diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..0fbf0b2
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,18 @@
+### macOS
+# Finder metadata
+.DS_Store
+
+# Thumbnails
+._*
+
+# Custom folder icons
+Icon

+
+# Volume root files
+.DocumentRevisions-V100
+.fseventsd
+.Spotlight-V100
+.TemporaryItems
+.Trashes
+.VolumeIcon.icns
+.com.apple.timemachine.donotpresent
\ No newline at end of file
diff --git a/hooks/sync-memories.sh b/hooks/sync-memories.sh
index 0cc1c85..da5d29f 100755
--- a/hooks/sync-memories.sh
+++ b/hooks/sync-memories.sh
@@ -1,59 +1,108 @@
 #!/bin/bash
 
 # Hook: index .claude/memories/ into docs-mcp-server for semantic search.
-# Runs on SessionStart and UserPromptSubmit (async). Never fails the hook.
+# Runs on SessionStart and UserPromptSubmit (async). Never fails the hook and
+# never writes to stdout (UserPromptSubmit stdout would be injected into context).
 
 set -uo pipefail
 
-# Consume stdin (hook passes JSON context)
-cat >/dev/null 2>&1 || true
+# Read the hook's stdin JSON so we can fall back to its `cwd` field below.
+hook_input=$(cat 2>/dev/null || true)
+hook_event=$(printf '%s' "$hook_input" | jq -r '.hook_event_name // empty' 2>/dev/null)
 
-project_root=$(git rev-parse --show-toplevel 2>/dev/null) || exit 0
+# --- Resolve the project root: the directory that contains .claude/ ---
+# In a git repo, anchor to the repo toplevel (original behavior): the same repo
+# may be checked out under different folder names, and the repo root is the stable
+# anchor so they map to one library. For a non-git folder (git rev-parse exits
+# non-zero), fall back to where Claude was launched — CLAUDE_PROJECT_DIR, then the
+# hook's stdin `cwd`, then $PWD.
+project_root=$(git rev-parse --show-toplevel 2>/dev/null)
+if [ -z "$project_root" ]; then
+    cwd=$(printf '%s' "$hook_input" | jq -r '.cwd // empty' 2>/dev/null)
+    project_root="${CLAUDE_PROJECT_DIR:-${cwd:-$PWD}}"
+fi
 
 MEMORIES_DIR="$project_root/.claude/memories"
 TIMESTAMP_FILE="$project_root/.claude/.memories-last-indexed"
+LOG_FILE="$project_root/.claude/.memories-index.log"
 
-# Exit early if no memories directory
+# Nothing to do without a memories directory that actually holds .md files.
 [ -d "$MEMORIES_DIR" ] || exit 0
+[ -n "$(find "$MEMORIES_DIR" -name '*.md' -print -quit 2>/dev/null)" ] || exit 0
 
-# Exit early if Ollama is not running
+# Exit early if Ollama is not running.
 curl -s --max-time 2 http://localhost:11434/api/tags >/dev/null 2>&1 || exit 0
 
-# --- Staleness check ---
-# If timestamp file exists and nothing changed, nothing to do.
-# If timestamp file doesn't exist, this is the first run — do a full index.
-if [ -f "$TIMESTAMP_FILE" ]; then
-    newest=$(find "$MEMORIES_DIR" -name "*.md" -newer "$TIMESTAMP_FILE" -print -quit 2>/dev/null)
-    # Also check if the directory itself was modified (file added/removed)
-    dir_changed=""
-    [ "$MEMORIES_DIR" -nt "$TIMESTAMP_FILE" ] && dir_changed="yes"
-    [ -n "$newest" ] || [ -n "$dir_changed" ] || exit 0
+# Start each session with a fresh log so it can't grow unbounded across sessions;
+# UserPromptSubmit runs during the session then append to it. Logging is
+# best-effort — a failed write here must never stop indexing.
+if [ "$hook_event" = "SessionStart" ]; then
+    : 2>/dev/null >"$LOG_FILE" || true
 fi
 
-# --- Index ---
+log() {
+    printf '%s %s\n' "$(date '+%Y-%m-%dT%H:%M:%S')" "$1" 2>/dev/null >>"$LOG_FILE" || true
+}
+
 repo_name=$(basename "$project_root")
+memories_url="file://$MEMORIES_DIR"
 
+embedding_model="openai:nomic-embed-text"
 export OPENAI_API_KEY=ollama
 export OPENAI_API_BASE=http://localhost:11434/v1
+# `list` reads the model from this env var and rejects --embedding-model. Without
+# it, `list` initializes with the default model, mismatches the nomic-built store,
+# and throws EmbeddingModelChangedError — silently breaking the count check below.
+export DOCS_MCP_EMBEDDING_MODEL="$embedding_model"
 export DOCS_MCP_SCRAPER_SECURITY_FILE_ACCESS_MODE=unrestricted
 export DOCS_MCP_SCRAPER_SECURITY_FILE_ACCESS_INCLUDE_HIDDEN=true
 export DOCS_MCP_SCRAPER_SECURITY_FILE_ACCESS_FOLLOW_SYMLINKS=true
-embedding_model="openai:nomic-embed-text"
 
-# Check if library already indexed with the same source URL
-existing_url=$(npx -y @arabold/docs-mcp-server@latest list 2>/dev/null \
-    | jq -r --arg name "$repo_name" '.[] | select(.name == $name) | .versions[0].sourceUrl // empty')
+# Echo the indexed document count for this library+source (0 if absent or empty).
+# Spawns npx, so callers keep it off the per-prompt hot path. docs-mcp-server
+# lowercases library names on storage, so match the name case-insensitively.
+library_doc_count() {
+    npx -y @arabold/docs-mcp-server@latest list --output json 2>/dev/null \
+        | jq -r --arg name "$repo_name" --arg url "$memories_url" \
+            'map(select((.name | ascii_downcase) == ($name | ascii_downcase)))[0].versions[]?
+             | select(.sourceUrl == $url) | .documentCount // 0' 2>/dev/null \
+        | head -1
+}
+
+# --- Decide whether to (re)index ---
+# The cheap local checks (timestamp vs. file mtimes) run every invocation. The
+# store-existence probe spawns npx, so it runs only at SessionStart — enough to
+# re-add a library whose store was cleared while its timestamp lingered, without
+# paying npx on every prompt. Anything changed, or no timestamp yet, => index.
+should_index=1
+if [ -f "$TIMESTAMP_FILE" ]; then
+    newest=$(find "$MEMORIES_DIR" -name '*.md' -newer "$TIMESTAMP_FILE" -print -quit 2>/dev/null)
+    if [ -z "$newest" ] && [ ! "$MEMORIES_DIR" -nt "$TIMESTAMP_FILE" ]; then
+        should_index=0
+        if [ "$hook_event" = "SessionStart" ]; then
+            count=$(library_doc_count)
+            [ "${count:-0}" -gt 0 ] 2>/dev/null || should_index=1
+        fi
+    fi
+fi
+[ "$should_index" -eq 1 ] || exit 0
+
+# --- Index ---
+# `scrape` (re)indexes in place whether or not the library exists — adds when
+# missing, updates when present — so no separate refresh command is needed.
+output=$(npx -y @arabold/docs-mcp-server@latest scrape "$repo_name" "$memories_url" \
+    --embedding-model "$embedding_model" 2>&1)
+rc=$?
 
-if [ "$existing_url" = "file://$MEMORIES_DIR" ]; then
-    npx -y @arabold/docs-mcp-server@latest refresh "$repo_name" \
-        --embedding-model "$embedding_model" \
-        --silent >/dev/null 2>&1
+# Confirm content actually landed using the same trusted count the gate uses: a
+# failure — or a #/? truncated path that "succeeds" with 0 pages — leaves the
+# count at 0, so we never record a false success that hides the library at zero.
+count=$(library_doc_count)
+if [ "$rc" -eq 0 ] && [ "${count:-0}" -gt 0 ] 2>/dev/null; then
+    touch "$TIMESTAMP_FILE"
+    log "index ok: '$repo_name' ($count docs) <- $memories_url"
 else
-    npx -y @arabold/docs-mcp-server@latest scrape "$repo_name" \
-        "file://$MEMORIES_DIR" \
-        --embedding-model "$embedding_model" \
-        --silent >/dev/null 2>&1
+    log "index FAILED (rc=$rc, docs=${count:-0}) '$repo_name' <- $memories_url :: $(printf '%s' "$output" | tr '\n' ' ' | tail -c 300)"
 fi
 
-# Mark indexing time for subsequent staleness checks
-touch "$TIMESTAMP_FILE"
+exit 0
diff --git a/techpack.yaml b/techpack.yaml
index 719a4d6..1c62462 100644
--- a/techpack.yaml
+++ b/techpack.yaml
@@ -160,6 +160,7 @@ components:
     description: Ignores memory files from version control
     gitignore:
       - ".claude/.memories-last-indexed"
+      - ".claude/.memories-index.log"
 
 # ---------------------------------------------------------------------------
 # Ignore — maintainer-only files not shipped with the pack