docs: add P6 AI knowledge context implementation plan for v0.7.0

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-07 14:24:19 +08:00
parent b5ae04af18
commit 2afbb83c43
1 changed files with 927 additions and 0 deletions
@@ -0,0 +1,927 @@
 # 笔临 P6 AI 知识上下文整合 Implementation Plan
 > **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
 **Goal:** 交付 v0.7.0：已审核知识库条目半自动推荐 + 用户勾选确认 + 注入对话/交互/自动/向导四种 AI 模式；修复写作流 `contextJson` 仅存标题摘要的问题。
 **Architecture:** 新增 `KnowledgeRetrievalService` 对 `approved` 知识条目综合打分；扩展 `AiContextBinding.knowledgeIds` 与 `AiContextBuilderService` 合并知识块；`FlowContextPayload` 统一写作流上下文存储；`KnowledgeSuggestList` + `ContextEditorModal` 知识 Tab 供四种模式共用。测试 **禁止 Mock** AI 集成/E2E 打真实 LM Studio。
 **Tech Stack:** Electron 36、electron-vite、React 18、TypeScript、node:sqlite、Zustand、i18next、Vitest、Playwright
 **Spec:** `docs/superpowers/specs/2026-07-07-bilin-p6-ai-context-design.md`
 **LM Studio 默认：** `baseUrl=http://127.0.0.1:1234`，`model=gemma-4-e4b-it`
 ---
 ## File Map
 | 文件 | 职责 |
 |------|------|
 | `src/shared/types.ts` | `KnowledgeSuggestOptions`、`ScoredKnowledge`、`FlowContextPayload`；扩展 `AiContextBinding` |
 | `src/shared/ipc-channels.ts` | `KNOWLEDGE_SUGGEST_CONTEXT` |
 | `src/main/services/knowledge-retrieval.service.ts` | 打分 + TopN 推荐 |
 | `src/main/services/ai-context-builder.service.ts` | knowledge 块 + 预算裁剪 |
 | `src/main/services/flow-context.util.ts` | `buildFlowContextPayload()` 共享 |
 | `src/main/services/interactive-writing.service.ts` | `readContextText` → `systemPrompt` |
 | `src/main/db/repositories/ai-session.repo.ts` | `parseContext` 补 `knowledgeIds` |
 | `src/main/ipc/handlers/knowledge.handler.ts` | `suggestContext` |
 | `src/main/ipc/handlers/interactive.handler.ts` | `FlowContextPayload` |
 | `src/main/ipc/handlers/auto.handler.ts` | `readContextText` 修复 |
 | `src/main/ipc/handlers/wizard.handler.ts` | 同上 + preview 知识列表 |
 | `src/main/ipc/handlers/ai.handler.ts` | binding 默认 `knowledgeIds: []` |
 | `src/main/services/global-settings.ts` | `knowledgeAutoSuggest` / `knowledgeSuggestTopN` 默认值 |
 | `src/preload/index.ts` | `knowledge.suggestContext` |
 | `src/shared/electron-api.d.ts` | 类型 |
 | `src/renderer/components/ai/KnowledgeSuggestList.tsx` | 共享勾选 UI |
 | `src/renderer/components/ai/ContextEditorModal.tsx` | 知识 Tab + 推荐区 |
 | `src/renderer/components/ai/InteractivePanel.tsx` | context_confirm 知识区 |
 | `src/renderer/components/ai/AutoPanel.tsx` | goal 后知识确认 |
 | `src/renderer/components/ai/WizardPanel.tsx` | wizard_context 知识区 |
 | `src/renderer/lib/ai-context-summary.ts` | 摘要含知识计数 |
 | `src/renderer/stores/useInteractiveStore.ts` | binding 含 knowledgeIds |
 | `public/locales/zh-CN/translation.json` | P6 文案 |
 | `public/locales/en/translation.json` | P6 文案 |
 | `tests/main/knowledge-retrieval.test.ts` | UT-CTX-01/04 |
 | `tests/main/ai-context-builder.test.ts` | UT-CTX-02/03（扩展） |
 | `tests/main/flow-context.test.ts` | FlowContextPayload 单元 |
 | `tests/main/ai-context-integration.test.ts` | IT-CTX-01/02/03 |
 | `e2e/ai-context.spec.ts` | E2E-CTX-* |
 ---
 ## Task 1: 类型 + IPC 通道 + 默认 binding
 **Files:**
 - Modify: `src/shared/types.ts`
 - Modify: `src/shared/ipc-channels.ts`
 - Modify: `src/main/db/repositories/ai-session.repo.ts`
 - [ ] **Step 1: 扩展 types**
 `src/shared/types.ts` 追加：
 ```typescript
 export interface KnowledgeSuggestOptions {
  currentChapterId?: string
  goalText?: string
  topN?: number
 }
 export interface ScoredKnowledge {
  entry: KnowledgeEntry
  score: number
  reasons: string[]
 }
 export interface FlowContextPayload {
  binding: AiContextBinding
  systemPrompt: string
  contextSummary: string
 }
 ```
 扩展 `AiContextBinding`：
 ```typescript
 export interface AiContextBinding {
  chapterIds: string[]
  outlineIds: string[]
  settingIds: string[]
  inspirationIds: string[]
  knowledgeIds: string[]
 }
 ```
 扩展 `ContextPreviewKind`：`| 'knowledge'`
 扩展 `ContextPreviewItem`：`knowledgeType?: KnowledgeType`
 扩展 `GlobalSettings`（可选，本 Task 仅类型，Task 2 实现默认值）：
 ```typescript
 knowledgeAutoSuggest: boolean
 knowledgeSuggestTopN: number
 ```
 - [ ] **Step 2: IPC 通道**
 `src/shared/ipc-channels.ts`：
 ```typescript
 KNOWLEDGE_SUGGEST_CONTEXT: 'knowledge:suggestContext',
 ```
 - [ ] **Step 3: ai-session.repo 默认 knowledgeIds**
 ```typescript
 const EMPTY_CONTEXT: AiContextBinding = {
  chapterIds: [],
  outlineIds: [],
  settingIds: [],
  inspirationIds: [],
  knowledgeIds: []
 }
 function parseContext(json: string): AiContextBinding {
  try {
    const parsed = JSON.parse(json) as Partial<AiContextBinding>
    return {
      chapterIds: parsed.chapterIds ?? [],
      outlineIds: parsed.outlineIds ?? [],
      settingIds: parsed.settingIds ?? [],
      inspirationIds: parsed.inspirationIds ?? [],
      knowledgeIds: parsed.knowledgeIds ?? []
    }
  } catch {
    return { ...EMPTY_CONTEXT }
  }
 }
 ```
 - [ ] **Step 4: 全局修补 EMPTY_BINDING**
 搜索 `inspirationIds: []` 且无 `knowledgeIds` 的 `AiContextBinding` 字面量（`ContextEditorModal.tsx`、`useInteractiveStore.ts` 等），全部补上 `knowledgeIds: []`。
 - [ ] **Step 5: build 验证**
 ```bash
 npm run build
 ```
 - [ ] **Step 6: Commit**
 ```bash
 git add src/shared/types.ts src/shared/ipc-channels.ts src/main/db/repositories/ai-session.repo.ts
 git commit -m "feat: extend AiContextBinding with knowledgeIds for P6"
 ```
 ---
 ## Task 2: KnowledgeRetrievalService
 **Files:**
 - Create: `src/main/services/knowledge-retrieval.service.ts`
 - Create: `tests/main/knowledge-retrieval.test.ts`
 - [ ] **Step 1: UT-CTX-01 / UT-CTX-04 失败测试**
 `tests/main/knowledge-retrieval.test.ts`：
 ```typescript
 import { describe, it, expect, afterEach } from 'vitest'
 import { DatabaseSync } from 'node:sqlite'
 import { migrate } from '../../src/main/db/migrate'
 import { VolumeRepository } from '../../src/main/db/repositories/volume.repo'
 import { ChapterRepository } from '../../src/main/db/repositories/chapter.repo'
 import { KnowledgeRepository } from '../../src/main/db/repositories/knowledge.repo'
 import { KnowledgeRetrievalService } from '../../src/main/services/knowledge-retrieval.service'
 import type { SqliteDb } from '../../src/main/db/types'
 describe('KnowledgeRetrievalService', () => {
  let db: SqliteDb
  afterEach(() => db?.close())
  it('UT-CTX-01: ranks by importance + chapter proximity + goal match', () => {
    db = new DatabaseSync(':memory:')
    migrate(db)
    const volId = new VolumeRepository(db).create('卷一', 0).id
    const chapterRepo = new ChapterRepository(db)
    const cur = chapterRepo.create(volId, '当前章', 5, '<p>x</p>')
    const near = chapterRepo.create(volId, '近章', 4, '<p>x</p>')
    chapterRepo.create(volId, '远章', 0, '<p>x</p>')
    const repo = new KnowledgeRepository(db)
    const low = repo.create({
      type: 'fact',
      title: '无关事实',
      content: '普通内容',
      importance: 2,
      status: 'approved'
    })
    const high = repo.create({
      type: 'foreshadow',
      title: '测灵石异常',
      content: '测灵石在入门考核中发热',
      importance: 5,
      status: 'approved',
      foreshadowStatus: 'buried',
      sourceChapterId: near.id
    })
    const service = new KnowledgeRetrievalService(db)
    const result = service.suggest({
      currentChapterId: cur.id,
      goalText: '测灵石入门考核',
      topN: 5
    })
    expect(result[0].entry.id).toBe(high.id)
    expect(result[0].score).toBeGreaterThan(
      result.find((r) => r.entry.id === low.id)!.score
    )
    expect(result[0].reasons.length).toBeGreaterThan(0)
  })
  it('UT-CTX-04: resolved foreshadow scores lower than buried', () => {
    db = new DatabaseSync(':memory:')
    migrate(db)
    const volId = new VolumeRepository(db).create('卷一', 0).id
    const chId = new ChapterRepository(db).create(volId, '章', 0).id
    const repo = new KnowledgeRepository(db)
    const buried = repo.update(
      repo.create({
        type: 'foreshadow',
        title: '埋设伏笔',
        importance: 3,
        status: 'approved',
        foreshadowStatus: 'buried',
        sourceChapterId: chId
      }).id,
      { status: 'approved' }
    )
    const resolved = repo.create({
      type: 'foreshadow',
      title: '已回收伏笔',
      importance: 3,
      status: 'approved',
      foreshadowStatus: 'resolved',
      sourceChapterId: chId
    })
    const service = new KnowledgeRetrievalService(db)
    const scores = service.suggest({ currentChapterId: chId, topN: 10 })
    const buriedScore = scores.find((s) => s.entry.id === buried.id)!.score
    const resolvedScore = scores.find((s) => s.entry.id === resolved.id)!.score
    expect(buriedScore).toBeGreaterThan(resolvedScore)
  })
 })
 ```
 - [ ] **Step 2: 运行确认 FAIL**
 ```bash
 npm run test -- tests/main/knowledge-retrieval.test.ts
 ```
 - [ ] **Step 3: 实现 KnowledgeRetrievalService**
 `src/main/services/knowledge-retrieval.service.ts` 核心：
 ```typescript
 import type { KnowledgeEntry, KnowledgeSuggestOptions, ScoredKnowledge } from '../../shared/types'
 import type { SqliteDb } from '../db/types'
 import { ChapterRepository } from '../db/repositories/chapter.repo'
 import { KnowledgeRepository } from '../db/repositories/knowledge.repo'
 const STOP_WORDS = new Set(['的', '了', '在', '是', '与', '和', 'the', 'a', 'an'])
 function tokenize(text: string): string[] {
  return text
    .toLowerCase()
    .split(/[\s,，。！？、；;:：]+/)
    .map((t) => t.trim())
    .filter((t) => t.length >= 2 && !STOP_WORDS.has(t))
 }
 function overlapRatio(a: string, b: string): number {
  const ta = new Set(tokenize(a))
  const tb = tokenize(b)
  if (ta.size === 0 || tb.length === 0) return 0
  let hit = 0
  for (const t of tb) if (ta.has(t)) hit++
  return hit / tb.length
 }
 function chapterProximityBonus(
  entry: KnowledgeEntry,
  currentSort: number | null,
  chapterSort: Map<string, number>
 ): number {
  if (currentSort == null) return 0
  const ids = [entry.sourceChapterId, entry.lastMentionChapterId].filter(Boolean) as string[]
  if (ids.length === 0) return 0
  let best = Infinity
  for (const id of ids) {
    const order = chapterSort.get(id)
    if (order != null) best = Math.min(best, Math.abs(currentSort - order))
  }
  if (best === Infinity) return 0
  if (best === 0) return 30
  if (best <= 3) return 20
  if (best <= 10) return 10
  return 0
 }
 export class KnowledgeRetrievalService {
  constructor(private db: SqliteDb) {}
  suggest(opts: KnowledgeSuggestOptions = {}): ScoredKnowledge[] {
    const topN = opts.topN ?? 8
    const entries = new KnowledgeRepository(this.db).list({ status: 'approved' })
    if (entries.length === 0) return []
    const chapters = new ChapterRepository(this.db).list()
    const sortMap = new Map(chapters.map((c) => [c.id, c.sortOrder]))
    const currentSort = opts.currentChapterId
      ? (sortMap.get(opts.currentChapterId) ?? null)
      : null
    const goalText = opts.goalText ?? ''
    const scored: ScoredKnowledge[] = entries.map((entry) => {
      let score = entry.importance * 20
      const reasons: string[] = []
      const prox = chapterProximityBonus(entry, currentSort, sortMap)
      score += prox
      if (prox > 0) reasons.push('knowledge.reason.chapterRelevant')
      if (entry.type === 'foreshadow') {
        const fs = entry.foreshadowStatus ?? 'buried'
        if (fs === 'buried') {
          score += 25
          reasons.push('knowledge.reason.foreshadowBuried')
        } else if (fs === 'partial') score += 15
      }
      if (goalText) {
        const text = `${entry.title} ${entry.content}`
        const ratio = overlapRatio(goalText, text)
        score += Math.round(ratio * 30)
        if (ratio > 0.2) reasons.push('knowledge.reason.goalMatch')
      }
      if (entry.importance >= 4) reasons.push('knowledge.reason.highImportance')
      return { entry, score, reasons: [...new Set(reasons)] }
    })
    return scored.sort((a, b) => b.score - a.score).slice(0, topN)
  }
 }
 ```
 - [ ] **Step 4: 运行 PASS**
 ```bash
 npm run test -- tests/main/knowledge-retrieval.test.ts
 ```
 - [ ] **Step 5: Commit**
 ```bash
 git add src/main/services/knowledge-retrieval.service.ts tests/main/knowledge-retrieval.test.ts
 git commit -m "feat: add KnowledgeRetrievalService with scoring for context suggest"
 ```
 ---
 ## Task 3: 扩展 AiContextBuilder + flow-context 工具
 **Files:**
 - Modify: `src/main/services/ai-context-builder.service.ts`
 - Create: `src/main/services/flow-context.util.ts`
 - Modify: `tests/main/ai-context-builder.test.ts`
 - Create: `tests/main/flow-context.test.ts`
 - [ ] **Step 1: UT-CTX-02 / UT-CTX-03 测试**
 在 `tests/main/ai-context-builder.test.ts` 追加：
 ```typescript
 import { KnowledgeRepository } from '../../src/main/db/repositories/knowledge.repo'
 it('UT-CTX-02: build includes knowledge preview items', () => {
  const know = new KnowledgeRepository(db)
  const entry = know.create({
    type: 'foreshadow',
    title: '测灵石',
    content: '石头发热',
    status: 'approved'
  })
  const result = aiContextBuilder.build(
    { chapterIds: [], outlineIds: [], settingIds: [], inspirationIds: [], knowledgeIds: [entry.id] },
    registry,
    bookId,
    '作者'
  )
  expect(result.preview.some((p) => p.kind === 'knowledge')).toBe(true)
  expect(result.systemPrompt).toContain('【知识·伏笔】')
  expect(result.systemPrompt).toContain('测灵石')
 })
 it('UT-CTX-03: caps knowledge section at 4000 chars', () => {
  const know = new KnowledgeRepository(db)
  const ids: string[] = []
  for (let i = 0; i < 20; i++) {
    const e = know.create({
      type: 'fact',
      title: `条目${i}`,
      content: '长'.repeat(400),
      importance: 5,
      status: 'approved'
    })
    ids.push(e.id)
  }
  const result = aiContextBuilder.build(
    { chapterIds: [], outlineIds: [], settingIds: [], inspirationIds: [], knowledgeIds: ids },
    registry,
    bookId,
    '作者'
  )
  const knowledgeChars = result.preview
    .filter((p) => p.kind === 'knowledge')
    .reduce((s, p) => s + p.charCount, 0)
  expect(knowledgeChars).toBeLessThanOrEqual(4000)
 })
 ```
 `tests/main/flow-context.test.ts`：
 ```typescript
 import { describe, it, expect, beforeEach, afterEach } from 'vitest'
 import { mkdtempSync, rmSync } from 'fs'
 import { join } from 'path'
 import { tmpdir } from 'os'
 import { BookRegistryService } from '../../src/main/services/book-registry'
 import { buildFlowContextPayload } from '../../src/main/services/flow-context.util'
 import { KnowledgeRepository } from '../../src/main/db/repositories/knowledge.repo'
 describe('flow-context.util', () => {
  let userDir: string
  let registry: BookRegistryService
  let bookId: string
  beforeEach(() => {
    userDir = mkdtempSync(join(tmpdir(), 'bilin-flow-ctx-'))
    registry = new BookRegistryService(userDir)
    bookId = registry.create({ name: '书', category: '玄幻' }).id
  })
  afterEach(() => rmSync(userDir, { recursive: true, force: true }))
  it('buildFlowContextPayload stores systemPrompt and summary', () => {
    const db = registry.getDb(bookId)
    const entry = new KnowledgeRepository(db).create({
      type: 'fact',
      title: '测试知识',
      content: '内容',
      status: 'approved'
    })
    const payload = buildFlowContextPayload(
      {
        chapterIds: [],
        outlineIds: [],
        settingIds: [],
        inspirationIds: [],
        knowledgeIds: [entry.id]
      },
      registry,
      bookId,
      '作者'
    )
    expect(payload.systemPrompt).toContain('测试知识')
    expect(payload.contextSummary).toContain('1')
    db.close()
  })
 })
 ```
 - [ ] **Step 2: 运行确认 FAIL**
 ```bash
 npm run test -- tests/main/ai-context-builder.test.ts tests/main/flow-context.test.ts
 ```
 - [ ] **Step 3: 扩展 ai-context-builder**
 要点：
 - `KNOWLEDGE_CHAR_LIMIT = 500`，`KNOWLEDGE_TOTAL_LIMIT = 4000`
 - `knowledgeTypeLabel()` 映射五种类型
 - 在 `candidates` 循环后追加 knowledge 块，带 `sortKey: entry.importance * 1000`（高 important 先保留）
 - 知识区单独裁剪至 4000 后再参与总 16KB 裁剪
 - 所有现有 `build()` 调用 binding 需含 `knowledgeIds`（测试已补）
 `src/main/services/flow-context.util.ts`：
 ```typescript
 import type { AiContextBinding, FlowContextPayload } from '../../shared/types'
 import type { BookRegistryService } from './book-registry'
 import { aiContextBuilder } from './ai-context-builder.service'
 export function buildFlowContextPayload(
  binding: AiContextBinding,
  registry: BookRegistryService,
  bookId: string,
  penName: string
 ): FlowContextPayload {
  const built = aiContextBuilder.build(binding, registry, bookId, penName)
  const counts = { chapter: 0, outline: 0, setting: 0, inspiration: 0, knowledge: 0 }
  for (const p of built.preview) {
    if (p.kind in counts) counts[p.kind as keyof typeof counts]++
  }
  const contextSummary = `${counts.chapter}章·${counts.outline}纲·${counts.setting}设·${counts.inspiration}感·${counts.knowledge}知`
  return { binding, systemPrompt: built.systemPrompt, contextSummary }
 }
 ```
 - [ ] **Step 4: 运行 PASS**
 ```bash
 npm run test -- tests/main/ai-context-builder.test.ts tests/main/flow-context.test.ts
 ```
 - [ ] **Step 5: Commit**
 ```bash
 git add src/main/services/ai-context-builder.service.ts src/main/services/flow-context.util.ts tests/main/
 git commit -m "feat: extend AiContextBuilder with knowledge blocks and flow payload helper"
 ```
 ---
 ## Task 4: IPC suggestContext + preload
 **Files:**
 - Modify: `src/main/ipc/handlers/knowledge.handler.ts`
 - Modify: `src/preload/index.ts`
 - Modify: `src/shared/electron-api.d.ts`
 - Modify: `src/main/services/global-settings.ts`
 - Modify: `src/renderer/stores/useSettingsStore.ts`
 - [ ] **Step 1: knowledge.handler 追加**
 ```typescript
 ipcMain.handle(
  IPC.KNOWLEDGE_SUGGEST_CONTEXT,
  (
    _e,
    { bookId, opts }: { bookId: string; opts?: KnowledgeSuggestOptions }
  ) => wrap(() => {
    const settings = /* inject via handler param or read topN from settings */
    new KnowledgeRetrievalService(registry.getDb(bookId)).suggest(opts)
  })
 )
 ```
 `registerKnowledgeHandlers` 签名改为接收 `GlobalSettingsService`，`topN` 默认 `settings.get().knowledgeSuggestTopN ?? 8`。
 - [ ] **Step 2: global-settings 默认值**
 ```typescript
 knowledgeAutoSuggest: true,
 knowledgeSuggestTopN: 8,
 ```
 `get()` merge 补默认。
 - [ ] **Step 3: preload + electron-api.d.ts**
 ```typescript
 suggestContext: (
  bookId: string,
  opts?: KnowledgeSuggestOptions
 ): Promise<IpcResult<ScoredKnowledge[]>>
 ```
 - [ ] **Step 4: build**
 ```bash
 npm run build
 ```
 - [ ] **Step 5: Commit**
 ```bash
 git commit -m "feat: add knowledge suggestContext IPC and global settings defaults"
 ```
 ---
 ## Task 5: 写作流 systemPrompt 修复
 **Files:**
 - Modify: `src/main/services/interactive-writing.service.ts`
 - Modify: `src/main/ipc/handlers/interactive.handler.ts`
 - Modify: `src/main/ipc/handlers/auto.handler.ts`
 - Modify: `src/main/ipc/handlers/wizard.handler.ts`
 - Modify: `src/main/services/wizard-writing.service.ts`（`buildPreview` 追加知识列表）
 - [ ] **Step 1: interactive confirmContext 存 FlowContextPayload**
 `interactive.handler.ts` `INTERACTIVE_CONFIRM_CONTEXT`：
 ```typescript
 const payload = buildFlowContextPayload(binding, registry, bookId, settings.get().penName)
 const flow = service.confirmContext(flowId, payload)
 ```
 `InteractiveWritingService.confirmContext` 改为：
 ```typescript
 confirmContext(flowId: string, payload: FlowContextPayload): InteractiveFlow {
  this.repo.updateStep(flowId, 'plot_suggest', {
    contextJson: JSON.stringify(payload)
  })
  return this.repo.get(flowId)!
 }
 private readContextText(flowId: string): string {
  try {
    const parsed = JSON.parse(this.repo.getContextJson(flowId)) as FlowContextPayload
    return parsed.systemPrompt || '（无上下文）'
  } catch {
    return '（无上下文）'
  }
 }
 enrichFlow(flow: InteractiveFlow): InteractiveFlow {
  try {
    const parsed = JSON.parse(this.repo.getContextJson(flow.id)) as FlowContextPayload
    return { ...flow, contextSummary: parsed.contextSummary }
  } catch {
    return flow
  }
 }
 ```
 - [ ] **Step 2: auto.handler / wizard.handler readContextText**
 替换本地 `readContextText` 为读 `systemPrompt`：
 ```typescript
 function readContextText(registry, bookId, flowId): string {
  const row = registry.getDb(bookId)
    .prepare('SELECT context_json FROM interactive_flows WHERE id = ?')
    .get(flowId) as { context_json: string } | undefined
  try {
    const parsed = JSON.parse(row?.context_json ?? '{}') as FlowContextPayload
    return parsed.systemPrompt ?? '（无上下文）'
  } catch {
    return '（无上下文）'
  }
 }
 ```
 - [ ] **Step 3: wizard confirmContext 同 interactive**
 `wizard.handler.ts` `WIZARD_CONFIRM_CONTEXT` 使用 `buildFlowContextPayload`。
 - [ ] **Step 4: buildPreview 追加知识标题**
 `WizardWritingService.buildPreview` 返回前，从 `payload.binding.knowledgeIds` 查标题，append `\n将注入知识：A、B、C`。
 - [ ] **Step 5: auto planScenes 前自动 confirm**
 `AutoPanel` 在 `planScenes` 调用前，若 flow 无 `systemPrompt`，用 session binding + 默认推荐 knowledgeIds 调 `interactive.confirmContext` 等价逻辑（可新增 `auto:confirmContext` IPC 或复用 `buildFlowContextPayload` + `setContextSummary` 扩展为存完整 payload）。
 推荐：新增 `auto.handler` `AUTO_CONFIRM_CONTEXT` 与 interactive 对称。
 - [ ] **Step 6: Commit**
 ```bash
 git commit -m "fix: store full systemPrompt in writing flow contextJson for P6"
 ```
 ---
 ## Task 6: KnowledgeSuggestList + ContextEditorModal
 **Files:**
 - Create: `src/renderer/components/ai/KnowledgeSuggestList.tsx`
 - Modify: `src/renderer/components/ai/ContextEditorModal.tsx`
 - Modify: `src/renderer/lib/ai-context-summary.ts`
 - [ ] **Step 1: KnowledgeSuggestList**
 Props：
 ```typescript
 interface KnowledgeSuggestListProps {
  items: ScoredKnowledge[]
  selectedIds: string[]
  onToggle: (id: string) => void
 }
 ```
 渲染 checkbox + title + `t(\`knowledge.type.${entry.type}\`)` + reason badges `t(reason)`。
 - [ ] **Step 2: ContextEditorModal 扩展**
 新增 props：`goalText?: string`
 打开时：
 ```typescript
 const chapterId = useBookStore.getState().selectedChapterId ?? undefined
 const suggested = await ipcCall(() =>
  window.electronAPI.knowledge.suggestContext(bookId, { currentChapterId: chapterId, goalText })
 )
 // 默认 merge suggested ids into binding.knowledgeIds
 ```
 新增 Tab `knowledge`；顶部推荐区 + `context-refresh-suggest`；下方 approved 全列表（`knowledge.list(bookId, { status: 'approved' })`）可勾选。
 - [ ] **Step 3: formatContextSummary 含 knowledge**
 ```typescript
 const counts = { chapter: 0, outline: 0, setting: 0, inspiration: 0, knowledge: 0 }
 // ...
 return t('ai.contextSummary', { ...counts })
 ```
 更新 `ai.contextSummary` i18n 为 `{{chapters}}章·{{outlines}}纲·{{settings}}设·{{inspirations}}感·{{knowledge}}知`。
 - [ ] **Step 4: Commit**
 ```bash
 git commit -m "feat: add KnowledgeSuggestList and knowledge tab in context editor"
 ```
 ---
 ## Task 7: 交互 / 自动 / 向导 Panel 集成
 **Files:**
 - Modify: `src/renderer/components/ai/InteractivePanel.tsx`
 - Modify: `src/renderer/components/ai/AutoPanel.tsx`
 - Modify: `src/renderer/components/ai/WizardPanel.tsx`
 - Modify: `src/renderer/stores/useInteractiveStore.ts`
 - Modify: `src/renderer/stores/useAutoStore.ts`（confirmContext）
 - Modify: `src/renderer/components/ai/AiPanel.tsx`（传 goalText 空）
 - [ ] **Step 1: InteractivePanel context_confirm**
 - mount 时 `suggestContext` → state `suggested`
 - 内联 `<KnowledgeSuggestList>`，`data-testid="interactive-knowledge-suggest"`
 - 勾选同步到 session `context.knowledgeIds`（`ai.sessionUpdate`）
 - `confirmContextAndStart` 传含 `knowledgeIds` 的 binding
 - [ ] **Step 2: AutoPanel**
 - `goal_setup` 区域下方折叠「知识上下文」`data-testid="auto-knowledge-suggest"`
 - `goalText` = `chapterGoal` + 选中大纲 description
 - `planScenes` 前调用 `auto.confirmContext`（Task 5 新增 IPC）
 - [ ] **Step 3: WizardPanel wizard_context**
 - 同 Interactive 内联推荐列表 `data-testid="wizard-knowledge-suggest"`
