| name | direct-voice |
| description | Choose the narration voice, pace, energy, and emotion for a content run from its vibe and script, by calling Hermes's voice-director judgment agent. Use when the Content Factory agent must produce the voice_assignment that Hermes's content-run pipeline needs for narration synthesis. |
| metadata | {"agent":"content-factory","source":"Project Atlas Requirements §3.5/10.2","layer":"L5 Content Factory","host":"railway"} |
direct-voice
Pick the ElevenLabs voice_id, pace, energy, and emotion matching the run's vibe and archetype.
When to use
- After
select-vibe has produced a vibe, before the content-run pipeline's narration-synthesis stage.
Inputs / Sources
Vibe output + script text + archetype context.
Output
{voice_id, pace, energy, emotion, rationale} (see hermes/agents/voice-director/schema.json) —
this is exactly the shape ContentRunRequest.voice_assignment expects (voice_id, energy
are read directly by orchestrator/temporal_workflows.ContentRunWorkflow).
Function signature (manifest contract)
def direct_voice(vibe: dict, script_text: str, archetype_context: dict) -> dict:
"""Returns {"voice_id": str, "pace": str, "energy": str, "emotion": str, "rationale": str}."""
Implementation
Call the hermes-bridge MCP tool run_judgment_agent(insight_type="voice-director", query=<ask>, context={"vibe": vibe, "script_text": script_text, **archetype_context}). Hits POST /v1/agents/voice-director — same self-correcting loop / mock-fallback pattern as select-vibe, no new backend.
Model
deepseek-direct/deepseek-chat — Cheap high-volume tier (spec §11.3).
