Skip to content

papperrollinggery/Paperrolling-DIRcreative-SKILL / skills/dircreative

dircreative

Verified

Use after explicit $dircreative for film concepts, stories, scripts, director or creative groups, shot/asset planning, reference-video analysis, and image/video prompts. Real media execution requires the corresponding authorization. Validated ADCO handoffs are supported. DIRcreative source maintenance is outside this Skill.

FileKindSize
SKILL.mdskill md8.3 KB
agents/openai.yamlconfig239 B
assets/visualizations/decision-surface.cssother18.4 KB
assets/visualizations/decision-surface.jsscript32.6 KB
assets/visualizations/stage-surface-registry.jsonconfig7.2 KB
references/asset-foundation-pass.mdmarkdown5.3 KB
references/character-master-sheet.mdmarkdown21.5 KB
references/client-story.mdmarkdown2.3 KB
references/copy-script.mdmarkdown3.7 KB
references/film-development.mdmarkdown8.3 KB
references/generation-delivery.mdmarkdown8.6 KB
references/humanization-workflow.mdmarkdown14.2 KB
references/image-execution.mdmarkdown20.5 KB
references/image-generation-diagnostics.mdmarkdown2.8 KB
references/product-cg-direction.mdmarkdown9.7 KB
references/product-cg/capsules/elastic-fiber-cg.jsonconfig4.8 KB
references/product-cg/capsules/graphic-modular-cg.jsonconfig4.9 KB
references/product-cg/capsules/pastel-dry-powder-cg.jsonconfig5.7 KB
references/product-cg/capsules/precision-hard-surface-cg.jsonconfig4.9 KB
references/product-cg/capsules/sculptural-luxury-cg.jsonconfig4.8 KB
references/product-cg/capsules/viscous-sensory-cg.jsonconfig4.8 KB
references/product-cg/catalog.jsonconfig3.9 KB
references/project-hygiene.mdmarkdown2.7 KB
references/prompt-model.mdmarkdown3.9 KB
references/prompt-structure.mdmarkdown15.7 KB
references/script-to-seedance.mdmarkdown13.1 KB
references/shot-development.mdmarkdown8.0 KB
references/shot-storyboard.mdmarkdown3.2 KB
references/spatial-discussion.mdmarkdown10.0 KB
references/specialist-exchange.mdmarkdown2.3 KB
references/storyboard-coverage.mdmarkdown8.8 KB
references/storyboard-frame-to-jingzao.mdmarkdown12.5 KB
references/storyboard-motion-planning.mdmarkdown8.1 KB
references/video-distillation.mdmarkdown7.2 KB
references/video-quality-prefix.mdmarkdown2.5 KB
references/visual-asset-to-jingzao.mdmarkdown5.6 KB
references/visual-skill-stack.mdmarkdown13.1 KB
routes/delivery-audit.mdmarkdown1.7 KB
routes/fast-task.mdmarkdown1.2 KB
routes/studio-development.mdmarkdown2.4 KB
runtime/review-trust-registry.jsonconfig102 B
runtime/routing-policy.yamlconfig10.6 KB
runtime/state-snapshot.schema.jsonconfig7.3 KB
runtime/visual-asset-plan.schema.jsonconfig15.1 KB
runtime/visual-skill-policy.jsonconfig36.2 KB

references/script-to-seedance.md

View on GitHub
# Script-to-Seedance Handoff

Read this contract only when the selected Skill Stack scenario is
`script_to_seedance`. It lets `convert-script-to-seedance` compile locked
DIRcreative work without becoming a second story, shot, asset, or state owner.
Use `prompt-structure.md` for the copyable scene: real reference roles and stable
facts, followed by causal events and their camera, environment, performance and
sound. Preserve an explicit user structure, including an authored quality block,
without imposing twelve default headings or conflicting cinematic defaults on
all scenes. `video-quality-prefix.md` covers source-derived compiler formatting.
Persist and validate the machine packet with
`docs/film-preproduction/schemas/script-to-seedance-handoff.schema.json` and
`python3 scripts/dircreative_script_to_seedance_handoff.py validate <packet>`.

## Ownership

- DIRcreative owns the approved story, authoritative script, canonical shot IDs
  and timecodes, generation units, entity ownership, visual and continuity
  locks, asset roles, audio policy, and current artifact state.
- `convert-script-to-seedance` owns only the bounded conversion of those inputs
  into Seedance-ready unit prompts, continuity snapshots, and stable asset/audio
  slot ledgers.
- The converter may budget performance, split an overloaded unit at a natural
  boundary, and make the smallest executable wording change. It must not invent
  or re-author dialogue, actions, props, wardrobe, geography, identities,
  claims, or downstream approval.
- DIRcreative consumes the converted prompts back into the video prompt
  manifest and remains responsible for Prompt IR validation, exact capability
  card resolution, direct-input policy, rights, QA, retry, and status.

For a Seedance 2.5 target, apply the installed `mr-li-seedance-25` 2.0 method
first: confirm the project visual baseline, identify speaking-role voice inputs,
estimate real performance capacity, bind assets/audio once per requested scope,
and preserve its heading-free natural-paragraph format. The method body and its
selected references use isolated method context when paired with this compiler.
The handoff records `applied_unverified`; only the host's matching Skill Stack
body/reference readback may promote the method to applied for that run.
Keep project and one-off segment limits separate; run capacity preflight before
writing and again after; cut on speaker changes without forcing one sentence per
shot; bind only current-segment assets and do not freeze changing held/damaged
states; return one natural generation unit plus its exact next source beat.
Target intent alone never establishes model facts; the exact card/version/surface,
DIR ownership, asset gate, handoff validator and preflight remain authoritative.
If the provider is absent, retain the converter and report non-adoption.
Encode that branch as `method_application.status: not_applied` with the observed
version and reason; never forge a 2.0 application block to satisfy the schema.

Read the current body and four production references: `prompt-writing`,
`continuity-and-duration`, `context-protocol`, `segment-ending`. Replace this pack
per stage; do not add regression instructions, changelog or aggregate runtime.
The host body-read request carries DIR ownership; selection is not adoption.

Use the 2.0 continuity, natural-ending, voice and medium-aware methods. DIR retains
its adopted source, five-view human master, canonical IDs, camera decisions,
reference promotion, state and authorization. Focal-length preferences remain
advisory. Do not import onboarding, repeated menus, foreign character layouts or
a second controller. Existing explicit creation/adoption authority needs no new
provider approval.

Optional previews use DIR narrative-frame/coverage roles and inherited all/key/skip
scope; they do not block prompt delivery or become clean/model inputs automatically.
Supported annotated references retain DIR's explicit promotion/QA/Prompt IR gates.
Keep current IDs and project paths. Apply the chosen medium: 2D work does not
acquire photographic blur, grain or atmosphere by default. Inspect pixels and
native detail separately.

## Required input packet

Pass only current, approved material:

1. authoritative script scope and immutable dialogue lines, each with a stable
   `dialogue_line_id` plus owning shot IDs;
2. canonical `S01`-style shot IDs with continuous start/end timecodes;
3. explicit narrative-node to shot to generation-unit mappings;
4. generation-unit IDs, target duration, incoming/outgoing handoff keys, and
   visible incoming/outgoing states;
5. entity ownership for every person, vehicle, product, prop, action, dialogue,
   and event sound;
6. continuity locks for identity, wardrobe, props, scene geography, axis,
   screen direction, light, damage/state progression, and audio;
7. asset ledger entries with source status, rights, production role,
   direct-input policy, user lock, and required shot/unit bindings;
8. audio policy, provider reference-slot limit, and the selected exact model
   card and provider surface.
9. the complete `asset_foundation_pass_v1` file and bound asset stress report,
   including real file hashes and scoped compile permission. Legacy v1/v2
   reports remain valid; v3 character sheets follow the single contract in
   `character-master-sheet.md` before this handoff consumes their active hash.

If the source consists only of a nine-grid or another generated image sequence,
first classify it as `planning_only`, recover narrative beats, and derive the
technical shot, asset-foundation, and generation-unit truth. Cross-frame visual
similarity is not identity, vehicle, prop, scene, or clean-frame evidence.

## Asset foundation before direct inputs

Use the smallest sufficient asset plan, in this dependency order:

1. recurring character, costume-state, vehicle/product, critical-prop, and
   irreversible damage/open/closed state-family identity references;
2. distinct scene geography and camera-FOV references;
3. a professional storyboard/motion map for human planning when needed;
4. only the clean first/key/end frames required by the selected generation
   route.

Separate `human_planning_board` from `model_layout_reference`. Boards with
labels, arrows, grids, panels, floor plans, or timing notes stay human-only.
Narrative frames stay human planning evidence unless an explicit promotion
receipt and independent QA authorize a narrower model-reference role. Neither
class becomes a literal first frame. A clean frame is a separate text-free,
border-free, QA-passed asset.

## Deterministic slot crosswalk

Do not renumber or compress existing slots. Every recurring character has one
active master sheet per generation unit: headed by default, or its headless-safe
derivative when explicitly selected. Do not attach both as equal identity
references. Add a detail sheet only when the shot can resolve one of its named
callouts. Every attached reference must have one row containing:

```text
source_asset_id: <DIR asset id>
reference_id: <null for active canonical asset; stress-report reference ID for a supporting detail>
asset_version: <immutable version>
relative_path: <project-relative canonical file>
sha256: <actual canonical bytes>
converter_slot: 【图片N】 | 【音频N】
platform_slot: @Image N | @Audio N
global_number: N
local_order_by_gu: {GUxx: N}
role: <identity / scene_fov / clean_frame / audio reference>
shot_ids: [Sxx]
unit_ids: [Uxx]
direct_input_policy: allowed | conditional | forbidden | planning_only
attached_to_run: true | false
status: available | planned | optional | post_dub | rejected | planning_only
primary_owner: <one Skill or controller>
assist: [<optional helpers>]
validator: <optional validator>
preserve: [...]
anti_misread: [...]
must_not_control: [...]
literal_frame_boolean: true | false
```

`【图片N】` and `【音频N】` are converter-ledger notation. The final Seedance
terminal surface uses only the exact current platform notation selected by the
DIRcreative adapter, such as `@Image N`, `@Video N`, and `@Audio N`. Internal
asset IDs, paths, hashes, QA fields, missing slots, and planning-only assets stay
out of the pasted prompt.

