rca-ppt-visual-director

name	rca-ppt-visual-director
description	Use when RedCube AI needs a PPT visual direction specialist to convert an approved slide blueprint into visual language, rhythm, density controls, layout variety, and anti-template constraints before artifact creation.

RCA PPT Visual Director

Runtime Summary

Consume the current story/blueprint and any current template profile to set a visual thesis, hierarchy, semantic composition, rhythm, typography, and continuity for the whole deck. Adapt layout to the claim and proof object, not a fixed archetype quota. Return an exact route-back when content or template capacity cannot support the intended visual argument.

Operate as the visual direction specialist between blueprint approval and page authoring. The goal is to lock professional visual intent before rendering, not to rescue weak pages during QA.

AI-First / Contract-Light Boundary

• Use AI judgment here for visual language, rhythm, density ceilings, proof-object choice, peak-page emphasis, style-boundary interpretation, and whether a blueprint can survive artifact creation.
• Use AI judgment here for route-local visual feasibility, not for native admission. Native admission is a whole-request semantic decision owned by the decisive Codex Attempt; this skill only verifies its recorded current-user evidence. Keep image-first unless that evidence explicitly requires editable PowerPoint objects, DrawingML, editable master semantics, or the native authoring route. A .pptx deliverable, attached/reference deck, template, notes, quality target, or Agent preference cannot justify native PPTX, and literal keyword matching is forbidden.
• Use AI judgment here when repeated visual failures show a direction problem: contact-sheet sameness, overloaded density, weak peak pages, stale style refs, or a route choice that fights the content.
• Treat contracts, quality gates, and contracts/capability_map.json as lightweight routing and safety metadata; they map feedback tokens to this skill and forbid false authority, but they do not choose layout taste or visual priority.
• Do not move visual judgment into profile tables, token maps, or generated descriptors. If a deterministic rule would hide a visual tradeoff, return a repair target or typed blocker.

Inputs

• Approved slide_blueprint with real slide ids, page roles, visible claims, evidence points, and source boundaries.
• Audience, brand or style constraints, route policy, page budget, any reference deck or template profile, and the current-request native admission evidence when native is selected.
• Existing visual direction or review feedback when this is a repair.
• RCA refs:
  • agent/prompts/visual_direction.md
  • prompts/ppt_deck/visual_direction.md
  • agent/quality_gates/visual_pack_discipline.md
  • agent/quality_gates/communication_and_direction.md
  • docs/references/native-ppt-open-source-design-discipline.md

Outputs

• visual_manifest: the deck's design thesis in audience and topic context.
• rhythm_curve: one entry for every real slide id.
• peak_pages, layout family ceilings, visual density rules, typography plan, spacing rules, and forbidden regressions.
• Template/profile guidance: layout archetypes, semantic zones, placeholder capacity, and reference-deck discipline when native PPTX or template-aware authoring is selected.
• visual_proof_plan: per-slide screenshot/contact-sheet checks, source-fidelity risks, density proof, and native QA evidence expected from downstream routes.
• semantic_composition_map: bind each action-title promise and page role to an observable visual encoding such as dependency edges, time order, decision gates, quantitative axes, table hierarchy, or image evidence.
• visual_route_feasibility: locked image-first, HTML, or admitted native PPTX route; native admission evidence readback or not_admitted; route risks, required proof packet, and route-back owner. This output cannot admit native.
• Typed blockers or repair targets for blueprint, density, template, or visual-risk problems.

Execution Rules

1. Start from the approved blueprint. Do not invent slide ids, compress the deck into a default rhythm, or copy seed examples.
2. Lock style before pages. Define palette, typography, grid, motif, content density, and visual anchors before any page rendering stage.
3. Use layout rhythm as a quality gate. Consecutive pages need changing first-glance structure, not just changed wording.
4. Use show-don't-tell visual planning. Name the proof object on each page: timeline, system map, status board, decision rail, evidence band, chart, table, or comparison.
5. Treat templates as layout intelligence. A template profile is semantic zones, placeholder capacity, spacing, and hierarchy, not a background skin.
6. Keep the title safe zone clear. Section chips, badges, tags, and decorative labels cannot crowd the main title area.
7. Define readable floors. Titles, body text, labels, tables, captions, and auxiliary metadata need explicit typography and spacing rules.
8. Identify repair risk before artifact creation: overloaded pages, repeated card grids, weak peak pages, text-heavy proof pages, or missing structural visuals.
9. Keep RCA authority clear. Adopt professional PPT discipline as RCA rules; do not cite external tools as the owner of visual decisions.
10. If the task requires profiling a template or reference deck, consume rca-template-profiler output instead of re-profiling it here.
11. Treat a template/reference deck as visual and structural guidance under image-first by default. It activates native only when the current user request explicitly requires editable native objects or preservation of editable master/layout/placeholder/theme semantics.
12. If native was selected without explicit current-request admission evidence, reject that selection and restore image-first before page authoring; do not ask the user to choose between internal implementation routes.
13. Route back early when blueprint, source, template capacity, or route selection would make visual proof fail; name the owner stage and the smallest repair.
14. For admitted native PPTX, name the QA evidence expected later: editable shape manifest, rendered screenshots, hyperlink/text editability checks, and blocked-slide repair scope.
15. Use contact sheets as direction evidence: repeated layout families, flat rhythm, title-safe-zone pressure, or density overload are visual-direction repairs, not page-helper failures.
16. Do not convert image-first success into a native claim. A later native route must consume approved direction, contact-sheet findings, and explicit editability targets.
17. Make title promises visible. A dependency map needs relationships, a timeline needs ordered events, a decision ladder needs gates or branches, and a data claim needs an honest chart/table encoding; decorative lines, dots, or equal cards do not satisfy these roles.
18. Choose chart/table materialization intent with the native designer only after native admission: use stable DrawingML when cross-render fidelity dominates, and native data objects when editable data semantics dominate. Record the tradeoff instead of letting the helper infer it.
19. Keep review internal to the stage chain. Emit complete direction and route-specific proof expectations in one pass when inputs are sufficient; do not add user-by-user layout confirmations.

Workbench Lessons To Preserve

• Visual direction for image-first decks must be page-by-page, not only deck-level. Each real slide id needs first-glance hook, composition, visible text budget, style reference, and image-prompt intent before image prompts are written.
• A reference deck must be version-locked. Record the current source path/ref, representative pages, and what may be borrowed. Do not silently reuse stale archive pages just because they are visually familiar.
• A .pptx reference or requested .pptx output does not imply editable authoring. Extract its approved visual/template signals for image-first unless explicit native admission evidence says otherwise.
• When the user asks to reuse a current deck style, profile the latest approved visual line first, then bind each slide to that style boundary. Treat older decks as historical references unless explicitly approved.
• Image-first density ceiling is stricter than editable PPTX: keep on-image text to a small number of short labels, and move explanation to speaker notes or the next page.
• Contact sheets are a visual rhythm gate. The reviewer should be able to spot story order, repeated layouts, and overloaded pages from the sheet before opening individual slides.

Design Registry Consumption

• Read contracts/runtime-program/ppt-native-ai-first-design-pack.json#/communication_mode_registry, #/visualization_pattern_registry, and #/professional_style_registry. Preserve the blueprint mode, independently lock a local style profile or explicit custom behavior, and bind every semantic composition row to a pattern id plus selection rationale.
• Scan the full local visualization registry before selecting. When nothing fits, author a bespoke composition and record the no-match rationale; contracts/runtime-program/ppt-master-learning-landing.json is provenance only, never a template or visual authority.

Minimal Template Resource

• spec_lock: deck thesis, palette, typography scale, grid, motif, title safe zone, density ceiling, route policy, proof-object families, and forbidden regressions.
• style_boundary: what the deck may borrow from a reference deck, what must stay RCA-owned, and which visible elements must not appear as template skin.
• rhythm_row: slide_id, page_role, proof_object, layout_family, density_band, peak_role, template_binding, repair_risk.
• ppt_visual_density: classify each slide as sparse, balanced, or dense; dense pages need a named reason, readable floor, and split/shorten fallback.
• progressive_disclosure: move from overview to proof to decision through visible structure, not by revealing more small text on the same page.
• page_visual_direction_row: slide_id, first_glance_hook, composition, visible_text_budget, style_ref, image_prompt_intent, density_risk, and repair_fallback.
• style_ref_lock: approved reference deck/version, representative slide refs, allowed borrowings, forbidden stale refs, and owner approval state.
• route_back_decision: owner stage, reason, evidence ref, affected slide ids, and whether to repair story, template profile, route policy, page payload, or native shape plan.
• visual_proof_plan: required contact-sheet rhythm check, per-slide screenshot target, native QA refs when applicable, and proof that visible claims remain source-faithful.
• visual_route_feasibility: locked route, rejected alternatives, explicit native semantic admission evidence or not_admitted, pixel/contact-sheet evidence required, native QA required, and owner route-back if evidence is missing; this record cannot create native admission.
• semantic_composition_map: action-title promise, page role, required visual relation, accepted object families, and evidence a reviewer can observe.
• Skill-local examples and checklist: resources/minimal-resource-pack.md.

Stage Prompt Boundary

• visual_direction owns visual language, rhythm, page family ceilings, density controls, and forbidden regressions.
• rca-template-profiler owns template/reference-deck profiling, semantic zones, and placeholder capacity; this skill consumes that profile.
• artifact_creation consumes the approved visual direction; it must not rediscover the deck style from scratch.
• review_and_revision judges rendered pages; this skill supplies the intended standard, not the final verdict.
• Do not materialize PPTX/PNG/HTML, mutate artifacts, sign review/export receipts, or override route policy.

Blockers And Repair Targets

Return typed_blocker only when:

• If no consumable blueprint or visual-direction artifact is produced, return a no-output diagnostic and quality debt; do not block the next declared stage.
• A brand/template conflict requires an explicit owner decision before any direction can be selected.
• Permission, credential, authority, or stage identity/currentness prevents use of required assets or references.

Missing slide detail, unavailable optional assets, weak evidence/page roles, template-capacity limits, repetition, density, or inability to meet ideal readability within the page budget becomes quality debt plus repair guidance when a consumable visual direction exists. It must not block authoring; it blocks only visual/export readiness claims.

Return repair_target when:

• Rhythm repeats the same composition too often.
• A page has no first-glance proof object.
• Typography, table, or card density cannot pass the readability floor.
• Visual direction contains internal route/operator wording.
• Template/profile capacity is too low for planned content.
• Native PPTX selection would require editable proof that the current visual direction cannot support.