cinematic-screenplay-incubation git:20261020.9ab055cA

Immutable. This exact content never changes and is served at /api/v1/blob/658f2436e1b61c22.

---
name: cinematic-screenplay-incubation
description: Support Host-first full-film and scene screenplay creation, then source-bound dramaturgy, professional exploration and Host-selected revision; production remains downstream.
---

## Real author inputs and collaboration

Use the [existing author entry](../../docs/host-first-creation.md). Identify whether
Host or a configured independent author actually writes each body. Provide source,
project adaptation authority, creative preferences, selected professional methods
and feedback through `authorContext`; they participate in first creation rather
than arriving only after the author has committed to a restricted draft. Advisory
methods leave room for the author to choose scenes, words, actions and structure.
Preserve the complete raw author output before Host edits. Director treatment may
return screenplay changes; reconcile both complete bodies before adoption, then
project exact selected sources into scene/shot production. Record actual authorship
and selected differences, not just that a call succeeded.

Formal source/rights compilation still governs Canon writes and production.
A designated-source text candidate can be written before a professional creative
interpretation package is settled; it never silently replaces that authority.

## Creative Source dispatch

For formal source compilation after autonomous candidate writing, HISTORICAL retains the historical-first module and evidence rules below. LITERARY obtains a reviewed package through literary-source-analysis → philosophical-core → literary-adaptation → literature-to-cinema and `source.prepare_screenplay`. For LITERARY, replace only historical grounding with that package and exact screenplayInput; shared knowledge, continuity, scene craft and review checks remain active. Never apply Historical Scope/Spine as fictional truth requirements. PhilosophicalCore is an interpretive constraint, not a dialogue template. Follow [Creative Source](../../docs/creative-source.md); screenplay revisions cannot edit upstream decisions in place.



# Authorial closure handoff

At an already-developed closure point, `authorial-voice` may compare silence with
sparse literary intervention. It does not rewrite character dialogue or story.
Keep Character Voice and World Voice in the screenplay; an authorial candidate
remains a separate, unadopted sidecar until the user selects it.

# Cinematic Screenplay Incubation · 影视编剧孵化

During screenplay creation and director design, consume the unified `creativeRhythm` from the actual creation context for professional realization planning, even when source text is already supplied. Follow [narrative rhythm](../../docs/narrative-rhythm.md); Skills never read env directly. Pin this profile to the creative revision; refresh/resume keeps it, while a new revision reads current configuration.

Own the creative working process from Historical Material to a readable, reviewed screenplay. The six modules below are stages of one Agent's work, not separate agents or an automatic skill chain. Preserve an already approved foundation; choose only the stages affected by the request.

## Scope and compatibility

The existing hierarchy is Work → Script → Episode → Scene → Shot; a Work is not limited to one Episode. Work retains the literary foundation, Script the adaptation architecture, Episode its dramatic unit, Scene the canonical action and `spokenContent`. The existing work-creation and script-adaptation entrypoints remain valid for narrow requests. Their historical attribution, persistence and revision gates still apply; this Skill shares their results in one working context rather than redefining their contracts.

Use supplied context first. For existing identities, read `work.get_work`, `script.get_script`, `episode.get_episode`, `scene.get_scene` only as needed. Use `context.build_context` for missing parent context. For material evidence gaps use `research.search_sources` and `research.verify_claim`; unresolved contested history requires an adaptation position, not silent certainty. Do not require a live domain object to write a requested standalone screenplay artifact.

For an explicit persistent deliverable, apply the relevant existing owner stage's write rules after review. Formal domain reads remain read-only here; the screenplay tools retain local originals and decisions, without automatically creating Work, Script, Episode or Scene. A local creative E2E is actual authored work, not a mocked provider response, and does not require media generation.

## Meta dramaturgy after the complete original

For a new work derive a lightweight load profile, explicit narrative aperture, POV contract and style from the request and evidence, using [meta dramaturgy](references/meta-dramaturgy.md). These select emphasis within the existing twenty dimensions; they are not another mandatory exposition checklist. Core grounding/causality/character/knowledge/continuity/review remain active; conditional war, ensemble, logistics, parallel narrative or spectacle may be N/A. No subject, period, person, tragedy, restraint, silence or spectacle is the default model of a good screenplay. Test stories are replaceable fixtures, never production rules.

For long work, use Global Bible → Episode Working Set → Scene Working Set. Read only relevant facts and creative constraints; author facts never automatically become character knowledge. Use [projection and dependency conventions](references/working-sets.md) to identify context and invalidated consumers before revising. Episode boundaries do not reset body, status, relationship, information or promises.

## Create V1 → Preserve → Explore support → Host decision → Handoff

Create a compact Dramatic Bible using [working state](references/bible.md). Keep one source of truth for historical constraints, characters, knowledge and continuity. Use stable local IDs before domain persistence; never pretend local IDs are durable entity IDs. Retain initial draft and review evidence. Read [craft controls](references/craft.md) for the relevant modules; all twenty dimensions receive an applicability decision, not twenty mandatory blocks of exposition.

### 1. Historical & Logic

Establish scope, evidence-backed causal spine and actor authority before thesis or protagonist. Preserve who actually decided, what they knew and what constrained them. Separate historical causality from dramatized presentation: deleting a fictional action may remove this scene's experiential logic, but cannot become the sole explanation of an established historical outcome. Historical Event != Scene.

Trace travel and message arrival in relative time when dates/distances are uncertain. Do not create precision from a map guess. Freeze forbidden historical contradictions and record alternatives where sources conflict.

### 2. Character & Relationship

Turn traits into behavior under pressure, disagreement, humiliation, uncertainty and unequal rank. Give each principal a few invariants, qualitative capability strengths and limits, and a distinct textual voice. Record elasticity where pressure, fatigue, injury or power changes can produce an earned deviation and a recovery pattern; consistency is not identical behavior. Distinguish formal authority from practical military, political and information power. Private psychology and relationship interpretation are creative positions, not biographical discoveries.

### 3. Story & Rhythm

Derive thesis, narrative POV and the smallest coherent scope from the causal spine. Build story and episode jobs before selecting counts. Compare Beat → Scene → Sequence → Episode → Multi-Episode Arc; each episode records its start/end state, turn, peak, breathing space, after-effect and next pressure. Track pressure, development, breathing, escalation, turn, climax and aftermath as changes in options and knowledge, not timing quotas. Review adjacent scenes for repeated argument, meeting, battle or emotional peak. A quiet scene still changes something; do not remove needed causality merely to accelerate.

For an authorized film-opening redesign, apply the Director's
[film-opening authorship](../director/SKILL.md#film-opening-authorship) at the script
level. Reconsider the entry event and spectator position against the complete film
and ending before preserving the existing first scene. Select a complete opening
passage and its continuation; retain before/after text and affected downstream
references. Reusable assets and successful technical tests remain available without
obliging the new screenplay to keep their former dramatic role.

### 4. Scene & Dialogue

For literary register, long speeches, in-scene documents or a closure comparison, read
[literary craft](references/literary-craft.md). Record who speaks to whom and what the
words change before choosing classical or plain language.

For complete-passage creation or craft repair, read [whole-scene craft proof](references/scene-craft-proof.md). Preserve adopted versions while distinguishing historical constraints from revisable AI expression; write the contest before its explanatory notes.

For each scene establish owner, goal, obstacle, tactic, counter-tactic, information/relationship/power/emotional change, turn and exit hook. Answer why the next scene follows. Track each speaker's knowledge at the moment of choice, not just at scene end. Give necessary dialogue exact words, differentiated diction, surface meaning, intent and subtext; allow a silent scene.

Keep control notes out of the audience-facing body. Preserve stable speaker and spoken-item identity during a local wording revision. Formal Scene persistence still uses the [existing dialogue convention](../../docs/dialogue-layer-content-convention.md), not a second dialogue store. Do not fabricate DIRECT_QUOTE status for adapted words.

### 5. War & Cinematic Expression

If conflict is armed, connect terrain/objective/command to unit behavior and bodily experience, then show the unit and operational consequence. Apply army training, fatigue, cohesion and supply to behavior. A retreat is neither automatically cowardice nor automatically a flawless maneuver. Do not insert a battle to satisfy a checklist; state N/A when war execution lies outside scope.

Use a small number of purposeful motifs and a chosen narrative texture. Test whether action, object, environment or silence can replace explanatory speech. Distinguish shootable visual intention from camera/production specifications. Derive episode opening/closing anchors from aperture, POV, texture and episode job. Subjective emphasis needs an anchor, a narrative purpose and a return to objective context; it cannot alter facts, outcomes or knowledge. See the meta reference for bounded subjectivity and text/artifact provenance.

### 6. Review & Revision

Apply [layered review](references/review.md) to the complete draft and the Bible. Record `problem`, `severity`, `layer`, exact `evidence`, `recommendedRevisionScope`, then declare changed facts, owner, invalidated consumers, unaffected scopes and required rechecks before revising. Recheck affected dependencies even if their words remain unchanged; do not regenerate them automatically. A dialogue defect does not authorize whole-Work regeneration.

The Host chooses revision scope after reading the complete work and advice. It may retain V1 unchanged, expand or reconstruct as well as make focused revisions. Preserve before/after bodies and unchanged-scope hashes. Review actual changed dependencies, without regenerating them automatically. Keep unresolved readings and limitations; ordinary artistic disagreement does not prevent a text candidate. A regression needs genuine creative work and comparison, not a forced improvement or manufactured defect.

Before freeze, run the optional offline [artifact checker](scripts/check_incubation.py) when using the Bible convention. It verifies references, recorded continuity/knowledge, optional meta control records, projection references, revision scope and budget; it cannot judge acting, prose quality or historical truth. Machine success never substitutes for reading the complete script. Freeze records the exact revision and outstanding limitations; user artistic acceptance stays separate from author review.

