chore(m5): cut connector main IPC + preload + agent prompt

packages/app/src/main/acp.ts

@@ -870,50 +870,29 @@ export class AcpManager {
    */
   private buildPrompt(userQuery: string): string {
     return [
-      'You have access to a local knowledge base called Spool that indexes:',
-      '  1. The user\'s AI coding sessions (Claude Code, Codex CLI, Gemini CLI)',
-      '  2. Platform data synced via connectors (X/Twitter bookmarks, GitHub stars, etc.)',
+      'You have access to a local knowledge base called Spool that indexes the user\'s AI coding sessions (Claude Code, Codex CLI, Gemini CLI).',
       '',
       'The database is at ~/.spool/spool.db (SQLite with FTS5). You can query it directly with the `sqlite3` CLI.',
       '',
-      '── Agent session schema ──',
+      '── Schema ──',
       '  sources(id, name TEXT, base_path TEXT)  -- "claude", "codex", or "gemini"',
       '  projects(id, source_id, slug, display_path, display_name, last_synced)',
       '  sessions(id, project_id, source_id, session_uuid TEXT, title TEXT, started_at TEXT, ended_at TEXT, message_count INT, has_tool_use INT)',
       '  messages(id, session_id, source_id, role TEXT, content_text TEXT, timestamp TEXT, tool_names TEXT)',
       '  messages_fts(content_text)  -- FTS5 virtual table, content synced from messages',
       '',
-      '── Connector captures schema ──',
-      '  connector_sync_state(connector_id TEXT PRIMARY KEY, head_cursor, tail_cursor, tail_complete INT, last_forward_sync_at, total_synced INT, enabled INT, config_json TEXT)',
-      '  captures(id, source_id, capture_uuid TEXT, url TEXT, title TEXT, content_text TEXT, author TEXT, platform TEXT, platform_id TEXT, content_type TEXT, metadata TEXT, captured_at TEXT, raw_json TEXT)',
-      '  captures_fts(title, content_text)  -- FTS5 virtual table, content synced from captures',
-      '',
       'Example queries:',
       '  # FTS search on agent sessions',
       '  sqlite3 ~/.spool/spool.db "SELECT m.content_text, s.title, s.started_at, p.display_name FROM messages_fts f JOIN messages m ON m.id = f.rowid JOIN sessions s ON s.id = m.session_id JOIN projects p ON p.id = s.project_id WHERE messages_fts MATCH \'search terms\' ORDER BY rank LIMIT 10"',
       '',
       '  # Recent sessions',
       '  sqlite3 ~/.spool/spool.db "SELECT session_uuid, title, started_at, message_count FROM sessions ORDER BY started_at DESC LIMIT 20"',
       '',
-      '  # FTS search on captures (bookmarks, saved web content)',
-      '  sqlite3 ~/.spool/spool.db "SELECT c.title, c.author, c.url, c.content_text, c.platform, c.captured_at FROM captures_fts f JOIN captures c ON c.id = f.rowid WHERE captures_fts MATCH \'search terms\' ORDER BY rank LIMIT 10"',
-      '',
-      '  # List captures for a specific connector',
-      '  sqlite3 ~/.spool/spool.db "SELECT c.title, c.author, c.url, c.content_text, c.captured_at, c.platform FROM captures c JOIN capture_connectors cc ON cc.capture_id = c.id WHERE cc.connector_id = \'twitter-bookmarks\' ORDER BY c.captured_at DESC LIMIT 20"',
-      '',
-      '  # What connectors are enabled',
-      '  sqlite3 ~/.spool/spool.db "SELECT connector_id, total_synced, last_forward_sync_at FROM connector_sync_state WHERE enabled = 1"',
-      '',
       'Important:',
       '- Interpret the user\'s intent and decide what to search. Don\'t just match their exact words.',
-      '- For questions about bookmarks, saved content, or web platforms → query captures/captures_fts.',
-      '- For questions about coding sessions, projects, or what the user built → query messages/sessions.',
-      '- Treat source boundaries as part of the query semantics, not as an implementation detail.',
-      '- For connector captures, JOIN `capture_connectors cc ON cc.capture_id = c.id` and filter by `cc.connector_id` to distinguish different connectors for the same platform. One capture can belong to multiple connectors.',
-      '- If the user names a source, only return results from that source unless they explicitly ask for cross-source search.',
+      '- If the user names a specific source (claude/codex/gemini), only return results from that source unless they explicitly ask for cross-source search.',
       '- For cross-source questions, first identify the relevant sources, then query each source separately, confirm hits or no-hits per source, and only then merge them into one answer.',
       '- For temporal queries ("what did I do recently"), use explicit date filters and be conservative when comparing times across different sources.',
-      '- Different sources may store `captured_at` differently. For cross-source time questions, verify each source separately before merging results. Do not assume one source\'s time semantics apply to another.',
       '- You may run multiple queries to find relevant information.',
       '- Synthesize a concise answer. Reference specific items, URLs, or sessions when relevant.',
       '- Keep the result layer separate from the explanation layer: answer the user\'s question directly first, then add brief notes about what you checked if helpful.',