Token导航 LogoToken导航TokenDH.com
研究检索只读github未标认证来源可访问clear审计通过

fluff-detector绒毛探测器

Agent Skill

fluff-detector 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

1,140

周安装

48

GitHub Stars

37

下载量

399
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

3

许可证

MIT

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:fluff-detector(绒毛探测器)
来源仓库:https://github.com/majesticlabs-dev/majestic-marketplace
仓库路径:skills/fluff-detector
安装命令:
npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill fluff-detector
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill fluff-detector

简介

fluff-detector 用于查找、检索和筛选相关信息。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词、任务场景或来源线索快速定位候选结果。
  • 通过 npx skills add 命令从指定 GitHub 仓库安装并使用。
  • 安装前需确认权限范围、维护状态,注意是否触发联网、命令执行或文件读写。
  • 建议结合原始 README 和仓库内容核验具体用法和功能边界。

SKILL.md

Fluff Detector

Purpose

Verify that agents, commands, and skills are written for LLM consumption, not human readers. Every token should improve execution, not provide comfort or credibility.

When to Use

  • Creating new agents, commands, or skills
  • Reviewing PRs that add/modify LLM artifacts
  • Quality gate before merge
  • Auditing existing artifacts for token waste

Fluff Categories

1. Attribution Signals

Content that establishes credibility for humans but has no LLM value.

PatternExampleProblem
"Inspired by X""Inspired by DHH's approach"LLM can't retrieve by author
"Based on X's methodology""Based on Martin Fowler's patterns"Pure credibility signal
"According to X""According to industry experts"Appeal to authority
"As recommended by""As recommended by Google"No execution value

Exception: Names that define a style (e.g., "DHH-style Rails") are functional, not attribution.

Detection pattern:

(inspired by|based on|according to|as recommended by|following|using|applying)\s+[A-Z][a-z]+('s)?

2. Decorative Quotes

Inspirational or philosophical quotes that add no functional value.

PatternExample
Epigraph quotes"The best code is no code" - Someone
Philosophy"In the spirit of simplicity..."
Motivation"Great engineers always..."

Detection pattern:

^>\s*[""].*[""].*[-—]\s*[A-Z]
[""][^""]{20,}[""][\s]*[-—]

3. Persona Fluff

Role descriptions that tell the LLM what it "is" instead of what to do.

PatternExampleBetter
Identity statements"You are a senior engineer"(delete)
Capability claims"You have deep expertise in..."(delete)
Personality traits"You are helpful and thorough"(delete)
Self-reference"As an AI assistant..."(delete)

Detection pattern:

^(You are|As an?|I am|Acting as)\s+(a |an |the )?[a-z]+(ly)?\s+(and\s+[a-z]+\s+)?

4. Redundant Explanations

Explaining what LLMs already know or can infer.

PatternExample
Tool explanations"The Read tool allows you to read files"
Obvious context"This agent is used for..."
Meta-commentary"The following section describes..."
Format explanations"The output should be formatted as..." (when showing format)

Detection pattern:

(this (agent|skill|command) (is used|helps|allows|enables))
(the following (section|content|instructions))
(as (shown|described|mentioned) (above|below|previously))

5. Marketing Language

Promotional or sales-oriented language inappropriate for LLM instructions.

PatternExample
Superlatives"Best-in-class", "world-class", "cutting-edge"
Promises"Ensures perfect results", "guarantees success"
Buzzwords"Revolutionary", "game-changing", "next-gen"
Vague claims"Highly efficient", "extremely powerful"

Detection pattern:

(best-in-class|world-class|cutting-edge|state-of-the-art|revolutionary|game-chang|next-gen|highly efficient|extremely powerful|ensures? (perfect|optimal)|guarantee)

6. Hedging Language

Uncertainty markers that add no value.

PatternExampleBetter
"Try to""Try to follow best practices""Follow best practices"
"Attempt to""Attempt to minimize errors""Minimize errors"
"Consider""Consider using X""Use X when Y"
"Maybe""Maybe add validation""Add validation if X"

Detection pattern:

\b(try to|attempt to|consider\s+(using|adding|implementing)|you (might|may|could) want to|perhaps|maybe)\b

7. Filler Phrases

Zero-information phrases that pad content.

PatternDelete
"It's important to note that"
"Please keep in mind that"
"It goes without saying"
"Needless to say"
"In order to"Use "to"
"Due to the fact that"Use "because"
"At this point in time"Use "now"

Detection pattern:

(it('s| is) (important|worth|crucial) to (note|mention|remember)|please (keep in mind|note|remember)|goes without saying|needless to say|in order to|due to the fact|at this point in time|as a matter of fact|for all intents and purposes)

Usage

Validate Single File

./scripts/detect-fluff.sh path/to/file.md

Validate All Agents

for agent in plugins/*/agents/**/*.md; do
  ./scripts/detect-fluff.sh "$agent"
done

CI Integration

- name: Detect Fluff
  run: |
    find plugins -name "*.md" -path "*/agents/*" -o -name "*.md" -path "*/skills/*" -o -name "*.md" -path "*/commands/*" | while read f; do
      .claude/skills/fluff-detector/scripts/detect-fluff.sh "$f" || exit 1
    done

Output Format

Analyzing: plugins/majestic-engineer/agents/qa/test-reviewer.md

[WARN] Line 12: Attribution signal
  > "Based on Kent Beck's TDD principles"
  Suggestion: Remove attribution or rephrase as pattern name

[WARN] Line 45: Persona fluff
  > "You are an expert test reviewer with deep knowledge"
  Suggestion: Delete - LLM doesn't need identity statements

[WARN] Line 78: Filler phrase
  > "It's important to note that tests should be isolated"
  Suggestion: "Tests should be isolated"

Summary: 3 fluff instances found

Severity Levels

LevelActionExample
ERRORMust fixPersona statements, marketing language
WARNShould fixAttribution, hedging, fillers
INFOConsiderRedundant explanations (may be intentional)

False Positive Handling

Some patterns have legitimate uses:

PatternLegitimate Use
"Based on""Based on user input" (not attribution)
"Consider"In decision trees with conditions
NamesWhen defining a style (DHH, Warren Buffett)

The script uses context analysis to reduce false positives.

Error Codes

CodeMeaning
0No fluff detected
1Warnings found (non-blocking)
2Errors found (blocking)

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

04

需要参考平台分布和安装热度时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Antigravity

29.88%
按下载量换算119

OpenCode

22.26%
按下载量换算89

Gemini CLI

18.28%
按下载量换算73

Claude Code

12.64%
按下载量换算50

trae

7.07%
按下载量换算28

windsurf

3.91%
按下载量换算16

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。

来源信息

继续浏览同类 Skills