Only `status: available` plus `attached_to_run: true` may enter a terminal
prompt. Every such image binding must byte-match the same version/path/hash in
the canonical foundation provenance and stress-tested asset record. One
character asset may therefore have one canonical active binding plus a
shot-needed `character_detail_sheet` binding whose `reference_id`, path, hash,
role and detail contract match the same stress report. Exact-graphic references
remain planning-only proof for deterministic detail construction and can never
be attached directly to Seedance. Planned or optional audio remains in the ledger and is never compiled
as an existing `@Audio` slot. Before export, reject duplicate/mismatched global
numbers, missing or duplicate per-unit local order, and any unit whose attached
references exceed the verified provider limit. Resolve overflow by explicit
reference sovereignty, not silent deletion: identity/state > vehicle/product >
scene > model-layout > composition.

## Unit compilation

- Seedance 2.5 `authoritative_script.next_source_start` binds the first still unmade
  source line, exact text and line hash; `null` requires actual scene-coverage
  review. The final export unit uses `source_line:N`/`segment_only` if source
  remains. The validator checks real locators and completion consistency, not
  full narrative coverage or an automatic end-of-scene assumption.
- Known character voice bindings carry `voice_owner_entity_id` and their applicable
  unit scope. Each speaking unit needs one usable scoped voice when available;
  silent units cannot attach it. Planned/unavailable/planning-only audio creates
  no binding requirement or repeated question. Verify media before execution.
- Compile one locally rebased generation unit at a time.
- Preserve shot IDs and time ranges outside or alongside the prompt so the edit
  assembly remains traceable.
- Each unit has one recoverable initial visible state, pressure/trigger, action
  path, camera start/end target, timing beat, final visible state, and sound or
  silence policy.
- Adjacent units must agree on entity positions, held props, wardrobe, damage,
  light, screen direction, emotional state, and audio handoff.
- Validate every within-unit state change against a declared predecessor and
  require exact outgoing-to-incoming state equality between adjacent units.
- When the requested performance exceeds the selected route, return a split
  recommendation tied to a natural shot/unit boundary; never squeeze the whole
  assembly into one executable prompt.
- For Seedance 2.5, 30 seconds is an exact-card ceiling, not a target. End at a
  natural action, dialogue, discovery, or dramatic boundary and never fill the
  remaining capacity with unsourced acting, reactions, props, geography, or
  decorative empty shots.
- User-supplied dialogue remains exact. Proposed optional empty shots or
  connective beats must be visibly optional and cannot silently change the
  authoritative script.
- Before compiling any unit, run the handoff validator with
  `--asset-foundation-pass`, `--asset-stress-report`, `--asset-artifact-root`,
  `--asset-review-receipt`, and `--asset-review-signature`. The review authority
  must already be configured in the host-owned trust registry; CLI input cannot
  add it. Missing project/asset bindings, host-pinned signatures, stages,
  unresolved gaps, incomplete
  stress matrices, unverified verdicts, or requested shots outside the allowed
  scope fail closed.
- Each `prompt_unit` carries the actual non-empty terminal `prompt_text`, a
  unique `artifact_id`, the exact packet status, and a SHA-256 over the UTF-8
  prompt bytes. A list of binding IDs without the compiled prompt is not a
  completed handoff.
- The platform slots visible in each prompt must equal that unit's locally
  rebased attached bindings. Missing, extra, stale, or wrong-media slots fail
  validation. Dialogue is validated by stable line identity and owning shot/unit
  scope; if two distinct lines use the same text, the prompt must contain two
  occurrences in that unit rather than collapsing them into one.

## Output packet

Return:

- one hash-bound Seedance prompt payload per generation unit;
- the complete image/audio slot crosswalk and upload order;
- the validated machine handoff packet with node/shot/unit coverage, provider
  limits, and binding availability;
- incoming/outgoing continuity snapshots per unit;
- any split or missing-asset blocker tied to exact shot/unit IDs;
- prompt-only or instructions-only status unless a separate authorized
  execution has a verified external receipt.

Run `ai-video-prompt-preflight` after conversion when validation is requested.
Generated video review and edit assembly remain later stages; the converter
must not claim that prompt export produced a video.

Keep prompt delivery, optional saving, image work and video execution as separate
existing artifact states. A later image/video/save failure does not rewind an
already delivered source cursor. A prompt revision preserves its unit ID and
marks only affected images or descendants stale. Display the requested prompt
or usable files; do not substitute a successful manifest for the user's result.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall