Hubbestaiskill.com

Home / All Skills / Content & Media Creation / dbs-content-risk-check

Content & Media Creation · dontbesilent2025/dbskill

dbs-content-risk-check

Pre-publishing content risk checker for creators. Performs sentence-by-sentence checks across titles, copy, image text, subtitles, voiceover scripts, comments, profile bios, and video frames on self-media, social media, and content platforms. Separately evaluates what platform algorithms might flag versus substantive policy issues within the content itself, pinpointing exact locations and providing minimal necessary edits. When video contextual boundaries need clarification, it generates truthful and accurate three-line video overlay disclaimers. Use when users ask to "check for sensitive words," "screen risks before publishing," "find potential policy violations in this draft," "check if this RED/Xiaohongshu post is safe to publish," or "suggest disclaimers for my video." Preserves strong opinions, emotional tension, personal style, and viral hooks.

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

Author dontbesilent2025GitHub dontbesilent2025/dbskill ↗Stars 10121License CC BY-NC 4.0（Attribution-NonCommercial 4.0 International；仓库根 LICENSE 明文 NonCommercial — 非商业授权，不得商用；GitHub API license=NOASSERTION/Other，因 CC BY-NC 不在 SPDX 常见许可标识内；Copyright (c) 2026 dontbesilent）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

Pure prompt-based pre-publishing risk check skill: Outputs a natural language report ready for creators by default, without deciding whether to publish or rewriting the entire piece.

skills/dbs-content-risk-check/SKILL.md

默认输出普通创作者可以直接使用的自然语言报告，不替用户作出发布决定，也不默认重写全文。

Positioned as a "pre-publish health check + minimal edits," not an automated ghostwriting or publishing tool.

Core dual-track methodology: One track checks algorithmic trigger risks, while the other audits substantive content issues; both conclusions stand independently.

skills/dbs-content-risk-check/SKILL.md

1. **平台机器可能识别什么**：广告结构、联系方式、外链、二维码、连续数字、敏感词面、OCR 噪声和重复模式可能触发自动审核；

The second track evaluates substantive issues based on complete semantics, including fraud, false advertising, prohibited traffic redirection, restricted promotions, medical efficacy claims, infringement, privacy, attacks, and hazardous behavior.

No conflation between the two layers: Machine risks must only be labeled as "may trigger automated review" without fabricating violation reasons; substantive issues must explain their independent mechanisms.

skills/dbs-content-risk-check/SKILL.md

机器误伤只能写成「可能触发机器审核」，不能补造合理违规原因。

Crucial constraint to prevent the model from fabricating violations based on sensitive words, distinguishing it from simplistic blocklist tools.

Fixed five-surface inspection (text, full semantics, audio/visuals, publishing context, platform conditions); if materials are incomplete, audit what is available and flag uninspected scopes.

skills/dbs-content-risk-check/SKILL.md

### 第二步：检查五个表面

When inputs are incomplete, guidelines state: text-only reviews must end with a reminder that images/profiles/comments were not inspected; OCR alone cannot prove visual safety; e.g., "Only text was checked this time; images, homepage profiles, and comment sections were not included in this evaluation."

Algorithmic risks evaluated across three evidence tiers: Explicit trigger signals / List-based signals / Weak lexical cues; remain silent on weak cues.

skills/dbs-content-risk-check/SKILL.md

只有弱词面线索时保持沉默。

List-based signals (brands, competing platforms, niche acronyms, drug names, etc.) are only flagged as "may hit a list match" and not inflated into substantive content violations.

Substantive risks categorized into three tiers (High Risk / Needs Confirmation / Low-Confidence Clues), requiring "clear subject + clear action, result, or promotional intent"; secondary risk categories require independent evidence.

skills/dbs-content-risk-check/SKILL.md

除明确硬红线外，每项风险至少需要「明确对象＋明确动作、结果或传播目的」。

High-risk items are handled with: "Recommended to prioritize deletion, modification, supplementary proof, official filing, or changing publication channels."

