dbs-ai-check (dbskill): a skill that marks the AI fingerprints in your writing
dbs-ai-check makes your agent scan your text for 22 AI-writing traits and report each one with a quote, and it does not rewrite unless you ask.
- Installs
- 27.6K
- on skills.sh
- Stars
- 11K
- on GitHub
- Files
- 1
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
You paste a draft and ask for an AI-trace check. The agent returns a report in text order: each hit quotes the original passage, says what is wrong in one or two sentences, and gives the trait number and a severity of strong, medium or weak. It ends with a one-line summary and its guess at the genre. If the text looks fine, it says so.
The skill is built to diagnose, not to disguise. If you say "I want to edit it", it asks one question per hit, from the most severe, about what you meant. It gives edit directions, and gives a concrete rewrite only when you say "help me edit". It replies in your language. Several traits describe Chinese writing habits, such as translation-style phrasing, so it suits Chinese text best.
Good for
- Writers and creators who want to see what reads as machine-written
- Editors checking Chinese articles, scripts and posts before publishing
- Anyone who wants questions about their own style, not an automatic rewrite
Install dbs-ai-check
Claude Code
- 1
Copy this prompt.
Install the dbs-ai-check skill in Claude Code. Run this command: npx skills add https://github.com/dontbesilent2025/dbskill --skill dbs-ai-check Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js.
From its listing on skills.sh. Source.
What's in the skill
When it loads
扫描文案中的 AI 写作特征并输出检测报告,默认只诊断不改写;用户看完报告说「我想改」后,逐条追问写法背后的意图并给出修改方向。用户要求检查 AI 味、AI 痕迹或机器化表达时使用。
Sections in SKILL.md
- 核心哲学
- 原则 1:AI 味的本质是「太完美」
- 原则 2:去 AI 味 ≠ 好内容
- 原则 3:改写必须基于用户自己的偏好
- 识别模式(默认)
- 检测报告格式
- 报告规则
- 严重度分级
- 改写引导模式(用户说「我想改」后触发)
- 核心逻辑
- 流程
- 追问映射表
- 误伤警告
- 体裁识别
- 追问的自检规则
- 特别警告(遇到就直说)
Key facts
- Author
- dontbesilent (dbskill)
- Installs on skills.sh
- 27,565 (as of Nov 2, 2026)
- GitHub stars
- 11,039 (shared repo, Nov 2, 2026)
- Last repo commit
- Oct 28, 2026
- skills.sh audits
- Gen Agent Trust Hub: pass. Socket: pass. Snyk: pass.
What it includes
Scans against 22 traits and writes a report
Reports each hit in text order with a quote, the problem, trait number and severity. Does not edit.
Read ·
Detection mode (default)Strong, medium, weak
Strong: almost only AI does this. Medium: AI does it often, humans can too. Weak: needs genre and context.
Severity levelsAdjusts what counts as a hit by genre
Short-video script, long article, social post, academic or formal text. It states its genre guess in the report.
Genre recognitionLists habits humans use too
For example the contrast pattern counts only at 3 or more in 800 characters, and punchline endings apply to long articles only.
False-positive warningsOne question per hit, most severe first
Asks about the intent behind the trait, then gives a direction. A concrete rewrite comes only after you say "help me edit".
Rewrite coaching ("I want to edit")Pushes back on "remove the AI flavor" requests
Says removing it does not make the content better and asks who you want to sound like before editing.
Special warningsMentions /dbs for other tasks
When the request is out of scope, it says what ability is needed and mentions /dbs if installed.
Follow-up routing
From the official docs
Questions
What does dbs-ai-check do?
It scans a text for 22 traits that read as AI-written, quotes each hit with a severity, and ends with a summary and a genre guess. It does not rewrite by default.
How do I install it in Claude Code?
Use the install command in the box above, which installs this skill alone. The dbskill README also lists an install-everything command for all 34 skills across every agent, and a Claude Code plugin.
Does it work on English text?
It replies in English if you write in English, per SKILL.md. Several of its 22 traits target Chinese writing, such as translation-style phrasing, so expect the best results on Chinese text.
Will it rewrite my text?
Not by default. SKILL.md says it will not hand over an unprompted rewrite or one meant to disguise the text. After the report, say "I want to edit" and it asks questions first.
Is it safe?
SKILL.md bundles no scripts and names no network calls, and the skill works on the text you paste. The text goes to your assistant's model like any other message.
Do I need the rest of dbskill?
No. This skill works on its own. SKILL.md mentions the /dbs router only as an optional next step, and the README says dbs-ai-check can be paired with dbs-plain-writing.
What is the license?
The LICENSE file and README badge say CC BY-NC 4.0, which bars commercial use. GitHub's API shows no recognized license, so read the LICENSE file before using it at work.
How is it different from a humanizer skill?
It is built to diagnose, not to disguise. SKILL.md says it opposes removing AI traces to pass as human, and it asks about your intent before any edit.
Information
updated Nov 2, 2026
- Repository
- dontbesilent2025/dbskill
- Folder
skills/dbs-ai-check
Sources
Docs, issue trackers and discussions this page is built from.
Show all 3Hide
- SKILL.md · github.com
- dbskill README · github.com
- dbskill LICENSE · github.com