| name | plan-assets |
| description | Determine B-roll, music, images, and templates; generate Google Veo and Higgsfield AI prompts for visual gaps; construct HeyGen presenter inputs, by calling Hermes's asset-planner judgment agent. Use when the Content Factory agent must produce the asset_planner_output that Hermes's content-run pipeline needs to drive visual/presenter/music/asset-sourcing generation. |
| metadata | {"agent":"content-factory","source":"Project Atlas Requirements §3.5/10.2","layer":"L5 Content Factory","host":"railway"} |
plan-assets
Plan every visual/audio asset the content-run pipeline will need: B-roll, music, images,
Veo/Higgsfield generation prompts, and the HeyGen presenter segment.
When to use
- After
select-vibe/direct-voice, before dispatching to Hermes's content-run pipeline.
Inputs / Sources
Script + Vibe + Voice assignment.
Output
{broll, music, images, veo_prompts, higgsfield_prompts, heygen_presenter} (see
hermes/agents/asset-planner/schema.json) — exactly the shape ContentRunRequest.asset_planner_output
expects (temporal_workflows.ContentRunWorkflow reads veo_prompts, higgsfield_prompts,
heygen_presenter, lyria_prompts, broll, music, images directly from it).
Function signature (manifest contract)
def plan_assets(script_text: str, vibe: dict, voice_assignment: dict) -> dict:
"""Returns the asset_planner_output dict Hermes's content-run pipeline consumes."""
Implementation
Call the hermes-bridge MCP tool run_judgment_agent(insight_type="asset-planner", query=<ask>, context={"script_text": script_text, "vibe": vibe, "voice_assignment": voice_assignment}). Hits POST /v1/agents/asset-planner — same self-correcting loop / mock-fallback pattern, no new backend. The Envato/Veo/Higgsfield/HeyGen calls themselves still happen deterministically inside Hermes's own production agents (asset-sourcing, visual-generation, presenter-generation) once the content run starts — this skill only produces the plan, not the generated assets.
Model
deepseek-direct/deepseek-chat — Cheap high-volume tier (spec §11.3); emit schema-only JSON, restrict to non-sensitive public data (§11.2).