Minimal necessary intervention principle prevents over-editing: Strong opinions, emotions, colloquialisms, and hyperbole are not risks by themselves; no edit is suggested unless a specific risk reduction is demonstrated.

skills/dbs-content-risk-check/SKILL.md

强观点、情绪化表达、冲突结构、偏激判断、口语、夸张修辞、隐喻和个人偏好本身不构成风险；

Accompanied by rules: "Do not provide edit suggestions if unable to specify what concrete risk is reduced," and "Do not rewrite entire paragraphs if deleting local words/sentences resolves the issue."

Strict prohibition on fabrication and audit evasion: Never fabricate evidence, credentials, or data; never generate homophones, character splitting, hidden contact info, or comment code words.

skills/dbs-content-risk-check/SKILL.md

也不能生成谐音、拆字、藏联系方式、评论暗号或其他绕审写法。

Defines capability boundaries (refusing to act as an evasion assistant), serving as the skill's clearest self-enforced safety constraint.

Criteria categorized by risk mechanism (content safety, network access, metaphysics, sexual suggestiveness, restricted promotions, advertising/hiring, traffic redirection, pricing claims, medical, visuals/audio/accounts, coordinated marketing, spam).

skills/dbs-content-risk-check/SKILL.md

### 网络访问与海外账号

Example: Merely mentioning ChatGPT or overseas accounts = list-based algorithmic signal; "providing specific VPN tools, download links, setup guides, nodes, purchasing channels, or methods to bypass network restrictions escalates to high risk."

Links and QR codes are not opened by default: URLs in copy are treated as review targets, while unknown short links and QR codes are labeled "destination unverified."

skills/dbs-content-risk-check/SKILL.md

去向不明的短链或二维码写「去向未检查」，不能自动打开或扫描；

Links are only accessed and scope expanded when the user explicitly requests inspection of a webpage, landing page, or video.

Three-line video overlay disclaimers are conditionally generated (not default disclaimers), only adding truthful contextual boundaries in plain text without generating images.

skills/dbs-content-risk-check/SKILL.md

只提供文字，不生成图片，也不调用生图或封面工具。

Example outputs include "Personal experience sharing / Does not guarantee results / Please evaluate based on individual circumstances," stipulating that "'No recruitment fees charged' can only be used if the business truly charges nothing."

Platform rules require timeliness and sourcing discipline: Consult official rules only when conclusions depend on platform policies, recording title, update date, subject, and direct links without inventing thresholds.

skills/dbs-content-risk-check/SKILL.md

只有结论依赖平台当前政策，或者用户询问准入、报备、处罚、导流、商业入口、直播、AI 标识等时效性问题时，才查询目标平台当前官方规则。

Accompanied by: "Do not predict violation probabilities with percentages, do not promise algorithmic outcomes," and "Do not write 'safe to post,' 'cannot post,' 'guaranteed pass,' or '100% compliant.'"

Clear boundary disclaimers: Does not provide legal advice, does not replace platform pre-reviews or professional legal counsel, and makes no guarantees regarding appeal outcomes.

skills/dbs-content-risk-check/SKILL.md

不提供法律意见，不替代平台预审、商品资质审核和专业律师判断；

A necessary disclaimer for content compliance skills, defining suitable use cases (self-checks rather than compliance certifications).

2Core Capabilities

01Dual-track evaluation: Algorithmic review risk vs. substantive content risk (separate conclusions)

02Five-surface inspection (text, semantics, visual/audio, publishing context, platform conditions)

03Three-tier algorithmic risk evidence evaluation (explicit trigger, list-based, weak cue)

04Substantive risk tiering (High Risk / Needs Confirmation / Low-Confidence Clues) and hard red-line list

05Minimal edit action library (add conditions, remove unproven claims, adopt approved phrasing, switch to official channels, add disclosures, localized deletions/revisions)

06Risk mechanism classification check (12 categories, including industry-specific restricted promotional criteria)

07Three-line video overlay text (conditionally triggered, text only)

08Standardized compact report template + zero-issue conclusion copy

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; no script execution or external impact.

Audit Notes: The evaluation target is the skill's own directory (dbs-content-risk-check/SKILL.md + agents/openai.yaml): zero scripts, zero network endpoints, zero credential reads, zero disk writes (dangerous token scan returned ZERO HITS), with all actions occurring strictly on the model output side. Key considerations are conclusion timeliness (dependence on platform policies) and risk of missed detections; platform rule lookups are performed by host search tools only when conditionally triggered, while the skill itself contains zero endpoints. Thus, following the standard of 'evaluating only the skill's own directory,' it remains Green. Host tools and external platform actions are not factored into the rating tier.

5Second-Pass Verification

[ok] 纯 prompt、无脚本、零端点 — 目录仅 SKILL.md(20766B) + agents/openai.yaml(416B)；危险 token 扫描（含 https?://）ZERO HITS —— 全文不出现任何平台域名/API，仅出现平台名称（闲鱼、淘宝、拼多多等）作为名单型信号示例。
[ok] 双轨判定与「不得互相顶替」 — 核心原则两条 +「机器误伤只能写成「可能触发机器审核」，不能补造合理违规原因。」逐字命中。
[ok] 五表面与三级证据强度 — 「### 第二步：检查五个表面」表格 5 行、「### 第三步」三级信号、「只有弱词面线索时保持沉默。」均在原文。
[ok] 反绕审与禁止编造 — 「严禁编造证据、资质、数据、评价、使用经历和活动期限。」与「也不能生成谐音、拆字、藏联系方式、评论暗号或其他绕审写法。」逐字命中。
[ok] 不打开链接/二维码 — 「链接属于待发布文案时，把链接本身作为审核对象，默认不打开网页；」与「去向不明的短链或二维码写「去向未检查」，不能自动打开或扫描；」均在原文。
[ok] 平台规则查询的触发条件与来源纪律 — 条件句与来源句逐字命中；据此判定「外部网络访问由宿主检索工具在条件触发时执行」，不是 skill 自身端点，故不升档（判级只看 skill 自身目录）。
[ok] description 与内容相符（无夸大） — description 声称覆盖标题/正文/图片文字/字幕/口播/评论/账号资料/视频画面，对应五表面 + 12 类机制；「需要明确视频语境边界时额外生成三行贴片」对应「## 视频三行贴片」条件节，未出现「保证通过」类越界承诺。
[ok] 输出层无绝对结论 — 「不写「能发」「不能发」「保证通过」或「100% 合规」。」与「不使用百分比预测违规概率」均在原文。

6Verdict & Scenarios

Dual-track evaluation avoids 'flagging violations via simple word lists': Algorithmic signals and substantive issues are separated and forbidden from serving as each other's rationale.

Minimal necessary intervention: Must articulate what specific risk an edit reduces, preserving strong stances and viral hooks.

Explicitly refuses evasion techniques: Rejects homophones, character splitting, code words, and obfuscated contact info.

Transparent uncertainty: Tiered outputs with no promises, no percentages, and no claims of 'guaranteed pass.'

Restrained operational attack surface: Does not open links or scan QR codes by default; unverified destinations are explicitly labeled 'destination unverified.'

Best for: Ideal for pre-publishing self-checks: short video, Xiaohongshu/RED, WeChat Official Account content, comments, profile bios, and sponsored/e-commerce promotional copy. Especially suited for creators who fear shadowbanning or reach throttling but refuse to dilute their sharp perspectives—providing exact locations, potential consequences, minimal edits, and clear guidance on what can be kept; generates practical, accurate three-line text overlays when video contextual boundaries are needed.

Not for: Not suitable as formal compliance certification or legal advice (explicitly does not replace platform pre-reviews or legal counsel); not suitable for requests seeking complete rewrites or full blocklists (focuses on risk localization and minimal edits without providing raw word banks); not suitable when internal platform risk control metrics, throttling percentages, or appeal success rates are needed (such thresholds are never fabricated); not suitable for generating evasion tactics (homophones, code words, etc.), which this skill strictly rejects.

Install Agent-Ready

① Verified Mirror 2026-09-15

Option A · Direct DownloadDownload dbs-content-risk-check.tar.gz

sha256: fc45138e77038aaa…

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.