| name | dircreative-script-treatment |
| description | Turn approved treatment into script-ready scenes, action, dialogue, voiceover, and no-dialogue decisions. |
Script Treatment
Required Knowledge
docs/film-preproduction/chat-co-creation-interface.md
docs/film-preproduction/chat-inline-visualization-interface.md
docs/film-preproduction/creative-copy-deck-capability.md
docs/film-preproduction/professional-agent-voice-standard.md
docs/film-preproduction/production-demo-retrospective.md
docs/film-preproduction/council-adversarial-review.md
docs/film-preproduction/client-film-hard-gates.md
docs/film-preproduction/schemas/story-package.yaml
docs/film-preproduction/research/audio-design-notes.md
docs/film-preproduction/research/film-production-glossary.md
Inputs
- approved treatment
- story review
- audio intent
- approved claim register
- real copy samples and
VOICE PROFILE when a client or brand voice is requested
Outputs
- script
- dialogue or no-dialogue policy
- voiceover or no-voiceover policy
- copy-development receipt
deck_narrative_spec when a deck handoff is requested; no PPT output
Chat Surface
Show the script in a readable preview:
阶段: 脚本预览
智能体创作内容: timed script, scene objective, conflict pressure, action beats, turn/reveal, dialogue/no-dialogue policy, VO budget when used, audio intent.
用户确认点: ask whether the script direction passes before shot design.
Keep the question about script approval only.
Copy Development Gate
Run audience-facing copy in this order:
- Collect 5 to 20 representative real samples when available; require
sample_id, source_ref, excerpt, why_representative, and recency, and record any shortfall.
- Build exactly one source-derived
VOICE PROFILE with the fields defined in project.yaml; cover all unique sample IDs or record coverage_shortfall_reason.
- Add DIR professional judgment: audience effect, channel fit, factual boundaries, tradeoff, and approved claim IDs.
- Run
de-AI-writing as a fidelity-preserving refinement. Protect meaning, evidence, and claim boundaries; add no unsupported fact or conclusion.
- Run
humanizer or humanizer-zh as a diagnostic only.
- Record manual craft review of rhythm, abstraction, and evidence alignment, then reconcile its status and trace clusters with
copy_execution.
An AI-trace cluster triggers manual review. A word-list or phrase hit cannot fail copy by itself and cannot prove authorship. Preserve intentional voice, quotations, technical terms, and approved claims.
The detected cluster set, diagnostic cluster set, manual-craft reviewed cluster set, and copy_execution.ai_trace_clusters must match. A cluster remains review-only, and the diagnostic may not add hidden blocking or authorship authority. Copy with no phrase hit can still require revision when manual craft review records repetitive rhythm, abstraction, or evidence detachment.
Every claim_id used by audience-facing dialogue, voiceover, supers, captions, CTA, or deck copy must have approval_status: approved and verified evidence. copy_execution.evidence_ids_used must cover used-claim evidence through the story source bindings. Pending, rejected, unreviewed, missing, or dangling claims block the relevant line. Audience-facing text must be non-empty, and copy statuses must match the validated pipeline.
Deck Narrative Handoff
When the user requests a deck, create deck_narrative_spec only after the script direction is approved. The spec must define the communication job, opening tension, cumulative story beats, closing resolution, one narrative job and authoritative primary_claim_id per slide, claim/evidence bindings, layout family, composition, primary visual role, text-density budget, and recomputed QA statuses.
Routing rules:
- Editable PowerPoint defaults to
Presentations with element-level editability.
codex-ppt is allowed only after the user explicitly accepts full-slide images whose contents are not separately editable.
high-end-visual-design may inform aesthetic references, but it is not a PPT construction, editability, or QA authority.
- Stop at the spec. Do not generate slides, slide images, or a
.pptx in this skill.
Visual Decision Contract
Use skills/dircreative/assets/visualizations/stage-surface-registry.json#script-timing-bands. Bind time bands, dialogue, voiceover, audio, duration budget, and overrun to the current script artifact; approval and revision remain conversation intents with a table fallback.
Rules
- Separate action, dialogue, voiceover, and sound notes.
- Preserve the approved facts, inferences, audience tension, business objective, proposition, avoid-cliches, concept dependencies, and claim boundaries.
- Write to the approved story duration, not automatically to 15s. If 15s is only the model generation-unit cap, write the full script/treatment first and hand longform splitting to sequence-planner.
- For 60-second client films, create a time budget and VO budget before shot design. Each VO line must include readable VO text, start/end seconds, covered shot IDs, and covered seconds. A VO line may cover multiple shots, but the mapping must be copyable and unambiguous.
- Do not write placeholders such as
VO marks this shot segment, VO标注本镜头句段, or internal production instructions inside VO text.
- Each scene must identify what the character wants, what blocks it, what changes by the end of the scene, and which visible action proves that change.
- Each scene must fill a scene beat card: scene objective, opposing force, tactic change, subtext under dialogue, visible action beat, turn/reveal, and end-state change.
- Dialogue must create pressure, misdirection, reveal, refusal, concession, or choice. Do not use dialogue only to state the theme or decorate the mood.
- The script must preserve the approved story engine: external pressure, hidden relationship engine, irreversible choice/reveal, escalation, consequence, and visible ending action.
- If the script reads as atmosphere without causality, return to story-development instead of continuing to shot design.
- If there is no dialogue, state that explicitly.
- Do not hide an unapproved claim inside dialogue, voiceover, a slogan, a supers line, or a slide title.
- Do not write shot list or prompt files.
skill_run_receipt
Record script scope, dialogue policy, copy source set and shortfalls, VOICE PROFILE coverage status, DIR judgment status, de-AI fidelity status, manual-craft status, humanizer review-only status, reconciled trace clusters, approved claim and evidence IDs used, deck claim/evidence/QA handoff status, and next_recommended_skill: script-breakdown.