Seedance Prompt ShortSAFE
Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0
Overview
Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0
b11e1df49a75OBSERVED · 2026-10-13Host compatibility
What the documentation claims. We have not run a compatibility test.
| Host | Status | Notes |
|---|---|---|
| openclaw | mentioned |
What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
---
name: seedance-prompt-short
description: "This skill should be used when the user asks for a compact Seedance 2.0 prompt, short Chinese prompt, prompt compression, 30-100 word output, or removal of unnecessary prompt language."
license: MIT
user-invocable: true
tags:
- prompt-compression
- chinese-prompt
- seedance-20
metadata:
version: "6.7.0"
updated: "2026-08-01"
parent: "seedance-20"
author: "Iamemily2050 (@iamemily2050)"
repository: "https://github.com/Emily2040/seedance-2.0"
openclaw:
emoji: "🎬"
homepage: "https://github.com/Emily2040/seedance-2.0"
---
# seedance-prompt-short
Before producing prompt text, a prompt-ready block, a rewrite, an example, or a compiled clip, load the [Director's Read](../../references/directors-read.md), classify the brief, and complete its canonical narrative or non-narrative record. Translate that record into visible or audible carriers and keep its internal labels out of final generation prose.
Compress Seedance prompts without losing the production signal. A short prompt still needs mode, subject, action, camera, light, sound when useful, and constraints. Remove filler before removing physical details.
Load the [Director's Read](../../references/directors-read.md) before compression, even when the requested output is only one line. Briefs with a dramatic story turn complete its ten-field narrative record first. Observation/performance without that turn and utility briefs use its two-line non-narrative intent and refusal without fabricated drama; short output must still preserve the requested expression or rhythm. Short means compressed output, not skipped judgment.
When sequence state is present, compression must preserve continuity locks, exact reference tags, actual opening state, current clip action, endpoint, completed beat exclusions, and reserved future beats. Do not compress away the words that keep a continuation from replaying completed action or leaking future action.
The [moderation pre-screen](../../references/moderation-prescreen.md) runs on the compressed prompt too: compression that drops a clarifying word can expose a cue, and a short prompt is refused as readily as a long one. Never write an involuntary outcome as the endpoint; the model stages accidents as intentions. The short route still builds the [shot table](../../references/shot-table.md) for any storyboard or multi-person clip and may deliver the prompt alone, but a cell it cannot fill is a question it must ask before delivering. Compression never removes a lock line, a stated position or a feeling word, never turns a literal action back into an idiom, and never replaces idle business with a freeze; see [direct-for-the-model](../../references/direct-for-the-model.md).
Storyboard clips keep their structure when compressed: numbered shots in event order, the cut written in words, no absolute seconds inside a block on Seedance 2.0. The shot-count ladder from [multishot-grammar](../../references/multishot-grammar.md) shrinks to one line per rung and one finished prompt at the chosen rung; when the user has fixed the count, state its rung in one line and write it.
## Intent
Compression is an act of judgment about what the user loves most. What survives the cut is the soul of their shot; everything else goes first. If the user would mourn a deleted word, it was never filler.
## Compression Priority
Preserve in this order:
1. Reference tags and their role.
2. Subject or product identity.
3. Action verb and visible endpoint.
4. For narrative work, the visible suppressed behavior and non-transferable detail that carry the Director's Read.
5. One camera move.
6. Physical light source or atmosphere.
7. Audio cue or silence instruction.
8. Safety, IP, or continuity constraint.
9. Sequence state clauses: actual opening state, continuity locks, completed beats, and reserved beats.
Delete generic adjectives, duplicate style labels, obvious background details, secondary camera moves, and secondary actions before deleting preservation constraints.
For bilingual or mixed-language compression, load [multilingual-community-examples](../../references/multilingual-community-examples.md). Keep only the language mix that clarifies reference roles, dialogue, camera terms, or safe production constraints.
## Compact Templates
| Need | Template |
|---|---|
| T2V | `[Subject] [action and endpoint] in [scene]. Camera: [one move]. Light/style: [physical source]. Sound: [cue]. Constraint: [risk/continuity].` |
| I2V | `@Image1 preserved; only [motion/light/camera] changes. Camera: [one move]. Sound: [cue]. Constraint: [what must not change].` |
| V2V | `@Video1 controls [motion/camera/timing] only; new subject [anchor]. [Action]. Do not transfer [identity/scene/logo].` |
| Chinese | `@Image1为参考,严格保持[主体]不变;仅加入[动作/光线/镜头]。声音:[提示]。` |
| Storyboard | `Shot 1: [cut or move], [action and expression], [space], [sound]. Shot 2: cut to [framing], [action to endpoint], [sound]. Constraint: [risk/continuity].` |
| Chinese storyboard | `镜头1:[运镜或切镜],[动作与表情],[声音]。镜头2:镜头切至[机位],[动作与终点],[声音]。保持无字幕。` |
## Output Contract
Return one compact prompt, ideally 30-100 English words or an equivalent Chinese prompt when the user asks for Chinese or maximum compression. Translate the Director's Read into visible or audible carriers and keep its internal labels out of the prompt. Include a one-line note only if something important was removed.Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | NA |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (0)
No findings outside the package's declared scope.
Gates applied: no_behavioural_pass.
b11e1df49a75full audit observations/trust-audit/skill/emily2040__seedance-prompt-short.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-13 | b11e1df49a75 | SAFE | B | 89 | first audit |
Questions
What does the Seedance Prompt Short skill do?
Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0
Is Seedance Prompt Short safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can Seedance Prompt Short access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
Which assistants does Seedance Prompt Short work with?
Its documentation mentions openclaw. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (b11e1df49a75), read on 2026-10-13. The repository is watched, and a new audit runs when it changes — this is the first audit.