （knowledge-base-search）走原生 web/intranet 关键词并发检索；其它技能用 其 description 生成定向 query 后调真实 provider 检索。 - - ``knowledge_base``（知识库检索）：**直连内网知识库 ES 接口** - （``ai_writing.intranet_search_url`` 配的 ``IntranetSearchProvider``）， - 不编排任何技能、也不跑通用 provider。未配置接口地址时本次无素材并给出告警。 + - ``knowledge_base``（知识库检索 / 内网直连检索）：**直连内网知识库 ES 接口** + （``IntranetSearchProvider``），不编排任何技能、也不跑通用 provider。接口地址取 + ``ai_writing.intranet_search_url``，**留空时回退 provider 内置默认内网端点**——与旧版 + ``get_search_provider(search_provider=intranet)`` 一致，即便没配地址内网直连仍可用。 - ``notebook``（我的空间）：仅用笔记 API 素材。 旧值 ``skill`` 仍按 ``general`` 兼容处理。 @@ -568,57 +569,50 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict # 仅使用 Step 1.5 已拉取的 notebook_materials。 _emit(f"仅使用笔记素材，跳过技能检索（{len(notebook_materials)} 条）") elif material_source == "knowledge_base": - # ── 知识库检索：直连内网 ES 接口 ── - # 用户在表单选了「知识库检索」→ 跳过技能编排、跳过通用 provider， - # 直接调 config.yaml 配的 ai_writing.intranet_search_url 内网知识库 ES 接口 - # （IntranetSearchProvider）对检索词并发检索。未配置地址 → 无素材 + 告警。 - intranet_url = (ai_writing_config.get("intranet_search_url") or "").strip() - if not intranet_url: - _emit("未配置内网知识库检索接口（ai_writing.intranet_search_url），本次未检索到素材") - search_errors.append("未配置 ai_writing.intranet_search_url，知识库检索不可用") - _emit_skill_step({ - "type": "skill_error", - "message": "「知识库检索」未配置内网接口地址（ai_writing.intranet_search_url），无法检索", - "skill_name": KNOWLEDGE_SEARCH_SKILL_NAME, - "reason": "intranet_search_url 为空", - }) - else: - from ..search.intranet_stub import IntranetSearchProvider + # ── 知识库检索：直连内网 ES 接口（恢复旧版「内网直连检索」可用逻辑）── + # 用户在表单选了「知识库检索」→ 跳过技能编排、跳过通用 provider，直接调内网知识库 + # ES 接口（IntranetSearchProvider）对检索词并发检索。接口地址优先取 config.yaml 的 + # ai_writing.intranet_search_url；**留空时回退到 IntranetSearchProvider 内置默认端点** + # （历史上一直可用的内网 ES 地址），与旧版 get_search_provider(search_provider=intranet) + # 完全一致——所以即便没配地址，内网直连依然可用，不再短路成「未配置 → 无素材」。 + from ..search.intranet_stub import IntranetSearchProvider - provider = IntranetSearchProvider( - base_url=intranet_url, - knowledge_base_name=ai_writing_config.get("intranet_knowledge_base", "samples"), - # 内网慢检索超时可在 config.yaml 的 ai_writing.intranet_search_timeout 调大，默认 600s。 - timeout=ai_writing_config.get("intranet_search_timeout", 600), - ) - # 用户输入框追加（is_user_query_search）时 keywords 已是 [用户原话]，全部检索； - # 首次检索按 keyword_count 截断检索词。 - search_keywords = keywords if is_user_query_search else keywords[:keyword_count] - total = len(search_keywords) - per_keyword = _per_keyword_limit(max_materials, total) - _emit(f"知识库检索：直连内网接口检索 {total} 个方向…") - _emit_skill_step({ - "type": "invoke_skill", - "message": f"使用「{_KNOWLEDGE_SKILL_DISPLAY}」直连内网接口检索 {total} 个方向：{('、'.join(search_keywords[:3]))[:60]}", - "skill_name": KNOWLEDGE_SEARCH_SKILL_NAME, - }) - gathered = await asyncio.gather( - *[provider.search(kw, max_results=per_keyword) for kw in search_keywords], - return_exceptions=True, - ) - for results in gathered: - if isinstance(results, Exception): - logger.warning("[knowledge-base] search failed: %s", results) - search_errors.append(f"{type(results).__name__}: {results}") - continue - new_results.extend(results or []) - search_errors.extend(getattr(provider, "errors", [])) - _emit_skill_step({ - "type": "skill_result", - "message": f"「{_KNOWLEDGE_SKILL_DISPLAY}」（内网接口）检索到 {len(new_results)} 条素材", - "skill_name": KNOWLEDGE_SEARCH_SKILL_NAME, - "matched_count": len(new_results), - }) + intranet_url = (ai_writing_config.get("intranet_search_url") or "").strip() + provider = IntranetSearchProvider( + base_url=intranet_url, # 空串 → provider 内部回退默认内网端点（旧逻辑） + knowledge_base_name=ai_writing_config.get("intranet_knowledge_base", "samples"), + # 内网慢检索超时可在 config.yaml 的 ai_writing.intranet_search_timeout 调大，默认 600s。 + timeout=ai_writing_config.get("intranet_search_timeout", 600), + ) + endpoint_label = intranet_url or getattr(provider, "endpoint", "内置默认内网端点") + # 用户输入框追加（is_user_query_search）时 keywords 已是 [用户原话]，全部检索； + # 首次检索按 keyword_count 截断检索词。 + search_keywords = keywords if is_user_query_search else keywords[:keyword_count] + total = len(search_keywords) + per_keyword = _per_keyword_limit(max_materials, total) + _emit(f"知识库检索：直连内网接口检索 {total} 个方向…") + _emit_skill_step({ + "type": "invoke_skill", + "message": f"使用「{_KNOWLEDGE_SKILL_DISPLAY}」直连内网接口（{endpoint_label}）检索 {total} 个方向：{('、'.join(search_keywords[:3]))[:60]}", + "skill_name": KNOWLEDGE_SEARCH_SKILL_NAME, + }) + gathered = await asyncio.gather( + *[provider.search(kw, max_results=per_keyword) for kw in search_keywords], + return_exceptions=True, + ) + for results in gathered: + if isinstance(results, Exception): + logger.warning("[knowledge-base] search failed: %s", results) + search_errors.append(f"{type(results).__name__}: {results}") + continue + new_results.extend(results or []) + search_errors.extend(getattr(provider, "errors", [])) + _emit_skill_step({ + "type": "skill_result", + "message": f"「{_KNOWLEDGE_SKILL_DISPLAY}」（内网接口）检索到 {len(new_results)} 条素材", + "skill_name": KNOWLEDGE_SEARCH_SKILL_NAME, + "matched_count": len(new_results), + }) elif is_append_only: # ── 兼容旧版追加关键词模式 ── # 直接对用户给的关键词做原生并发检索（不走技能编排），结果追加到旧素材。 diff --git a/offline-backend-20260512/backend/tests/test_ai_writing_knowledge_base.py b/offline-backend-20260512/backend/tests/test_ai_writing_knowledge_base.py index 99fa41b0..02a064f5 100644 --- a/offline-backend-20260512/backend/tests/test_ai_writing_knowledge_base.py +++ b/offline-backend-20260512/backend/tests/test_ai_writing_knowledge_base.py @@ -88,12 +88,16 @@ async def test_knowledge_base_hits_intranet_directly(monkeypatch): @pytest.mark.asyncio -async def test_knowledge_base_no_url_empty(monkeypatch): - """未配置内网接口地址：无素材，且不实例化 provider。""" +async def test_knowledge_base_no_url_falls_back_to_default_endpoint(monkeypatch): + """未配置接口地址：仍实例化 IntranetSearchProvider（base_url 传空串，由 provider 内部 + 回退默认内网端点），内网直连依然可用——恢复旧版 search_provider=intranet 的行为。""" _patch(monkeypatch, intranet_url="") out = await R.researcher_node(_state(), {"configurable": {}}) - assert out["material_package"]["materials"] == [] - assert _FakeIntranet.instances == [], "无地址时不应实例化内网 provider" + mats = out["material_package"]["materials"] + assert mats, "无地址时应回退默认内网端点，仍返回素材" + assert all(m["source_type"] == "intranet" for m in mats) + assert _FakeIntranet.instances, "应实例化 IntranetSearchProvider 直连内网接口" + assert _FakeIntranet.instances[0].kw.get("base_url") == "", "空地址应原样下传，由 provider 回退默认端点" @pytest.mark.asyncio