Skill
Spam escalation (LLM)
spam-llmllmper messagebuilt inAsk the model for a second opinion on the borderline mail the rules can't call.
How it works
Escalate to the tenant-local model ONLY when the algorithmic spam score is uncertain ([0.35, 0.7)). Raises 'spam' on a confident spam verdict, records the model's reasoning under extra.spam_llm either way; never clears a flag the rules already raised.
What it surfaces for search
The fields, flags, and categories this skill writes become filters in the inbox search bar.
Fields
has:spam_scorehas:spam_reasonshas:spam_llmFlags
is:spam