show-profile-manager

name	show-profile-manager
description	Create, load, validate, and version show profiles and creative config objects for the podcast pipeline.
user-invocable	true
model	opus
tools	Read, Write, Edit, Glob, Grep, Bash

/show-profile-manager

You manage the identity layer of the podcast pipeline. A profile is not a settings file — it is a creative worldview. When you create or edit a profile, you are defining how a show thinks, what it values, and what it refuses.

Commands

• /show-profile-manager create — Interactive profile creation (guided conversation)
• /show-profile-manager load [name] — Load and validate a profile
• /show-profile-manager edit [name] — Modify a profile field
• /show-profile-manager creative add [profile] — Add a creative config to a profile
• /show-profile-manager creative edit [profile] [creative] — Modify a creative config
• /show-profile-manager list — Show available profiles and their creatives
• /show-profile-manager validate [name] — Check profile schema completeness

Storage

Profiles are stored as JSON files in ./profiles/ relative to the working directory:

profiles/
  show-name.profile.json

Profile Schema

{
  "schema_version": "1.1.0",
  "profile_id": "uuid",
  "show_name": "string — the show's name",
  "show_slug": "string — filesystem-safe identifier",
  "mission": "string — what this show exists to do in the world",
  "audience_tribe": "string — who these people are, not just demographics but identity",
  
  "platform_stack": [
    {
      "platform": "instagram | tiktok | youtube_shorts | youtube | spotify | apple_podcasts",
      "priority": "primary | secondary | tertiary",
      "primary_ratio": "9:16 | 1:1 | 16:9 | 4:5",
      "content_approach": "string — how the show uses this platform specifically"
    }
  ],
  
  "aesthetic_invariants": {
    "visual": {
      "grain_vs_clean": "float 0-1 (0=clinical digital, 1=heavy film grain)",
      "typography_personality": "string — how text feels, not just which font",
      "color_anchors": ["string — 3-5 color descriptions, not hex codes"],
      "motion_style": "string — how things move (languid, snappy, weighted, floaty)",
      "texture_preference": "string — organic/synthetic, tactile/flat, lived-in/pristine"
    },
    "sonic": {
      "strategy": "string — what role sound design plays in the show's identity",
      "identity_description": "string — what the show sounds like between words",
      "vocal_processing": "string — how voice should feel (warm/present/distant/intimate)"
    },
    "anti_patterns": [
      "string — things this show NEVER does, visually or sonically"
    ]
  },
  
  "editorial_voice": {
    "host_sensibility": "string — how the host sees the world, what they find interesting",
    "what_the_show_rewards": "string — what makes a good moment on this show",
    "what_bores_the_show": "string — what the show has no patience for",
    "trusted_formats": ["string — formats/structures that work for this show"],
    "forbidden_moves": ["string — editorial choices this show would never make"]
  },
  
  "growth_scoring_weights": {
    "hook_density": "float 0-1 — weight for scroll-arrest moments",
    "quotability": "float 0-1 — weight for standalone phrases",
    "tension_resolution": "float 0-1 — weight for build-peak-release arcs",
    "loop_ability": "float 0-1 — weight for end-echoes-beginning structures",
    "emotional_peaks": "float 0-1 — weight for surprise/recognition/revelation",
    "tribal_signals": "float 0-1 — weight for audience-identity language"
  },
  
  "team": [
    {
      "name": "string — team member name",
      "role": "string — role in the show (host, writer, producer, editor)",
      "bio": "string — brief description relevant to creative decisions"
    }
  ],
  
  "show_stats": {
    "episodes": "int — total episodes released",
    "countries": "string — reach description (e.g., '25+ countries')",
    "platforms": "int — number of distribution platforms",
    "ai_knowledge_entries": "int — number of AI knowledge base entries, if applicable"
  },
  
  "release_schedule": [
    {
      "name": "string — content type name (e.g., 'Full Episode')",
      "day": "string — day of week",
      "platform": "string — primary platform for this release",
      "notes": "string — additional context"
    }
  ],
  
  "studio": "string — physical studio location description",
  
  "guest_deliverable": "string — description of what the pipeline produces FOR the guest",
  
  "website": "string — show website URL",
  
  "rss_feed": "string — podcast RSS feed URL",
  
  "creatives": []
}

Creative Schema

{
  "creative_id": "uuid",
  "name": "string — human-readable name (e.g., 'Instagram Reel - Standard')",
  "output_type": "reel | story | short | trailer | audiogram",
  "platform": "instagram | tiktok | youtube_shorts",
  "ratio": "9:16 | 1:1 | 16:9 | 4:5",
  
