Hubbestaiskill.com

Home / All Skills / Content & Media Creation / dbs-script-flow

Content & Media Creation · dontbesilent2025/dbskill

dbs-script-flow

Inspects short video verbatim scripts for paragraph transitions, information density, and spoken fluency, pinpointing moments where viewers might swipe away. Used when users request logic continuity checks, script smoothness reviews, or drop-off risk evaluations.

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit

Author dontbesilent2025GitHub dontbesilent2025/dbskill ↗Stars 10121License CC BY-NC 4.0（非商业授权）：仓库根 LICENSE 为 Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)，Copyright (c) 2026 dontbesilent；GitHub API license.spdx_id = NOASSERTION （CC BY-NC 4.0 非 SPDX 标准文本，GitHub 无法自动识别），不得用于商业用途Commit 8b8e33f1ec

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

The entire skill consists of fixed three-stage inspection instructions (segmentation → segment-by-segment 3D scanning → risk tiering); it contains no scripts, no templates, and no external dependencies, fulfilling its capabilities purely through prompt orchestration in SKILL.md.

skills/dbs-script-flow/SKILL.md

把逐字稿按自然段落（话题切换点）分成若干段，每段标注：

What it does: codifies subjective assessments like "whether a script flows well" into a repeatable mechanical procedure (segment first, run through the checklist segment-by-segment, then prioritize); the model executes this without any tools, relying solely on two files: SKILL.md and agents/openai.yaml.

Attributes viewer swipe-aways to three locatable root causes (logical breaks / declining information density / spoken tongue-twisters), ensuring every conclusion maps to specific sentences rather than generic feedback.

skills/dbs-script-flow/SKILL.md

**完播率的敌人不是内容差，而是某一个瞬间观众跟不上了。**

The three-factor model maps to three inspection dimensions; each dimension is paired with a "self-inquiry" checklist and a "fix method," forming a complete diagnosis-to-repair loop.

Risks are categorized into three tiers (🔴 High / 🟡 Medium / 🟢 Minor suggestion) with mandatory prioritization for high risks, preventing genuine issues from being drowned out by trivial suggestions.

skills/dbs-script-flow/SKILL.md

| 🔴 高风险 | 观众大概率在这里划走 | 逻辑断裂、信息密度严重下降 |

This tiering dictates both output ordering and revision priority list (rule 4 under cautions: "**Prioritize high-risk markings**").

The output structure is strictly defined as a rigid scaffold (paragraph structure table → issue list → overall evaluation), and it must proactively ask whether revisions are desired upon completing the inspection.

skills/dbs-script-flow/SKILL.md

**不要等用户主动要求**——大部分时候用户做检查的目的就是改稿，主动问省一次往返。

What it does: turns "whether to proceed" into an explicit confirmation while preserving user autonomy (the original text specifies "**do not revise without asking**").

The revision stage uses "markup-based overlay" rather than full-text replacement: strikethroughs indicate deletions, 🆕 indicates additions, ⚠️ indicates self-retracted assessments, and multi-round feedback appends without overwriting.

skills/dbs-script-flow/SKILL.md

**永不抹除用户原文**。改稿是在原文上叠加标记，不是用新版本替换旧版本。用户必须能一眼看出：

Supporting rule: changes appear in pairs (~~original~~ + 🆕 new version), followed by a "Changelog Table" at the end for effortless user review.

Explicitly declares output boundaries: does not create, save, or open local files, delivering Markdown strictly within the current conversation.

skills/dbs-script-flow/SKILL.md

不得依赖特定笔记软件，不负责创建、保存或打开本地审阅文件；文件渲染与应用打开由项目外部能力处理。

This marks the dividing line between this skill and "disk-persisting skills" (such as dbs-content-system), serving as the basis for its zero file-write risk.

Maintains clear content boundaries: only modifies "how it is said" without altering "what is said"; revisions must preserve the user's authentic tone and repair suggestions must be naturally readable aloud.

skills/dbs-script-flow/SKILL.md

不要改用户的论点、案例、数据

The same section specifies "keep the user's tone... do not turn the user's words into generic AI speak," preventing AI rewrites from erasing personal style.

2Core Capabilities

01Verbatim script segmentation and functional labeling (hook/background setup/argumentation/case study/transition/climax/outro)

02Dimension 1 - Logical cohesion: checks jumps across paragraphs and provides speakable transition phrases

03Dimension 2 - Information density: detects repetition, circling around points, and "truisms/filler", marking items as "can be deleted"

04Dimension 3 - Spoken fluency: detects overly long sentences (>30 characters/words), formal written phrasing, rhetorical self-answering, and fabricated jargon

05Three-tier risk labeling and priority sorting

06Markup-style revision (strikethrough / 🆕 / ⚠️) + end changelog table

07Proactively asks for revision consent after inspection and confirms shooting format before rewriting (whiteboard videos have different criteria)

08Misjudgment safeguards: conducts a "mental-fill test" before suggesting paragraph deletion to distinguish necessary context from bloat

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Can be auto-loaded

For Administrators

Register source and version

Evidence & Triggers (File Paths & Excerpts)

SKILL.md: Pure prompt instructions and rule text; executes no scripts and causes no external side effects

Audit Notes: Pure instruction-based skill: no script execution, no outgoing network calls, no credential reads, and no local file writes (SKILL.md explicitly states it does not create/save/open files); assets consist strictly of SKILL.md + agents/openai.yaml. The only considerations relate to the subjective nature of diagnostic conclusions and the credibility of revised copy, neither of which constitutes a code-level risk.

5Second-Pass Verification

[ok] 目录内仅 2 个文件、无脚本可执行 — find skills/dbs-script-flow -type f 仅返回 SKILL.md 与 agents/openai.yaml。
[ok] 无网络外发 / 无 CLI 依赖 — 两文件全文扫描 URL、curl、wget、npx、bash、python、API_KEY 等 token 零命中。
[ok] 无本地文件写入 — SKILL.md 明文"不负责创建、保存或打开本地审阅文件"，且无 write/save 类指令。
[ok] 三维度检查模型与风险分级的描述属实 — 原文含「### 第一步：分段」「#### 维度 1：逻辑衔接（前后段之间）」「🔴/🟡/🟢」三档定义，逐条与侦查结论吻合。
[ok] 主动询问改稿 + 保留原文的改稿机制 — 原文含"**不要等用户主动要求**"与"**永不抹除用户原文**"，及 ~~/🆕/⚠️ 三种标记符表。
[ok] 实战案例真实存在于源码 — SKILL.md 末尾「### 案例：「小红书被告了」逐字稿（2026-04-09）」含两处问题与改法，非侦查编造。
[ok] skill 定位与 README 一致 — README.md 把 dbs-script-flow 与 dbs-resonate/dbs-spread 并列在"检查文稿共鸣、逻辑与传播性"，分工为微观衔接检查。
[ok] description 与正文能力是否夸大 — "检查段落衔接、信息密度和口播流畅度，定位观众可能划走的位置"三项均在正文有对应的独立检查维度，无越界承诺。

6Verdict & Scenarios

Actionable diagnosis: breaks down "flow" into three dimensions, each with a self-inquiry checklist and fix strategy, anchoring conclusions directly to specific sentences.

Restrained revision methodology: markup overlays + preserving original text + changelog table; multi-round feedback appends without overwriting, preventing lost revision history.

Built-in anti-misdiagnosis mechanism: runs a "mental-fill test" before declaring off-topic points, applies separate standards for whiteboard videos, and prefers omitting minor suggestions over burying high risks.

Zero side effects & cross-host compatible: pure prompt, no scripts or network calls, reads/writes no files; works across any agent supporting SKILL.md.

Best for: Ideal for short-form video creators and directors conducting pre-publishing checkups: verifies whether a script loses viewer attention at any second and refines issues into directly speakable phrasing; also well-suited for cautious revision workflows requiring side-by-side original comparisons (strikethrough + 🆕 markup).

Not for: Not suitable for evaluations requiring empirical retention/completion metrics (skill explicitly does not predict platform analytics); not suitable for workflows needing disk-based file reviews or note-taking app integrations (outputs in-chat Markdown only); not suitable for changing core opinions or topics (modifies only "how it is said"); not suitable for scenarios with raw video files lacking transcripts (requires extracting transcripts first via dbs-video-extract, etc.).

Install Agent-Ready

① Verified Mirror 2026-09-15

Option A · Direct DownloadDownload dbs-script-flow.tar.gz

sha256: dcec68913a374fbf…

Option B · Copy Guide for Agent

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 8b8e33f1ec.