Atlas / Skills / emily2040 / Seedance Antislop

Seedance AntislopSAFE

skills/emily2040/seedance-antislop

Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0

Verdict
SAFE
Grade
B
Trust score
89 /100
Version
6.7.0
Hosts
1 documented
License
MIT
Stars
7,642
01

Overview

Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0

Read from source at commit b11e1df49a75OBSERVED · 2026-10-13
02

Host compatibility

What the documentation claims. We have not run a compatibility test.

HostStatusNotes
openclawmentioned
03

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-antislop
description: "This skill should be used when a Seedance 2.0 prompt contains generic AI filler, hollow superlatives, vague cinematic language, bloated adjectives, weak verbs, or needs sharper production-specific wording."
license: MIT
user-invocable: true
tags:
  - prompt-quality
  - anti-slop
  - 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-antislop

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.

Clarify wording that hides an unresolved production decision. Preserve useful style, genre and mood language; concrete detail should support the user's direction, not replace it with a stock look. These are editorial choices, not measured rules about which words improve Seedance output.

## Intent

Keep the intended energy, medium, era, palette and performance. A word such as "epic" may communicate a useful ambition even when the brief still needs a decision about scale or stakes. Do not infer the user's taste or confidence from their adjectives.

## Rewrite Pass

1. Identify what is already decided: creative intent, exact dialogue, actual reference tokens, continuity, safety boundaries, exclusions and requested settings. Preserve those through the edit.
2. Check whether an abstract phrase leaves a material choice unresolved. Keep it when context is sufficient. Otherwise use the brief to clarify it; if the user has delegated creative choices, propose a coherent interpretation and state any material assumption. Ask only when a necessary choice remains unresolved outside that scope.
3. Remove redundant praise where it adds no distinct intent. Surface conflicting requirements instead of silently choosing one. Add only details that serve this brief; neither a camera move nor a lighting recipe is required for every style word. Load [anti-slop-lexicon](../../references/anti-slop-lexicon.md) for conditional repairs and delivery requirements.
4. Apply a length limit only when the user requests it or the selected operation documents it. Do not invent a universal character budget or silently discard must-haves to fit one.

## Do Not Over-Correct

Keep useful labels such as noir, documentary or ultra-realistic when they describe the chosen look. Observable detail is a way to resolve ambiguity, not a test that every phrase must pass. A locked shot may already serve an energetic scene; "dynamic" does not automatically authorize a lateral track. Likewise, "professional" does not require a clean product tabletop.

Treat output resolution as a delivery requirement. Preserve the requested value and carry it into a supported setting when available. Prompt text alone does not guarantee that resolution. If support is unknown or the value is unavailable, state the gap and offer an explicit option within the user's scope; do not silently lower the target, switch providers, add an upscale or spend credits.

For English wording, load [seedance-vocab-en](../seedance-vocab-en/SKILL.md) with [English vocabulary](../../references/vocab/en.md). For another prompt language, consult its Slop Traps table only when needed: [Chinese](../../references/vocab/zh.md), [Japanese](../../references/vocab/ja.md), [Korean](../../references/vocab/ko.md), [Spanish](../../references/vocab/es.md), or [Russian](../../references/vocab/ru.md). These are contextual wording examples, not language-wide lists of useless words.

## Output Contract

Return the revised prompt in the requested format and briefly explain material edits when useful. Keep delivery settings separate from scene prose, retaining any unresolved requirement. Disclose content changes or assumptions; a wording rewrite does not authorize another submission. Provide a word-by-word removal table only when requested or needed for the review.
04

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.

LayerWhat it checksResult
L0Provenance & inventoryPASS
L1Static analysis of the codeNA
L2Instruction surface (what it tells the agent)PASS
L3Class-specific surfacePASS
L4Behavioural (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.

Audited 2026-10-13 · audit v0.4.1 · source sha b11e1df49a75full audit observations/trust-audit/skill/emily2040__seedance-antislop.json · Report an issue / request a re-scan
05

Audit history

Every audit this skill has had.

DateSourceVerdictGradeScoreChange
2026-10-13b11e1df49a75SAFEB89first audit
06

Questions

What does the Seedance Antislop skill do?

Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0

Is Seedance Antislop 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 Antislop access on my machine?

The audit observed no filesystem, network or shell use at all in its source.

Which assistants does Seedance Antislop 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.

Advertisement