get-claudia 1.55.13 → 1.55.14

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to Claudia will be documented in this file.
 
+## 1.55.14 (2026-03-16)
+
+- **LaunchAgent no longer bakes in --project-dir** -- The standalone background daemon now starts without a `--project-dir` argument. This forces a plist content change for all existing installs, which triggers an automatic LaunchAgent reload on next `claudia setup`, picking up the current Python daemon code. Previously, the plist could be identical across updates, leaving old daemon code running indefinitely even after `pip install --upgrade`.
+- **Cleanup of orphaned empty hash databases** -- On each startup, if the database is already unified and empty hash-named DB files exist (created by stale old-code daemon processes), they are silently removed. Prevents phantom databases from accumulating in `~/.claudia/memory/`.
+- **Root cause:** After the v1.55 consolidation, users whose LaunchAgent was running old pre-unified-DB code would see scheduled jobs (consolidation, backups, decay) operating against an empty `6af67351bcfa.db` while all real memories lived in `claudia.db`. Health check showed `schema_version: 0` and `-1` counts.
+
 ## 1.55.13 (2026-03-16)
 
 - **Gmail & Calendar MCPs as standard options** -- The standalone `gmail` (`@gongrzhe/server-gmail-autoauth-mcp`) and `google-calendar` (`@gongrzhe/server-calendar-autoauth-mcp`) servers are now first-class options alongside workspace-mcp. Two paths for Google integration: Option A (lightweight, focused, fewer tools) and Option B (all-in-one workspace-mcp with Drive, Docs, Sheets, etc.). Both can coexist.

package/bin/index.js

@@ -1004,7 +1004,7 @@ async function main() {
 
     // Register LaunchAgent for standalone daemon (macOS only)
     if (daemonOk && process.platform === 'darwin') {
-      await ensureLaunchAgent(venvPython, targetPath);
+      await ensureLaunchAgent(venvPython);
     }
 
     // MCP Config step: verify .mcp.json is correct and check stdio server count
@@ -1575,7 +1575,7 @@ function checkMcpConfig(targetPath) {
  * The standalone daemon runs 24/7 for scheduled jobs (consolidation, decay, vault sync).
  * This is separate from the MCP daemon that Claude Code spawns per-session.
  */
-async function ensureLaunchAgent(venvPythonPath, projectDir) {
+async function ensureLaunchAgent(venvPythonPath) {
   const plistDir = join(homedir(), 'Library', 'LaunchAgents');
   const plistPath = join(plistDir, 'com.claudia.memory.plist');
 
@@ -1591,8 +1591,6 @@ async function ensureLaunchAgent(venvPythonPath, projectDir) {
         <string>-m</string>
         <string>claudia_memory</string>
         <string>--standalone</string>
-        <string>--project-dir</string>
-        <string>${projectDir}</string>
     </array>
     <key>WorkingDirectory</key>
     <string>${join(homedir(), '.claudia', 'daemon')}</string>

package/memory-daemon/claudia_memory/__main__.py

@@ -195,7 +195,16 @@ def _auto_consolidate() -> None:
                 fetch=True,
             )
             if rows and rows[0]["value"] == "true":
-                logger.debug("Database already unified, skipping consolidation")
+                # Unified. Clean up any empty hash DBs that stale daemon instances may have
+                # created (old standalone daemons running pre-unified-DB code create a fresh
+                # empty hash DB on startup if the original was deleted by consolidation).
+                all_hash_dbs = scan_hash_databases(memory_dir)
+                empty_dbs = [d for d in all_hash_dbs if not d["has_data"]]
+                if empty_dbs:
+                    logger.info(
+                        f"Removing {len(empty_dbs)} empty hash DB(s) left by stale standalone daemon"
+                    )
+                    cleanup_old_databases(memory_dir, empty_dbs)
                 return
         except Exception:
             pass  # _meta table might not exist yet

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "get-claudia",
-  "version": "1.55.13",
+  "version": "1.55.14",
   "description": "An AI assistant who learns how you work.",
   "keywords": [
     "claudia",