## Downstream handoff

When handing a frozen screenplay to Scene/Shot planning, extract a thin source-pinned Cinematic Intent sidecar using [intent handoff](references/cinematic-intent.md). Carry relevant POV, opening/closing, subjective moments and bridges with MUST/SHOULD/FREE priorities through scene working sets. Review meaning against the complete shot coverage, not merely matching IDs. A correct screenplay stays frozen when only its shot interpretation needs repair.

An optional Historical Epilogue follows the completed dramatic ending as a separate reading/presentation block. Use the Historical Epilogue section of the [meta convention](references/meta-dramaturgy.md); it supplies concise evidence-backed context without adding explanatory dialogue or altering the closing anchor.

Provide Character + Scene Dramatic State + canonical Dialogue Intent/Subtext + sparse Performance Intent. DPD owns its existing Scene/Beat/Line direction; the screenplay cannot redesign it. Surface/hidden emotion, target and intent are interpretive inputs; tempo, pause, force, breathing and restraint are upstream advisory language for later projection, not additional DPD fields or numeric controls. Shot anchors express the first/last effective perception and its purpose, not lens, frame size, camera movement, exact duration or shot lists. Detailed execution remains with the existing Scene/Shot/Production owners. Do not select voice identities, synthesize speech, render video, choose reference assets or bind provider parameters here. See the [existing DPD boundary](../../docs/dpd-core-contract.md) only when preparing that handoff.

Complete with a readable screenplay, supporting Bible, scoped review/revision trail and a freeze decision. Keep rejected drafts and review reasoning in local working artifacts, outside formal Domain content.

## Earned dramatic peaks

Use the [dramatic peak contract](../../docs/dramatic-peak-editorial-contract.md) to audit CURRENT and propose RECOMMENDED DramaticPeakMap for an Episode/Sequence. Connect setup, pressure, trigger, escalation, payoff, aftermath and character change; MICRO/SCENE/EPISODE are review levels, not required counts. Identify an earned HeroMoment and potential SetPiece only where Canon supports them. Compare pressure/hold/release dynamic range rather than making every scene loud. Name a possible5–30 second memory excerpt, apply the delete-strongest30-seconds test, and explain why the central character earns visual/dramatic power. Missing memory is WARN/REVISE, not automatic FAIL. Keep proposals in working artifacts until story revision is authorized; no new business entities, coverage Skill or generation follows automatically.

When a visual route is explicitly selected, retain the same narrative, character knowledge and [Literary Craft](references/literary-craft.md). Pass the resolved [visual-route sidecar](../../docs/visual-route-contract.md) to design/director owners; route changes do not rewrite screenplay or turn character speech into narrator speech.

## Screenplay Readiness before direction

For full-film/episode direction, produce the source-bound [Screenplay Readiness Review](references/screenplay-readiness.md) before final preproduction. This is a blocking upstream challenge with evidence, not an average score or compulsory romance. Preserve historical opportunities and explicit exclusion reasons.

## Creative runtime and revision scope

Every new full Director Screenplay and Production Book must supply expected_runtime using the nested DirectorRuntimeEstimate facet. See [creative runtime accounting](../../docs/director-runtime.md). Derive target and range from actual scene obligations, never from provider clip limits or a fixed example length. Dialogue, action, silent performance and transition form one non-overlapping duration partition; music-bearing, intentional no-music and unresolved music form a separate overlay partition. Report Scene, Sequence and Act totals. A proposal estimate is not measured media or a mandate to fill time.

Performance architecture validation is not screenplay artistic approval. A user-authorized upstream rewrite may change scenes, relationships, action and rhythm; preserve prior revisions and explicitly mark downstream mappings pending. Keep historical attestation, reconstruction and invention separate; source-based words spoken by a character are not automatically the author's historical conclusion. All work-specific choices and runtime values belong to the work's artifacts. Stop at the requested review boundary.


## Actor-playable screenplay handoff (Screenplay R2)

Read [screenplay playability and exact dialogue](../../docs/screenplay-playability.md). Audit every Scene/important Beat using PLAYABLE, LITERARY_ONLY, INTERNAL_STATE, AMBIGUOUS_DIALOGUE, EXPLANATORY_PROSE, OVERDIRECTED and ABSTRACT_EMOTION. Author the clean screenplay plus the existing-DPD playability sidecar; review dialogue in its body state before downstream handoff.

## Local original and decision operations

Use `screenplay.retain_context` and `screenplay.retain_thread` for exact source originals and classified long-arc notes; `screenplay.begin` returns the autonomous author packet. Preserve the complete original with `screenplay.retain_v1` before requesting `screenplay.support_packet`; retain real advisory text with `screenplay.retain_support`. The Host selects with `screenplay.decide`, then supplies a complete scene through `screenplay.shot_handoff`. These local candidate operations grant no formal adoption or production permission.