fix(hooks): use SessionEnd not Stop for Claude Code transcript extraction

docs/architecture.md

@@ -40,7 +40,7 @@
 
 3. **Snapshot** (pre-compact): The PreCompact hook (Claude Code, Kimi) saves a lightweight snapshot of the conversation before context compression.
 
-4. **Extract** (session end): The Stop hook launches a background ingestion process that parses the transcript, runs the encoding gate on each fact, and stores high-quality memories.
+4. **Extract** (session end): The SessionEnd hook launches a background ingestion process that parses the transcript, runs the encoding gate on each fact, and stores high-quality memories. (On Claude Code, the `SessionEnd` event fires only when the session actually terminates — `Stop` fires per-turn and is therefore the wrong trigger for transcript-end extraction.)
 
 ## Package Structure
 

install.sh

@@ -12,7 +12,7 @@
 #   4. Runs `truememory-mcp --setup` (code from PyPI) to auto-configure
 #      Claude Code and/or Claude Desktop. Set TRUEMEMORY_SKIP_SETUP=1 to skip.
 #   5. Runs `truememory-ingest install` to wire up lifecycle hooks
-#      (SessionStart, Stop, UserPromptSubmit, PreCompact) and merge
+#      (SessionStart, SessionEnd, UserPromptSubmit, PreCompact) and merge
 #      CLAUDE.md instructions so Claude uses TrueMemory proactively.
 #
 # Environment overrides:

truememory/ingest/CLAUDE_TEMPLATE.md

@@ -33,7 +33,7 @@ MEMORY.md may contain personal facts that were cached from earlier sessions. **D
 
 ## Background Processing
 - Memories are also extracted automatically from conversations via background processing.
-- The Stop hook captures the full transcript and runs deep extraction after sessions end.
+- The SessionEnd hook captures the full transcript and runs deep extraction when the session terminates.
 - You do NOT need to store everything manually — focus on in-conversation corrections and explicit preferences.
 - The background extractor handles: personal facts, preferences, decisions, temporal facts, and technical context.
 

truememory/ingest/cli.py

@@ -791,7 +791,7 @@ def _run_install(args):
     Hooks installed:
     - SessionStart: injects relevant memories as additionalContext
     - UserPromptSubmit: buffers user messages (kept for future use)
-    - Stop: triggers background fact extraction after each session
+    - SessionEnd: triggers background fact extraction when the session terminates
     - PreCompact: saves context snapshot before Claude compresses conversation
 
     Note: the event is named ``PreCompact`` in Claude Code's settings.json
@@ -809,7 +809,7 @@ def _run_install(args):
     hook_files = {
         "SessionStart": hooks_dir / "session_start.py",
         "UserPromptSubmit": hooks_dir / "user_prompt_submit.py",
-        "Stop": hooks_dir / "stop.py",
+        "SessionEnd": hooks_dir / "stop.py",
         "PreCompact": hooks_dir / "compact.py",
     }
     missing = [name for name, path in hook_files.items() if not path.exists()]
@@ -1083,7 +1083,7 @@ def _run_status(args):
         try:
             settings = json.loads(settings_path.read_text(encoding="utf-8"))
             hooks = settings.get("hooks", {})
-            expected = ["SessionStart", "UserPromptSubmit", "Stop", "PreCompact"]
+            expected = ["SessionStart", "UserPromptSubmit", "SessionEnd", "PreCompact"]
             installed = []
             missing = []
             for event in expected: