Addressed bandit false positive

Author: travis-bauer

CHANGELOG.md

@@ -1,6 +1,8 @@
 # Changelog
 
 ## 0.12.0
+- Fixed Bandit security scan false positive on the LLM memory default
+  `context_token_trigger`.
 - Fixed `makevectordatabase` document indexing so stored LanceDB `content` contains the
   shingled text that was embedded, rather than only the final split chunk.
 - Added LLM memory controls to prompt adapters, `llmPrompt`, and guided-generation

src/talkpipe/llm/chat.py

@@ -24,7 +24,8 @@
 PROMPT_ADAPTER_COMPAT_KWARG_DEFAULTS = {
     "memory_mode": "full",
     "unsummarized_message_count": 6,
-    "context_token_trigger": None,
+    # Controls memory token-budget compaction; not a credential.
+    "context_token_trigger": None,  # nosec B105
     "memory_size": 512,
     "debug_messages": False,
 }