Wire sensory (sound/trace/awareness) narration through the pipeline

[dkirby-ms]

Summary

Full system prompts and timeout/model config exist for sound_narration/trace_narration/awareness_narration (llm-client.ts:127-160, narrative-types.ts:136-150), and there is a renderSensoryTemplate dispatch, but these types are never passed to narrate() — trace output goes straight through traceSystem (ZoneRoom.ts:2363-2371). Wire sensory output through the pipeline.

Scope / Tasks

• Route sendTraceNarrations (ZoneRoom.ts:2363-2371) and any sound/awareness emission through generateNarration('trace_narration' | 'sound_narration' | 'awareness_narration', …), with the current traceSystem text as fallback.
• Ensure the sensory-specific system prompts (llm-client.ts:127-160) are selected per type.
• Respect awareness/perception thresholds when deciding whether to narrate.
• Add wiring tests for each of the three sensory types.

Acceptance Criteria

• Trace/sound/awareness narration reaches narrate() and uses the matching sensory prompt.
• Template fallback preserves current behavior when LLM is off.
• Tests cover all three sensory types.

Dependencies

#513; Milestone A.

Size

M

Derived from the LLM narration gap analysis (Volo) and decomposition (Elminster).