fix(plugin): hook cleanup + identity + compact-summary flow (#5076)

mem0-plugin/scripts/on_session_start.sh

@@ -11,9 +11,34 @@
 # even if jq is missing or stdin is malformed.
 set -uo pipefail
 
+if [ -n "${MEM0_DEBUG:-}" ]; then
+  mkdir -p "$HOME/.mem0" && exec 2>>"$HOME/.mem0/hooks.log"
+fi
+
+# Skip the bootstrap entirely if no API key is configured -- the agent
+# would otherwise be told to call mem0 MCP tools that will all fail.
+if [ -z "${MEM0_API_KEY:-}" ]; then
+  exit 0
+fi
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+# shellcheck source=_identity.sh
+. "$SCRIPT_DIR/_identity.sh"
+
 INPUT=$(cat)
 SOURCE=$(echo "$INPUT" | jq -r '.source // "startup"' 2>/dev/null || echo "startup")
 
+# Identity line is emitted before every bootstrap variant so the agent
+# uses the same user_id the hooks resolved. Without this, the agent's
+# search_memories/add_memory MCP calls may bind to a different bucket
+# than what the hooks write to.
+echo "## Mem0 Identity"
+echo ""
+echo "Active user_id: \`$MEM0_RESOLVED_USER_ID\`"
+echo ""
+echo "Always include \`{\"user_id\": \"$MEM0_RESOLVED_USER_ID\"}\` (wrapped in an \`AND\` clause) in every \`search_memories\` filter and as \`user_id\` on every \`add_memory\` call. This keeps memories under one bucket regardless of which machine you're on."
+echo ""
+
 if [ "$SOURCE" = "startup" ]; then
   cat <<'EOF'
 ## Mem0 Session Bootstrap
@@ -40,14 +65,22 @@ Continue where you left off.
 EOF
 
 elif [ "$SOURCE" = "compact" ]; then
+  # Capture the just-generated compact summary in the background.
+  # PreCompact fires too early to see this entry; SessionStart-compact
+  # is the first place isCompactSummary=true is in the transcript.
+  echo "$INPUT" | python3 "$SCRIPT_DIR/capture_compact_summary.py" 2>/dev/null &
+
   cat <<'EOF'
 ## Mem0 Post-Compaction Recovery
 
-Context was just compacted. You may have lost important session context.
+Context was just compacted. The Claude Code-generated compact summary
+is being captured to mem0 in the background as `metadata.type=compact_summary`.
 
-1. Call `search_memories` with queries related to what you were working on to reload relevant knowledge.
-2. Check for any session state memories that were saved before compaction.
-3. Continue working based on the recovered context.
+1. Call `search_memories` to reload context, layering up to three angles:
+   - `metadata.type=session_state` -- the rich pre-compaction summary you wrote
+   - `metadata.type=compact_summary` -- the platform-generated condensed summary just now
+   - `metadata.type=decision` / `anti_pattern` -- specific facts you stored during the session
+2. Continue working from the recovered context.
 EOF
 fi