 - `handleContextClose` 传 `goalText` = goal + styleNote
 - [ ] **Step 4: Commit**
 ```bash
 git commit -m "feat: integrate knowledge suggest into interactive auto wizard panels"
 ```
 ---
 ## Task 8: i18n
 **Files:**
 - Modify: `public/locales/zh-CN/translation.json`
 - Modify: `public/locales/en/translation.json`
 - [ ] **Step 1: 追加 spec §9 全部键 + `ai.contextSummary` 更新**
 - [ ] **Step 2: Commit**
 ```bash
 git commit -m "feat: add P6 AI context i18n strings"
 ```
 ---
 ## Task 9: 集成测试
 **Files:**
 - Create: `tests/main/ai-context-integration.test.ts`
 - [ ] **Step 1: IT-CTX-01**
 ```typescript
 it('IT-CTX-01: suggestContext returns entries when approved exist', () => {
  // setup approved entry → suggest → length >= 1
 })
 ```
 - [ ] **Step 2: IT-CTX-02**
 通过 `aiContextBuilder.build` 含 knowledgeIds 断言 `systemPrompt` 含 `【知识·`（无需 Electron IPC）。
 - [ ] **Step 3: IT-CTX-03（240s）**
 ```typescript
 it('IT-CTX-03: interactive confirm then suggestPlots uses knowledge context', async () => {
  // setup book + knowledge + interactive flow
  // confirmContext with knowledgeIds
  // suggestPlots returns 3 options
 }, 240_000)
 ```
 - [ ] **Step 4: 运行**
 ```bash
 npm run test -- tests/main/ai-context-integration.test.ts
 ```
 - [ ] **Step 5: Commit**
 ```bash
 git commit -m "test: add P6 AI context integration tests"
 ```
 ---
 ## Task 10: E2E
 **Files:**
 - Create: `e2e/ai-context.spec.ts`
 - [ ] **Step 1: E2E-CTX-01**
 ```typescript
 test('E2E-CTX-01: context editor knowledge tab shows suggestions', async () => {
  // create book → add approved knowledge via panel → open AI context editor → context-tab-knowledge visible
 })
 ```
 - [ ] **Step 2: E2E-CTX-02**
 ```typescript
 test('E2E-CTX-02: interactive context_confirm shows knowledge suggest', async () => {
  // open interactive → interactive-knowledge-suggest visible
 })
 ```
 - [ ] **Step 3: E2E-CTX-03（300s）**
 ```typescript
 test('E2E-CTX-03: chat with knowledge context gets AI reply', async () => {
  test.setTimeout(300_000)
  // add knowledge → select in context → send chat → non-empty reply
 })
 ```
 - [ ] **Step 4: 运行**
 ```bash
 npm run test:e2e -- e2e/ai-context.spec.ts
 ```
 - [ ] **Step 5: Commit**
 ```bash
 git commit -m "test: add P6 AI context E2E specs"
 ```
 ---
 ## Task 11: v0.7.0 交付
 **Files:**
 - Modify: `package.json`
 - Modify: `README.md`
 - Modify: `src/renderer/components/settings/SettingsPage.tsx`（about 版本号）
 - Modify: `docs/superpowers/plans/2026-07-07-bilin-p6-ai-context.md`（勾选完成项）
 - [ ] **Step 1: 版本号 `0.7.0`**
 - [ ] **Step 2: README 追加**
 ```markdown
 - 知识库 AI 上下文：半自动推荐 + 勾选注入四种写作模式（P6）
 ```
 - [ ] **Step 3: 全量测试**
 ```bash
 npm run build
 npm run test
 npm run test:e2e
 ```
 - [ ] **Step 4: Commit**
 ```bash
 git commit -m "feat: ship v0.7.0 with AI knowledge context integration"
 ```
 ---
 ## Spec Coverage Checklist
 | Spec 要求 | Task |
 |-----------|------|
 | KnowledgeRetrievalService 打分 | Task 2 |
 | AiContextBinding.knowledgeIds | Task 1 |
 | AiContextBuilder 知识块 | Task 3 |
 | FlowContextPayload / systemPrompt 修复 | Task 3, 5 |
 | knowledge:suggestContext IPC | Task 4 |
 | ContextEditorModal 知识 Tab | Task 6 |
 | 对话静默注入 | Task 5（AI_CHAT 已用 build）+ Task 6 |
 | 交互 context_confirm | Task 7 |
 | 自动写作知识确认 | Task 5, 7 |
 | 向导 wizard_context | Task 7 |
 | global-settings 默认值 | Task 4 |
 | UT-CTX-01~04 | Task 2, 3 |
 | IT-CTX-01~03 | Task 9 |
 | E2E-CTX-01~03 | Task 10 |
 | v0.7.0 DoD | Task 11 |