fix(claude-plugin): rerank auto-injected memory context by default (#5690)

integrations/mem0-plugin/scripts/_search.py

@@ -7,12 +7,31 @@ All pre-fetch hooks use this instead of duplicating urllib boilerplate.
 from __future__ import annotations
 
 import json
+import os
 import urllib.request
 
 SEARCH_URL = "https://api.mem0.ai/v3/memories/search/"
 SEARCH_TIMEOUT = 5
 
 
+def should_rerank() -> bool:
+    """Whether auto-injection searches should request Platform reranking.
+
+    The REST search endpoint does not rerank when ``rerank`` is omitted, so
+    auto-injected context is ordered by raw vector similarity and the single
+    most relevant memory can fall outside the injected top_k window. We default
+    reranking ON for the hook-driven injection path (the extra ~150-200ms is
+    well within the hook's curl budget) and let users opt out via MEM0_RERANK.
+
+    MEM0_RERANK is read case-insensitively; ``0``, ``false``, ``no``, and
+    ``off`` disable reranking. Anything else (including unset) enables it.
+    """
+    raw = os.environ.get("MEM0_RERANK")
+    if raw is None:
+        return True
+    return raw.strip().lower() not in ("0", "false", "no", "off", "")
+
+
 def _do_search(api_key: str, payload: dict) -> list[dict]:
     body = json.dumps(payload).encode()
     req = urllib.request.Request(

integrations/mem0-plugin/scripts/file_context.py

@@ -23,7 +23,7 @@ sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
 from _formatting import TYPE_ICONS, format_age
 from _identity import resolve_api_key, resolve_user_id
 from _project import resolve_project_id
-from _search import search_memories
+from _search import search_memories, should_rerank
 
 FILE_READ_GATE_MIN_BYTES = 1500
 MAX_RESULTS = 5
@@ -93,6 +93,7 @@ def search_file_context(
         api_key, user_id, project_id, query,
         top_k=MAX_RESULTS, threshold=0.3,
         global_search=global_search,
+        rerank=should_rerank(),
     )
 
     results = results[:MAX_RESULTS]

integrations/mem0-plugin/scripts/on_bash_output.sh

@@ -77,15 +77,16 @@ RESULTS=$(PYTHONPATH="$SCRIPT_DIR" MEM0_SEARCH_QUERY="$ERROR_QUERY" MEM0_SEARCH_
   python3 -c "
 import os, sys
 sys.path.insert(0, os.environ.get('PYTHONPATH', '.'))
-from _search import search_memories, format_results_for_context
+from _search import search_memories, format_results_for_context, should_rerank
 
 api_key = os.environ.get('MEM0_API_KEY', '')
 user_id = os.environ.get('MEM0_SEARCH_USER', 'default')
 project_id = os.environ.get('MEM0_PROJECT_ID', 'unknown')
 query = os.environ.get('MEM0_SEARCH_QUERY', '')
+rerank = should_rerank()
 
-r1 = search_memories(api_key, user_id, project_id, query, metadata_type='anti_pattern', top_k=3)
-r2 = search_memories(api_key, user_id, project_id, query, metadata_type='bug_fix', top_k=3)
+r1 = search_memories(api_key, user_id, project_id, query, metadata_type='anti_pattern', top_k=3, rerank=rerank)
+r2 = search_memories(api_key, user_id, project_id, query, metadata_type='bug_fix', top_k=3, rerank=rerank)
 
 seen = set()
 combined = []

integrations/mem0-plugin/scripts/on_user_prompt.sh

@@ -120,14 +120,15 @@ if [ -n "$HAS_RESUME" ]; then
   RESUME_RESULTS=$(PYTHONPATH="$SCRIPT_DIR" MEM0_SEARCH_USER="$USER_ID" python3 -c "
 import os, sys
 sys.path.insert(0, os.environ.get('PYTHONPATH', '.'))
-from _search import search_memories, format_results_for_context
+from _search import search_memories, format_results_for_context, should_rerank
 
 api_key = os.environ.get('MEM0_API_KEY', '')
 user_id = os.environ.get('MEM0_SEARCH_USER', 'default')
 project_id = os.environ.get('MEM0_PROJECT_ID', 'unknown')
+rerank = should_rerank()
 
-state = search_memories(api_key, user_id, project_id, 'session state current task', metadata_type='session_state', top_k=3)
-decisions = search_memories(api_key, user_id, project_id, 'recent decisions and learnings', metadata_type='decision', top_k=3)
+state = search_memories(api_key, user_id, project_id, 'session state current task', metadata_type='session_state', top_k=3, rerank=rerank)
+decisions = search_memories(api_key, user_id, project_id, 'recent decisions and learnings', metadata_type='decision', top_k=3, rerank=rerank)
 
 all_r = state + decisions
 seen = set()