Token导航 LogoToken导航

Agent Skill · 研究检索

quality-loop(质量循环)

quality-loop 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。

quality-loop 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为只读;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 55,679。

总安装 470周安装 19GitHub Stars 4下载 147核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:quality-loop(质量循环)
来源仓库:https://github.com/cdeistopened/opened-vault
仓库路径:skills/quality-loop
安装命令:
npx skills add https://github.com/cdeistopened/opened-vault --skill quality-loop
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。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/cdeistopened/opened-vault --skill quality-loop

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

quality-loop 用于查找、检索和筛选相关信息,适合根据关键词或任务线索快速定位候选结果。适用于研究、调研或信息收集类场景,可提升外部资源发现效率。通过指定查询条件调用搜索接口,返回结构化候选列表供进一步处理。使用前建议确认是否允许联网及数据抓取边界,避免违反平台策略。

适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

来源仓库内路径:skills/quality-loop

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/cdeistopened/opened-vault --skill quality-loop;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

quality-loop 怎么安装?

在终端执行 npx skills add https://github.com/cdeistopened/opened-vault --skill quality-loop。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

quality-loop 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

quality-loop 需要什么权限?

权限级别是只读。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。

quality-loop 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

quality-loop 用的人多吗?

公开数据是累计安装 470 次,近一周安装 19 次,下载 147 次,GitHub 4 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

quality-loop 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 只想先试试效果、不希望它改动本地文件的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
quality-loop

质量循环 · 本页

4704只读未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Quality Loop

Iterative drafting system with backpressure from judge personas. Creates high-quality content through systematic source gathering, hook-first drafting, and multi-pass quality gates.

When to Use This Skill

UNIVERSAL QUALITY GATES: This skill triggers for ALL public-facing content.

Trigger Points by Content Type

Content TypeTrigger PointMode
NewsletterAfter draft, before sendFull 5-judge
Deep Dive/ArticleBefore Webflow publishFull 5-judge
Podcast blogBefore publishFull 5-judge
LinkedIn postAfter draft, before scheduleLite 3-judge
X postAfter draft, before scheduleLite 3-judge
InstagramAfter draft + visualLite 3-judge
FacebookAfter draft, before scheduleLite 3-judge

Full 5-Judge Mode

For long-form content (articles, newsletters, blog posts):

  • Human Detector (BLOCKING)
  • Accuracy Checker (BLOCKING)
  • OpenEd Voice (BLOCKING)
  • Reader Advocate (BLOCKING)
  • SEO Advisor (ADVISORY)

Lite 3-Judge Mode

For social posts (faster, focused on voice and AI tells):

  • AI-Tell Judge (BLOCKING) - Hard blocks only
  • Voice Judge (BLOCKING) - Brand alignment
  • Platform Judge (ADVISORY) - Platform-specific optimization

See "Lite Quality Loop for Social" section below.


The Quality Loop Process

SOURCES → HOOK → DRAFT → JUDGES → ITERATE
    ↑                         ↓
    └─────── (if blocked) ────┘

Phase 1: Source Compilation

Before drafting, gather all relevant sources into a single compiled file.

Source Search Order:

  1. Proprietary content first - Search OpenEd podcast transcripts, newsletters, Slack
  2. Content database - Grep Content/ for related themes (not just exact terms)
  3. External research - Web search for biography, facts, external validation

When no direct coverage exists: Expand the search to related themes. For example, if writing about "Daniel Greenberg":

  • Search for "democratic education"
  • Search for "self-directed"
  • Search for "trust children"
  • Search for related thinkers who discuss him (Peter Gray)

Source File Format:

# Source: [Topic] Compiled

## Biographical Facts
- Key dates, roles, locations
- Verified from multiple sources

## OpenEd Proprietary Content
**From Podcast [Episode]:**
> "Direct quote..."

**From Newsletter:**
> "Direct quote..."

## Key Themes
- Theme 1 with supporting evidence
- Theme 2 with supporting evidence

## SEO Notes
- Primary keyword: [keyword] ([volume]/mo)
- Secondary keywords: [list]

## Sources for Verification
- [URL 1]
- [URL 2]

Phase 2: Hook First

Never draft without user approval of the hook angle. Propose 4-6 hook options with clear differentiation.

Hook Proposal Format:

#HookOpening LineWhy It Works
1[Name]"[First sentence]"[Reasoning]
2[Name]"[First sentence]"[Reasoning]
............

Good hooks:

  • Start with a specific moment, fact, or tension
  • Create curiosity without clickbait
  • Connect to something the reader cares about
  • Differentiate from what's already ranking

Bad hooks:

  • Generic "In today's world..." openings
  • Definition-first approaches ("X is defined as...")
  • Vague statements that could apply to anyone

After user selects a hook, proceed to drafting.

For series content: When writing multiple pieces on similar topics (e.g., profiles of thinkers), ensure each piece has a structurally different approach to avoid sameness. Examples:

  • Profile A: Personal drama narrative
  • Profile B: Scientific/research angle
  • Profile C: Evolution story (phases of career)
  • Profile D: Institutional critique
  • Profile E: Evidence/longevity angle

Track the structural approach used for each piece to ensure variety.


Phase 3: Draft

Create the full draft following OpenEd style guidelines.

Draft Structure:

# Draft v[N]: [Title]

**Date:** [YYYY-MM-DD]
**Type:** [New / Enhancement]
**Target:** ~[X] words

---

## META ELEMENTS

**Title Options:**
1. [Option 1] ([char count])
2. [Option 2] ([char count])
...

**Meta Description (155 chars):** [Description]

**URL:** /blog/[slug]

---

## ARTICLE

[Full article content]

---

## NOTES FOR JUDGES

**Word count:** ~[X] words
**Internal links:** [N]
**External links:** [N]
**Target keywords:** [list]
**OpenEd connection:** [what makes this ours]
**Unique angle:** [what differentiates from competitors]

Phase 4: Five Judges

Run every draft through all five judges in order. If ANY blocking judge fails, fix and re-run that judge before proceeding.

Full judge references: See references/ folder for expanded criteria.

Judge 1: Human Detector (BLOCKING)

*See: references/human-detector.md*

Scans for AI tells. Zero tolerance for:

  • Correlative constructions ("X isn't just Y - it's Z")
  • Dramatic contrast reveals ("Not X. Y.")
  • AI vocabulary (delve, comprehensive, crucial, landscape, journey, tapestry, myriad)
  • Staccato patterns ("No fluff. No filler. Just results.")
  • Triple Threat Syndrome (forced three-adjective stacks)

VERDICT: PASS only if zero AI tells found.


Judge 2: Accuracy Checker (BLOCKING)

*See: references/accuracy-checker.md*

Verifies all factual claims against compiled sources:

  • Dates, names, quotes must match sources exactly
  • Statistics need citations
  • Timeline events in correct order
  • No unverifiable claims presented as fact

VERDICT: PASS only if all facts verified.


Judge 3: OpenEd Voice (BLOCKING)

*See: references/opened-voice.md*

Core stance: Pro-child, not anti-school.

  • Describes, doesn't prescribe
  • Practical takeaways present
  • 3+ internal links
  • Uses proprietary OpenEd content

VERDICT: PASS only if aligned with stance.


Judge 4: Reader Advocate (BLOCKING)

*See: references/reader-advocate.md*

Assesses engagement:

  • Hook creates curiosity (not definitions)
  • Logical section flow
  • Scannable structure
  • Appropriate length
  • Strong ending

VERDICT: PASS only if engaging throughout.


Judge 5: SEO Advisor (ADVISORY)

*See: references/seo-advisor.md*

Evaluates search optimization. Does not block.

  • Keyword in title, first 100 words, H2s
  • Meta elements optimized
  • 5+ internal links, 2-3 external
  • Featured snippet opportunities

VERDICT: Advisory feedback only.


Phase 5: Iterate

If any blocking judge fails:

  1. Make the specific fixes identified
  2. Re-run ONLY the failed judge
  3. If pass, continue to next judge
  4. If fail again, make additional fixes and repeat

After all judges pass:

  • Update status in tracking document
  • Move to next piece or finalize for publication

Folder Structure

The folder structure depends on the project type. Organize sources and drafts logically.

Example for profile series:

project/
├── sources/
│   ├── person-a/compiled-sources.md
│   └── person-b/compiled-sources.md
├── drafts/
│   ├── person-a/v1.md
│   └── person-b/v1.md
└── TRACKING.md

Example for guide project:

project/
├── sources/compiled-sources.md
├── drafts/
│   ├── v1.md
│   ├── v2.md
│   └── final.md
└── TRACKING.md

Tracking Progress

Maintain a tracking document showing status of each piece.

Tracking Table Format:

ItemSourcesHookDraftHDACOVRASEOStatus
[Name]DONE[Hook name]v1PASSPASSPASSPASSPASSCOMPLETE
[Name]DONEpending------NEEDS HOOK

Legend:

  • HD = Human Detector
  • AC = Accuracy Checker
  • OV = OpenEd Voice
  • RA = Reader Advocate
  • SEO = SEO Advisor

Lite Quality Loop for Social

Faster quality checks for social posts. Run after draft, before scheduling.

Judge 1: AI-Tell Judge (BLOCKING)

Hard blocks - auto-reject if ANY found:

  • Correlative constructions ("X isn't just Y - it's Z")
  • Banned words: delve, comprehensive, crucial, leverage, landscape
  • Setup phrases: "The best part?", "What if I told you", "Here's the thing"
  • Staccato patterns: "No fluff. No filler. Just results."
  • Em dashes without spaces (use " - " not "—")

VERDICT: PASS only if zero hard blocks found.

Judge 2: Voice Judge (BLOCKING)

Checks:

  • Sounds like OpenEd/brand account (not personal blog)
  • Aligns with Open Education values (parent empowerment, learner agency)
  • Appropriate tone for platform
  • No preachy or prescriptive language

VERDICT: PASS only if brand-aligned.

Judge 3: Platform Judge (ADVISORY)

Platform-specific checks:

PlatformChecklist
LinkedIn200-500 words, links in comments, 3-5 hashtags, hook in first 2 lines
X70-100 chars optimal, 1-2 hashtags, retweet-worthy
InstagramVisual-first, caption supports, 5-10 hashtags, first 150 chars hook
FacebookNo external links, no hashtags, ends with question/engagement prompt

VERDICT: Advisory feedback - does not block.

Lite Loop Process

DRAFT → AI-Tell Judge → Voice Judge → Platform Judge → SCHEDULE
             ↓ (fail)        ↓ (fail)
           FIX & RETRY     FIX & RETRY

Quick Reference: AI Patterns to Avoid

Correlative Constructions (Most Common Tell)

  • "X isn't just Y - it's Z"
  • "X didn't Y. It Z."
  • "The goal isn't X - it's Y"
  • "It's not about X, it's about Y"

Forbidden Words

delve, comprehensive, crucial, vital, leverage, landscape, navigate, foster, facilitate, realm, paradigm, embark, journey, tapestry, myriad, multifaceted, seamless, cutting-edge

Forbidden Phrases

  • "The best part?..." / "The secret?..."
  • "What if I told you..." / "Here's the thing..."
  • "In today's fast-paced..." / "In the ever-evolving..."
  • "In conclusion" / "In summary"
  • "Let that sink in" / "Now more than ever"

Dramatic Contrast Reveals (Priority #2)

  • "Not on lessons. On fear."
  • "Not the curriculum. The structure."
  • "He didn't teach. He observed."
  • Any "Not X. Y." fragment pattern

Forbidden Patterns

  • Staccato: "No fluff. No filler. Just results."
  • Triple adjectives: "Bold, beautiful, brilliant"
  • Negation structure: "No X. No Y. Just Z."

Formatting Rules

  • Use hyphens with spaces - like this - not em dashes
  • No emojis in body content
  • No bold for emphasis in articles

平台分布

Codex

37.02%
按下载量换算54

Claude

31.78%
按下载量换算47

Cursor

18.79%
按下载量换算28

Gemini CLI

9.48%
按下载量换算14

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills