Commit Graph

31 Commits

Author SHA1 Message Date
MIST
fa5d406118
fix(wiki): agent 通过 wiki_create_page 写入的页面缺少 raw/chunks/embeddings/citations,界面无法识别与操作 (#475)
## 背景

Agent 通过 `wiki_create_page` 工具写入知识库的报告、分析结果等页面,虽然能在"Wiki 页面"列表中看到,但:

- **识别不到**:不出现在"原始材料"面板
- **不可操作**:"查看引用"按钮消失(`sourceRawIds` 为空)
- **不可处理**:无 raw 可 reprocess、无 chunks 导致语义检索查不到
- **不可下载**:无 raw 行,下载端点无数据

## 根因

`wiki_create_page` 只调用 `WikiPageService.createPage()` 写了一张 `mate_wiki_page` 表(`sourceRawIds=null`),跳过了 UI 上传文本路径的全部"消化副产物"——raw material 创建、chunks 切片、embeddings 生成、citations 构建、lineage 血缘、`WikiPageCreatedEvent` 事件发布。

## 解决方案

让 `wiki_create_page` 在落 page 之后,同步创建 raw material 并补齐全部消化副产物,但**不重跑 LLM 页面生成**(agent 已提供最终内容)。`WikiTool.wiki_create_page` 的同步返回(`ok / pageId / slug`)不受影响。

### 改动(4 个文件,+190 / -1)

| 文件 | 改动 |
|---|---|
| `WikiRawMaterialService.java` | 新增 `addAgentAuthored(kbId, title, content)`:创建 `sourceType="text"` 的 raw 行,状态置 `processing`(不发布 `WikiProcessingEvent`,避免触发 LLM 重消化),按 content hash 去重 |
| `WikiProcessingService.java` | 新增 `linkAgentPageToRaw(pageId, kbId, rawId, rawTitle, pageType)`:编排 ①`mergeSourceLineage`(血缘)②`deriveKnowledgeLayer` ③`persistChunks`(切片)④`embedMissingChunks` + `embedPage` ⑤`buildCitationsAsync` ⑥发布 `WikiPageCreatedEvent` ⑦raw 置 `completed` + `lastProcessedHash`。每步独立 try/catch,单点失败不阻塞其它 |
| `WikiTool.java` | `wiki_create_page` 在 `createPage` 后调用 `addAgentAuthored` + `linkAgentPageToRaw`。`processingService` 为可选注入(`@Autowired(required = false)`),测试上下文中为 null 时退化为旧行为 |
| `WikiPageService.java` | `deleteExclusiveBySourceRawId` 新增跳过 `lastUpdatedBy = "ai"` 的页面:agent 直接创作的页面不应在 reprocess 时被自动清理 |

### 修复后效果

| 能力 | 修复前 | 修复后 |
|---|---|---|
| Wiki 页面列表可见 |  |  |
| 原始材料面板可见(可识别) |  |  |
| 下载按钮 |  |  |
| 查看引用按钮(可操作) |  隐藏 |  |
| CitationDrawer 有内容 |  空 |  |
| 可 reprocess(可处理) |  |  |
| 语义检索可命中 |  |  chunks+embeddings |
| pipeline 触发器评估 |  |  |
| 页面内容 = agent 原文 |  | (不重跑 LLM) |
| UI "AI 生成"标记 |  | (`lastUpdatedBy = "ai"`) |

## 验证

全量 wiki 模块测试通过:`Tests run: 474, Failures: 0, Errors: 0, Skipped: 0`

## 风险

- `processingService` 为可选注入,测试/轻量上下文下退化为旧行为,**向后兼容无破坏**
- 新增两处数据库写入(raw + chunks),每次 `wiki_create_page` 增加 1 条 raw + N 条 chunk 行
- reprocess 行为:reprocess 时 `lastUpdatedBy = "ai"` 的页面被保留不清理,LLM 管线会从 raw content 重新消化生成额外概念页
2026-07-03 10:37:16 +08:00
倪程伟
f6156f6093 feat(webchat): expose agent-bound wiki pages to API-Key callers
Add GET /api/v1/channels/webchat/wiki/pages mirroring /skills, so
downstream integrators can build a [[slug]] picker UI that points the
LLM at specific wiki pages. The picker token format is the universal
Obsidian/Wikipedia wikilink convention; the LLM consumes [[slug]] via
the existing wiki_read_page(slug=...) tool, so no agent-runtime changes
are needed.

- AgentBindingResolver.getBoundKbIds(agentId): three-state mirror of
  getBoundSkillIds. null = no rows (fall through to workspace-wide KBs),
  Set.of() = explicitly scoped to zero KBs, non-empty = explicit scope.
- WebChatController.listWikiPages: API Key + visitorToken auth chain,
  agentId workspace anti-escalation, visibility excludes pageType=
  synthesis (LLM intermediate artifacts), 100-page cap forces keyword
  filter, response carries only display-level metadata.
- WebChatWikiPageView DTO: kbId/kbName/slug/title/summary/pageType;
  content/embedding/sourceRawIds deliberately stay admin-console-only.
- WikiTool.wiki_read_page @Tool description: document the [[slug]]
  convention so the LLM treats each token as a wiki-page reference.
- WebChatWikiPageListTest: 8 cases covering happy path, keyword filter,
  synthesis exclusion, anti-escalation, auth failures, cap behavior,
  and the no-binding → workspace-wide fallback.

Closes #381.
2026-06-20 07:21:07 +08:00
jack
88be1f748a
[#305] [Feature] Add knowledge base trust verification (#334)
Co-authored-by: SuperCoderMan521 <SuperCoderManqq.com>
2026-06-16 07:47:04 +08:00
倪程伟
28f2ba973d feat(wiki): honour KB pageType profile in transformations, agent pages & UI
Wiki page classification was only profile-aware in the main ingest pipeline.
Transformation outputs hard-coded "synthesis", agent-created pages were left
untyped, and the frontend hard-coded the built-in ten types for ordering,
colouring and labels — so custom/synthesis types sank to the bottom, rendered
grey and showed raw keys.

Backend:
- Add nullable target_page_type column to mate_wiki_transformation (V142,
  mysql + h2) plus the entity field and CRUD normalization (blank = use
  profile fallbackType; membership validated at save time, not edit time).
- Route transformation single-run + KB-aggregate page saves through
  WikiPageTypeProfileService.normalizePageType so output joins the KB
  classification; agent wiki_create_page now lands on the profile fallbackType
  instead of an untyped page.

Frontend:
- Load + parse the KB pageType profile into the wiki store (order, labels,
  fallbackType) on KB select / refresh.
- New useWikiPageType composable: profile-driven label (3-tier fallback) and
  colour (built-in fixed + deterministic hash palette for custom types).
- Sidebar grouping order, graph colouring, node panel, graph filter and the
  page header badge now follow the profile; transformation editor gains a
  target-type dropdown sourced from the profile when output target is a page.

Refs #292
2026-06-08 21:02:50 +08:00
matevip
1388b6eec8 feat(wiki): add wiki_update_page (in-place edit) and wiki_stale_pages tools 2026-05-31 08:00:19 +08:00
matevip
e18e0f1029 feat(wiki): per-agent pageType permission config API and pending-approval recording 2026-05-31 07:59:48 +08:00
matevip
551321b542 fix(wiki): make pageType permission service a mandatory dependency 2026-05-31 07:58:59 +08:00
matevip
cc8c9cf951 feat(wiki): knowledge-layer filter on wiki search 2026-05-31 07:57:18 +08:00
matevip
ec47d19bf8 feat(wiki): gate wiki write/mutate tools by pageType permission 2026-05-31 07:53:59 +08:00
matevip
502b8406a6 feat(wiki): per-agent pageType read permission gate for wiki tools 2026-05-31 07:53:48 +08:00
matevip
ff2620dfcf fix(wiki): harden kbName/kbId routing — ambiguous fail-closed, kbId param, prompt cleanup (#224) 2026-05-26 14:31:07 +08:00
matevip
d2b23c049c fix(wiki): let agents reach every visible KB via kbName + wiki_list_kbs (#224) 2026-05-26 13:53:32 +08:00
matevip
d629c945ee fix(wiki): return the new page id from the create-page tool 2026-05-19 20:06:33 +08:00
matevip
c6d51f8ced fix(wiki): resolve an agent's own knowledge base before shared ones 2026-05-19 16:53:12 +08:00
matevip
90936dba4f feat(wiki): cross-material transformation aggregator — map-reduce all runs of a template into one KB page 2026-05-12 14:10:54 +08:00
matevip
e1759e0e3c feat(wiki): page-input transformations — run a template against an existing page 2026-05-12 14:10:40 +08:00
matevip
bae9f68261 feat(wiki): user-defined transformation templates with optional auto-save to synthesis pages 2026-05-12 10:49:58 +08:00
matevip
58b49f6e20 feat(wiki): structured no-evidence compile + ops admin endpoints 2026-04-25 19:02:33 +08:00
matevip
1e62dbad47 feat(wiki): PR-7 archived soft-archive 2026-04-25 19:02:32 +08:00
matevip
b41496ed46 feat(wiki): PR-4 on-demand compile + multi-page read tools 2026-04-25 09:56:19 +08:00
matevip
1281153aa8 feat(wiki): PR-2 system pages — overview/log scaffold + locked + filters 2026-04-25 09:56:18 +08:00
matevip
9746271ea5 feat(wiki): PR-1c preprocessor + chunk metadata + search exposure 2026-04-25 09:56:18 +08:00
matevip
c50977785a fix(wiki): internal link navigation and source citation guidance 2026-04-24 06:55:20 +08:00
matevip
3d213eb281 chore: sync multiple commits from private dev
Covers 15 upstream commits (private mirror → public):

Multi-provider failover (RFC-009):
- PR-0: extract ChatModelBuilder strategy seam
- PR-1a: AvailableProviderPool data structure
- PR-1b: startup provider liveness probe + 4 protocol strategies
- PR-1c: wire AvailableProviderPool into runtime chat-model selection
- PR-1d: provider pool REST endpoint + UI badges
- PR-1e: manual reprobe trigger + auto-reprobe on provider config change
- PR-3: per-agent provider preferences (agents can override the
  org-wide fallback chain)

Wiki subsystem (RFC-029~033):
- Relation model, resilient background jobs, light-weight processing
  path, retrieval enhancement, frontend redesign (single landing commit)
- Follow-up fixes: null guards + stats query + i18n polish, move
  WikiProcessingJobMapper to repository/ for @MapperScan, align
  implementation with RFC-029~031 spec
- Copy pass: replace "富化 / enrich" wording with clearer "链接 / link"
- Style: switch enrich/repair buttons to @element-plus/icons-vue
2026-04-19 18:37:44 +08:00
matevip
1fdf87b31e feat(wiki): semantic hybrid search + chunk persistence + deep research pipeline 2026-04-16 17:11:31 +08:00
matevip
d5e3502829 fix(wiki): backend hardening + batch delete UI 2026-04-13 15:37:12 +08:00
matevip
e4679796b8 feat(wiki): performance + quality + delete redesign (RFC-008) 2026-04-13 09:35:26 +08:00
matevip
5124842526 feat(platform): workspace isolation, info architecture, and UI redesign 2026-04-09 16:29:02 +08:00
matevip
3d58a48eae feat(platform): add workspace foundation and channel execution upgrades 2026-04-09 10:29:16 +08:00
matevip
f6102c4733 fix(wiki): auto-resolve kbId, improve search and processing robustness 2026-04-08 16:09:22 +08:00
matevip
642360a773 feat(wiki): add LLM Wiki knowledge base system 2026-04-08 10:55:50 +08:00