  "duration_target": {
    "min_seconds": 30,
    "max_seconds": 90,
    "sweet_spot": 60
  },
  
  "caption_style": {
    "key_semantic": {
      "position": "center | lower_third | upper_third | dynamic",
      "font_personality": "string — not a font name, a feeling",
      "size_behavior": "string — how size relates to emphasis",
      "animation": "string — how text arrives and departs",
      "diegetic_treatment": "string — how text integrates with b-roll (in-world vs overlay)"
    },
    "subtitle": {
      "position": "bottom | lower_third",
      "font_personality": "string — deliberately subordinate to key semantic",
      "size": "small | medium",
      "style": "string — how subtitles look and feel"
    }
  },
  
  "hook_style": "string — how this creative opens (hard cut, slow build, question, provocation, recognition)",
  "broll_direction": "string — overall b-roll philosophy for this creative",
  "copy_tone": "string — how the post copy sounds (conversational, authoritative, vulnerable, provocative)",
  
  "video_gen_tool": "seedance-2.0-i2v | seedance-2.0-r2v | kling-v3-pro-i2v | nano-banana-2 | none",
  // Must match a key in fal_router.py MODELS dict exactly. Abstract names (e.g. 'seedance', 'kling') will break envelope generation.
  "video_gen_notes": "string — tool-specific prompt optimization notes",
  
  "morph_config": {
    "enabled": "boolean — whether morph transitions are used for this creative (default true for morphing b-roll slots)",
    "generation_model": "seedance-2.0-i2v — deterministic start+end frame binding. Only model that guarantees morph endpoints. Do NOT use r2v for morphs (no frame binding).",
    "midpoint_model": "nano-banana-2 — generates the abstract midpoint image at show palette",
    "frame_binding": "deterministic — image_url (start) + end_image_url (end). No reference images alongside frame binding (API modes are mutually exclusive on Seedance).",
    "clip_duration_minimum": "4 — Seedance minimum. Generate at 5s, trim to fit overlay window half.",
    "departure_frame_offset": "-0.033 — seconds before b-roll IN timestamp (one frame at 30fps)",
    "return_frame_offset": "0.0 — at b-roll OUT timestamp exactly",
    "reframe_before_generation": "true — departure and return frames must be cropped/reframed to output aspect ratio before use as start/end frames"
  },

  "adjudication_config": {
    "enabled": "boolean — whether visual adjudication runs for this creative (default true)",
    "preset": "source_assessment | generated_assessment | draft_review — which adjudication preset to run",
    "model_tier": "small | medium | large — model tier for adjudication (default medium)",
    "auto_revise": "boolean — if true, pipeline automatically re-runs flagged stages on REVISE verdict (default false)",
    "morph_continuity_check": "boolean — if true, adjudicator specifically checks morph transition integrity (default true)",
    "caption_sync_check": "boolean — if true, adjudicator checks caption timing alignment (default true)"
  },
  
  "overrides": {
    "// any profile-level field can be overridden here for this creative": true
  }
}

Adjudication Model Tier Guidance

The model_tier field in adjudication_config maps to Overshoot model selection:

• medium (default) — uses qwen3.5-9b for source and generated assessment. Fast, strong vision benchmarks.
• large — uses qwen3.5-27b for draft review. Best quality for final editorial judgment. Recommended for the draft_review preset.
• small — uses qwen3.5-4b. For high-volume triage only; not recommended for production quality gates.

Profile Creation Flow

When creating a profile interactively, guide the user through these questions in this order. Do not present a form. Have a conversation.

1. What's the show? — Name, mission, what it exists to do
2. Who listens? — Not demographics. Identity. What do these people believe? What are they hungry for?
3. What does the show sound like when it's at its best? — The moments that make it this show and not any other
4. What would the show never do? — Anti-patterns, forbidden moves, editorial lines
5. Where does it live? — Platform stack, priorities, how each platform is used differently
6. What does it look like? — Visual identity: grain, color, motion, typography personality
7. How does it score? — Which growth dimensions matter most for this show? (Guide them through the six weights)

After gathering answers, synthesize into the full profile schema. Present it back for confirmation before writing to disk.

Validation Rules

A profile is valid when:

• All required fields have non-empty values
• growth_scoring_weights values sum to approximately 1.0 (within 0.1 tolerance)
• At least one platform in platform_stack
• At least one creative in creatives (warn but don't block if missing)
• anti_patterns has at least 2 entries (a show without limits has no identity)
• editorial_voice.forbidden_moves has at least 1 entry

Versioning

When a profile is modified:

• Increment the schema_version patch number
• Add a _changelog array entry with timestamp and description of what changed
• The previous version is not preserved (git handles history)