storyboard-from-script
The primary storyboarding skill. Takes a finished RMBC II script and produces both the human-readable storyboard.md AND the machine-readable scene-spec.json (schema v1.5+, valid against schemas/video-gen-spec.schema.json).
Pre-flight
script_pathexists and the script's status isreviewedorapproved.brief_pathexists and Brief 2.0 has all 25 sections filled.- Read
Brands/{brand_slug}/{BrandContext, voice-profile, compliance, knowledge-base}.md. - Read References/video-storyboarding/PLAYBOOK.md for the framework.
- Read References/video-storyboarding/schemas/video-gen-spec.schema.json — output MUST validate.
Procedure
1. Pick the template archetype
- If
template_archetypewas passed: validate it exists atReferences/video-storyboarding/templates/{archetype}/. - Else: scan all 22 templates'
video-theme.jsondescriptionfields. Pick the archetype whose narrative + visual + audio profile best matches:- The script's shape (UGC opener vs cinematic narrative vs claymation)
- The brand voice (warm-authentic vs premium-cinematic vs irreverent)
- The funnel position (TOFU = high-energy hook; BOFU = single-take demo)
- Surface 2-3 candidates to user if ambiguous; let user pick.
2. Pick the hook
- Read hooks/HOOK_CATALOG.json.
- Match the audience awareness level (from Brief 2.0 section 2 / research awareness assessment).
- Cross-check against hooks/ANTI_PATTERNS.json.
- Generate 3-5 hook variants for the
hook_variantsarray (per schema v1.8: each carries per-hook headlines + descriptions).
3. Decompose the script into ordered scenes
For each scene, populate:
scene_id(e.g.,SC01,SC02)start_sec+duration_sec(must sum to ≤ duration_target_sec)visual_prompt— descriptive, with realism cues per PLAYBOOK § 6 (UGC archetypes get the iPhone / no-cinematic-lighting / no-beauty-filter clauses)audio— VO line (verbatim from script) + music cue + SFXcaption— kinetic caption text + style preset (percaption_spec)camera_behavior— static / handheld / gimbal / dolly / rack-focusscene_archetype— proof-shot / pattern-interrupt / mechanism-reveal / offer / etc.
4. Author storyboard.md
Path: Brands/{brand_slug}/Marketing/storyboards/{date}-{slug}/storyboard.md
Format per the template in storyboarder.md. Sections: Frontmatter → Hook → Scene 1..N → Hook variants → Editing notes → Grounding.
5. Author scene-spec.json
Same directory: scene-spec.json. Schema-valid .video-gen-spec.json v1.5.
Required top-level: schema_version: "1.5", pipeline: "demandgen-video", brand, market, output_dir, generated_at, reference_images: [], scenes: [...], global_settings, script, audio_spec, caption_spec.
If schema validation fails, surface the error and fix before writing — never ship a broken spec.
6. Editing notes section in storyboard.md
Always include:
- Voice fit — does VO direction match brand voice-profile?
- Compliance flags — any visual or audio that may trip platform / niche compliance?
- Specificity gaps — any scenes still vague that need a real visual reference?
- Alternate hooks / CTAs — list the 3-5 hook variants for A/B testing
7. Append Log entry
Hard rules
- No storyboard without an approved script. Period.
- No fabricated visuals. If the brief lacks product details, scene visuals stay schematic; flag in editing notes.
- Realism cues mandatory for UGC archetypes — every scene visual_prompt carries them.
- Schema strictness — output MUST validate against video-gen-spec.schema.json.
- Compliance scan — every claim spoken or shown gets cross-checked against
Brands/{slug}/compliance.md.
Failure modes
- Script status is "draft" — abort, ask Adam to mark as reviewed/approved first.
- Brief incomplete — surface which sections are missing.
- Brand voice-profile / compliance still placeholder — surface as blocker (or proceed with explicit "voice TBD" flags).
- No template archetype fits — surface to Adam; consider running extract-style-from-reference to author a new archetype.
- Schema validation error — fix and re-write; don't write a broken spec.