agentquad 0.3.0

package/src/search/index.js

@@ -0,0 +1,199 @@
+import {
+  ensureFtsConsistency,
+  initFtsTables,
+  rebuildWikiFts,
+} from './fts.js'
+
+const ALL_SCOPES = ['todos', 'comments', 'wiki', 'ai_sessions']
+
+/**
+ * 把用户输入的自然语言 query 转成 FTS5 MATCH 友好的表达式。
+ *
+ * 策略：
+ *   - 单词按空白切分；
+ *   - 每个 token 加通配 `*` 做前缀匹配；
+ *   - tokens 之间用 AND；
+ *   - 若 token 含 FTS5 特殊字符（引号/操作符），用 `"..."` 引起来；
+ *   - 过短（1 字符）和全标点 token 丢弃。
+ */
+function sanitizeFtsQuery(raw) {
+  const trimmed = String(raw || '').trim()
+  if (!trimmed) return ''
+  // FTS5 支持的语法字符：AND/OR/NOT/NEAR + 双引号 + 括号 + *
+  // 我们简化：先去掉所有引号和括号，保留其它
+  const stripped = trimmed.replace(/["()]/g, ' ').replace(/\s+/g, ' ').trim()
+  const parts = stripped.split(/\s+/)
+  const terms = []
+  for (const part of parts) {
+    const clean = part.replace(/[*]+$/g, '')  // 先剥尾部星号避免 **
+    if (!clean) continue
+    if (clean.length === 0) continue
+    // 安全起见，所有 token 都套双引号 + 尾部加 *（前缀搜索）
+    // 双引号里双引号转义
+    const quoted = `"${clean.replace(/"/g, '""')}"`
+    terms.push(`${quoted}*`)
+  }
+  return terms.join(' AND ')
+}
+
+/**
+ * 工厂：创建 search service。
+ *
+ * 职责：
+ *   1. 启动时 initFtsTables + ensureFtsConsistency
+ *   2. 对外暴露 search({ query, scopes, includeArchived, limit })
+ *   3. 暴露 reindexWiki() 供 wiki run 完成时调用
+ */
+export function createSearchService({ db, wikiDir } = {}) {
+  if (!db) throw new Error('db_required')
+  const dbHandle = db.raw
+
+  function init() {
+    initFtsTables(dbHandle)
+    return ensureFtsConsistency(dbHandle, { wikiDir })
+  }
+
+  function reindexWiki() {
+    return rebuildWikiFts(dbHandle, { wikiDir })
+  }
+
+  function search({ query, scopes, includeArchived = false, limit = 20 } = {}) {
+    if (!query || !String(query).trim()) throw new Error('query_required')
+    const cappedLimit = Math.max(1, Math.min(Number(limit) || 20, 100))
+    const useScopes = Array.isArray(scopes) && scopes.length
+      ? scopes.filter((s) => ALL_SCOPES.includes(s))
+      : ALL_SCOPES.slice()
+    if (useScopes.length === 0) return { total: 0, results: [] }
+    const match = sanitizeFtsQuery(query)
+    if (!match) return { total: 0, results: [] }
+
+    const perScopeResults = []
+
+    if (useScopes.includes('todos')) {
+      // 注意 bm25 越小越相关，所以我们转成 1/(1+bm25) 归一化
+      const rows = dbHandle.prepare(`
+        SELECT f.todo_id AS todoId,
+               snippet(todos_fts, 1, '<mark>', '</mark>', '…', 12) AS titleSnippet,
+               snippet(todos_fts, 2, '<mark>', '</mark>', '…', 16) AS descSnippet,
+               bm25(todos_fts) AS bm25
+        FROM todos_fts f
+        JOIN todos t ON t.id = f.todo_id
+        WHERE todos_fts MATCH ? ${includeArchived ? '' : 'AND t.archived_at IS NULL'}
+        ORDER BY bm25 ASC
+        LIMIT ?
+      `).all(match, cappedLimit * 2)
+      for (const r of rows) {
+        perScopeResults.push({
+          scope: 'todos',
+          todoId: r.todoId,
+          snippet: r.descSnippet || r.titleSnippet || '',
+          title: null, // 前端 / MCP 再 JOIN 拿完整 title
+          score: 1 / (1 + Number(r.bm25)),
+        })
+      }
+    }
+
+    if (useScopes.includes('comments')) {
+      const rows = dbHandle.prepare(`
+        SELECT f.comment_id AS commentId,
+               f.todo_id AS todoId,
+               snippet(comments_fts, 2, '<mark>', '</mark>', '…', 16) AS snippet,
+               bm25(comments_fts) AS bm25
+        FROM comments_fts f
+        JOIN todos t ON t.id = f.todo_id
+        WHERE comments_fts MATCH ? ${includeArchived ? '' : 'AND t.archived_at IS NULL'}
+        ORDER BY bm25 ASC
+        LIMIT ?
+      `).all(match, cappedLimit * 2)
+      for (const r of rows) {
+        perScopeResults.push({
+          scope: 'comments',
+          todoId: r.todoId,
+          commentId: r.commentId,
+          snippet: r.snippet || '',
+          score: 1 / (1 + Number(r.bm25)),
+        })
+      }
+    }
+
+    if (useScopes.includes('wiki')) {
+      const rows = dbHandle.prepare(`
+        SELECT f.todo_id AS todoId,
+               snippet(wiki_fts, 1, '<mark>', '</mark>', '…', 16) AS snippet,
+               bm25(wiki_fts) AS bm25
+        FROM wiki_fts f
+        JOIN todos t ON t.id = f.todo_id
+        WHERE wiki_fts MATCH ? ${includeArchived ? '' : 'AND t.archived_at IS NULL'}
+        ORDER BY bm25 ASC
+        LIMIT ?
+      `).all(match, cappedLimit * 2)
+      for (const r of rows) {
+        perScopeResults.push({
+          scope: 'wiki',
+          todoId: r.todoId,
+          snippet: r.snippet || '',
+          score: 1 / (1 + Number(r.bm25)),
+        })
+      }
+    }
+
+    if (useScopes.includes('ai_sessions')) {
+      const rows = dbHandle.prepare(`
+        SELECT f.todo_id    AS todoId,
+               f.session_id AS sessionId,
+               snippet(ai_sessions_fts, 2, '<mark>', '</mark>', '…', 12) AS labelSnip,
+               snippet(ai_sessions_fts, 3, '<mark>', '</mark>', '…', 16) AS commandSnip,
+               bm25(ai_sessions_fts) AS bm25
+        FROM ai_sessions_fts f
+        JOIN todos t ON t.id = f.todo_id
+        WHERE ai_sessions_fts MATCH ? ${includeArchived ? '' : 'AND t.archived_at IS NULL'}
+        ORDER BY bm25 ASC
+        LIMIT ?
+      `).all(match, cappedLimit * 2)
+      for (const r of rows) {
+        perScopeResults.push({
+          scope: 'ai_sessions',
+          todoId: r.todoId,
+          sessionId: r.sessionId,
+          snippet: r.commandSnip || r.labelSnip || '',
+          score: 1 / (1 + Number(r.bm25)),
+        })
+      }
+    }
+
+    // 全局排序 + 截断
+    perScopeResults.sort((a, b) => b.score - a.score)
+    const top = perScopeResults.slice(0, cappedLimit)
+
+    // 把 todoId → title 查一次，避免 N+1
+    const todoIds = [...new Set(top.map((r) => r.todoId))]
+    const titleMap = new Map()
+    if (todoIds.length) {
+      const placeholders = todoIds.map(() => '?').join(',')
+      const rows = dbHandle.prepare(`
+        SELECT id, title, archived_at FROM todos WHERE id IN (${placeholders})
+      `).all(...todoIds)
+      for (const r of rows) {
+        titleMap.set(r.id, { title: r.title, archived: r.archived_at != null })
+      }
+    }
+    for (const r of top) {
+      const info = titleMap.get(r.todoId)
+      if (info) {
+        r.todoTitle = info.title
+        r.archived = info.archived
+      }
+    }
+
+    return {
+      total: perScopeResults.length,
+      results: top,
+    }
+  }
+
+  return {
+    init,
+    reindexWiki,
+    search,
+  }
+}

package/src/search/transcripts.js

@@ -0,0 +1,148 @@
+import { existsSync, readFileSync, statSync } from 'node:fs'
+import { join } from 'node:path'
+
+/**
+ * 在 AI 会话日志文件里做逐行关键词扫描。不使用 FTS，纯 Node fs。
+ *
+ * 查 transcripts 时的常见约束：
+ *   - 单个 log 文件可能几十 MB（node-pty 输出量大）；
+ *   - 查询返回太多命中会把 LLM 上下文拉爆。
+ * 所以这里用严格的 maxMatches / perFileLimit 保护。
+ */
+export function createTranscriptScanner({ db, logDir } = {}) {
+  if (!db) throw new Error('db_required')
+  if (!logDir) throw new Error('logDir_required')
+
+  /**
+   * 从 todo.ai_session 拿出所有 sessionId + 起止时间。
+   * 返回 [{ sessionId, startedAt, completedAt, tool, todoId, todoTitle }]
+   */
+  function collectCandidates({ todoId, afterTs, beforeTs }) {
+    const todos = todoId
+      ? [db.getTodo(todoId)].filter(Boolean)
+      : db.listTodos({ archived: 'all' })
+    const out = []
+    for (const t of todos) {
+      if (!Array.isArray(t.aiSessions)) continue
+      for (const s of t.aiSessions) {
+        if (!s?.sessionId) continue
+        const start = s.startedAt || 0
+        if (afterTs && start < afterTs) continue
+        if (beforeTs && start > beforeTs) continue
+        out.push({
+          sessionId: s.sessionId,
+          todoId: t.id,
+          todoTitle: t.title,
+          startedAt: s.startedAt || null,
+          completedAt: s.completedAt || null,
+          tool: s.tool || null,
+        })
+      }
+    }
+    return out
+  }
+
+  /**
+   * 对单个 log 文件扫 query。返回至多 perFileLimit 条命中。
+   * 命中条目：{ lineNumber, beforeLines[], matchLine, afterLines[] }
+   */
+  function scanFile(absPath, needle, { perFileLimit = 5, contextBefore = 1, contextAfter = 1 } = {}) {
+    if (!existsSync(absPath)) return []
+    let buf = ''
+    try {
+      const st = statSync(absPath)
+      // 文件超大（>8MB）只读尾部 4MB，避免 OOM
+      if (st.size > 8 * 1024 * 1024) {
+        const fd = readFileSync(absPath)
+        buf = fd.slice(fd.length - 4 * 1024 * 1024).toString('utf8')
+      } else {
+        buf = readFileSync(absPath, 'utf8')
+      }
+    } catch {
+      return []
+    }
+    const lines = buf.split('\n')
+    const lowerNeedle = needle.toLowerCase()
+    const hits = []
+    for (let i = 0; i < lines.length && hits.length < perFileLimit; i++) {
+      if (lines[i].toLowerCase().includes(lowerNeedle)) {
+        hits.push({
+          lineNumber: i + 1,
+          beforeLines: lines.slice(Math.max(0, i - contextBefore), i),
+          matchLine: lines[i],
+          afterLines: lines.slice(i + 1, Math.min(lines.length, i + 1 + contextAfter)),
+        })
+      }
+    }
+    return hits
+  }
+
+  /**
+   * 主入口。
+   */
+  function search({
+    query,
+    todoId,
+    afterTs,
+    beforeTs,
+    maxMatches = 30,
+    perFileLimit = 5,
+    contextBefore = 1,
+    contextAfter = 1,
+  } = {}) {
+    if (!query || !String(query).trim()) throw new Error('query_required')
+    const q = String(query).trim()
+    const candidates = collectCandidates({ todoId, afterTs, beforeTs })
+    const matches = []
+    let scanned = 0
+    for (const c of candidates) {
+      if (matches.length >= maxMatches) break
+      const file = join(logDir, `${c.sessionId}.log`)
+      scanned += 1
+      const hits = scanFile(file, q, {
+        perFileLimit: Math.min(perFileLimit, Math.max(1, maxMatches - matches.length)),
+        contextBefore,
+        contextAfter,
+      })
+      for (const h of hits) {
+        matches.push({ ...h, sessionId: c.sessionId, todoId: c.todoId, todoTitle: c.todoTitle, tool: c.tool })
+        if (matches.length >= maxMatches) break
+      }
+    }
+    return {
+      query: q,
+      scannedFiles: scanned,
+      totalMatches: matches.length,
+      matches,
+    }
+  }
+
+  /**
+   * 读取单个会话 transcript 的完整内容（可截断）。
+   * maxChars：返回字符数上限（粗略近似 token，默认 32000 ≈ 8k tokens）
+   */
+  function readSession({ sessionId, maxChars = 32_000 } = {}) {
+    if (!sessionId) throw new Error('sessionId_required')
+    const file = join(logDir, `${sessionId}.log`)
+    if (!existsSync(file)) {
+      return { exists: false, body: null }
+    }
+    const st = statSync(file)
+    const full = readFileSync(file, 'utf8')
+    if (full.length <= maxChars) {
+      return { exists: true, body: full, bytes: st.size, truncated: false, path: file }
+    }
+    const tail = full.slice(full.length - maxChars)
+    const truncatedChars = full.length - maxChars
+    return {
+      exists: true,
+      body: `…[truncated ${truncatedChars} chars from the start]\n${tail}`,
+      bytes: st.size,
+      truncated: true,
+      droppedChars: truncatedChars,
+      path: file,
+    }
+  }
+
+  return { search, readSession }
+}