重要前提
安装AI Skills的关键前提是:必须科学上网,且开启TUN模式,这一点至关重要,直接决定安装能否顺利完成,在此郑重提醒三遍:科学上网,科学上网,科学上网。查看完整安装教程 →
npx skills add https://github.com/jwynia/agent-skills --skill media-meta-analysis综合多个独立媒体分析中的模式和联系,以揭示更深层次的见解、概念网络和涌现的主题。该框架作用于已分析内容的集合,而非单个内容片段。
整体揭示部分无法揭示的内容。 在单个来源中不可见的模式,在集合中变得可见。
在对多个内容片段进行独立提取分析(例如,使用媒体内容提取框架)之后使用。本框架作用于分析的集合,而非原始媒体。
记录集合特征:
识别覆盖范围的偏见或空白
| 分析项 | 追踪内容 |
|---|---|
| 最频繁出现的概念 | 核心主题 |
| 高连接密度 | 枢纽概念 |
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
| 孤立概念 | 孤立想法 |
| 概念集群 | 相关想法组 |
| 术语变体 | 相同想法,不同表述 |
| 随时间演变 | 想法如何发展 |
描绘论证格局:
发现不同来源之间的联系:
映射连接强度和方向性
识别有意义的张力:
| 类型 | 示例 |
|---|---|
| 直接论断矛盾 | 来源 A 说 X,来源 B 说非 X |
| 竞争性解释 | 相同证据,不同结论 |
| 框架不兼容 | 根本方法差异 |
| 价值优先级差异 | 不同的层级结构 |
| 定义不一致 | 相同术语,不同含义 |
| 方法论分歧 | 如何研究该问题 |
注意矛盾是表面的还是根本性的
识别相互支持的元素:
评估强化强度和来源独立性
在单个内容中不突出的模式:
描绘"负空间":
根据重要性和可解决性进行优先级排序
跨来源获得重要性的元素:
| 结构 | 目的 |
|---|---|
| 概念到来源索引 | 查找想法出现的位置 |
| 论断验证路径 | 追踪证据链 |
| 矛盾地图 | 查看来源分歧之处 |
| 证据链 | 遵循证明模式 |
| 框架比较 | 比较不同方法 |
| 问题-答案网络 | 追踪探究路径 |
创建可导航的关系模型:
映射探索路线:
## 集合:[名称]
**来源:** [数量和类型]
**日期范围:** [发布日期]
**分析周期:** [分析时间]
**主要领域:** [主题领域]
**分析目的:** [预期用途]
## [要素类型]:[主题/连接/模式]
**来源:** [贡献来源及位置]
**证据:** [关键支持示例]
**重要性:** [为何重要]
**张力:** [矛盾或复杂性]
**探索向量:** [进一步调查方向]
用于开发新内容:
用于指导调查:
用于构建知识结构:
模式: 包含所有可用来源,而不评估其质量、相关性或冗余性。失败原因: 劣质来源污染综合结果。冗余来源造成虚假共识。不相关来源分散了对重要模式的注意力。修复方法: 明确评估语料库构成。移除低质量来源。根据独立性对来源进行加权。注意"多个来源"实际上可能是同一来源的重复。
模式: 发现的模式仅存在于所选来源中,而非存在于底层现实中。失败原因: 确认偏误影响了你找到的来源。如果你搜索"X 导致 Y",你会找到讨论 X 和 Y 的来源。这并非模式存在的证据。修复方法: 积极寻找证伪来源。注意在预期出现模式的地方其缺失情况。区分"我所有的来源都同意"和"我选择了同意的来源"。
模式: 将矛盾来源综合成一个中间立场——"真相在两者之间"。失败原因: 矛盾通常表明真实的分歧,而非测量误差。中间立场可能无人持有,也缺乏证据支持。修复方法: 明确映射矛盾。理解来源为何不同意。呈现立场的格局,而非人为的共识。
模式: 引用综合结果时,仿佛它是原始证据,丢失了回溯到原始来源的链条。失败原因: 元分析的质量仅与其来源相当。当链条坍塌时,你无法评估可靠性或识别分歧所在。修复方法: 维护来源到论断的索引。始终清楚哪个原始来源支持哪个综合论断。使验证路径明确。
模式: 专注于来源说了什么,而不映射它们没说什么——知识空白和盲点。失败原因: 缺失的内容往往比存在的内容更重要。系统性的空白揭示了偏见、研究不足的领域和机会。修复方法: 明确映射"负空间"。哪些问题没有来源涉及?哪些方法论缺失?哪些视角未被代表?
| 技能 | 提供内容 |
|---|---|
| research | 单个来源发现和查询扩展 |
| claim-investigation | 已验证的单个论断,用于综合 |
| fact-check | 经过质量检查的单个分析 |
| 技能 | 提供内容 |
|---|---|
| research | 识别出的空白,供进一步调查 |
| (content creation) | 综合知识,用于原创工作 |
| (knowledge organization) | 信息架构的结构 |
| 技能 | 关系 |
|---|---|
| research | research 负责寻找来源;meta-analysis 负责综合它们。迭代使用——综合揭示空白,research 负责填补 |
| claim-investigation | claim-investigation 验证单个论断;meta-analysis 追踪论断如何跨来源连接 |
每周安装量
64
代码库
GitHub 星标数
40
首次出现
2026 年 2 月 16 日
安全审计
安装于
codex59
gemini-cli58
github-copilot57
cursor57
opencode57
kimi-cli56
Synthesize patterns and connections across multiple individual media analyses to reveal deeper insights, conceptual networks, and emergent themes. Operates on collections of analyzed content, not individual pieces.
The whole reveals what the parts cannot. Patterns invisible in individual sources become visible across collections.
Use after analyzing multiple pieces with individual extraction (e.g., media content extraction framework). This framework operates on collections of analyses, not raw media.
Document collection characteristics:
Identify biases or gaps in coverage
| Analysis | What to Track |
|---|---|
| Most frequent concepts | Core themes |
| High connection density | Hub concepts |
| Isolated concepts | Orphan ideas |
| Concept clusters | Related idea groups |
| Terminology variations | Same idea, different words |
| Evolution over time | How ideas develop |
Map the argumentation landscape:
Discover connections between disparate sources:
Map connection strength and directionality
Identify meaningful tensions:
| Type | Example |
|---|---|
| Direct claim contradictions | Source A says X, Source B says not-X |
| Competing interpretations | Same evidence, different conclusions |
| Framework incompatibilities | Fundamental approach differences |
| Value priority differences | Different hierarchies |
| Definitional inconsistencies | Same term, different meanings |
| Methodological disagreements | How to study the question |
Note whether contradictions are apparent or fundamental
Identify mutually supporting elements:
Rate reinforcement strength and source independence
Patterns not prominent in individual pieces:
Map the negative space:
Prioritize by significance and addressability
Elements that gain significance across sources:
| Structure | Purpose |
|---|---|
| Concept-to-source index | Find where ideas appear |
| Claim verification pathways | Trace evidence chains |
| Contradiction maps | See where sources disagree |
| Evidence chains | Follow proof patterns |
| Framework comparisons | Compare approaches |
| Question-answer networks | Track inquiry paths |
Create navigable relationship models:
Map exploration routes:
## Collection: [Name]
**Sources:** [Number and types]
**Date Range:** [Publication dates]
**Analysis Period:** [When analyzed]
**Primary Domains:** [Subject areas]
**Analysis Purpose:** [Intended use]
## [Element Type]: [Theme/Connection/Pattern]
**Sources:** [Contributing sources with locations]
**Evidence:** [Key supporting examples]
**Significance:** [Why this matters]
**Tensions:** [Contradictions or complications]
**Exploration Vectors:** [Further investigation directions]
For developing new content:
For guiding investigation:
For structuring knowledge:
Pattern: Including all available sources without assessing their quality, relevance, or redundancy. Why it fails: Bad sources contaminate synthesis. Redundant sources create false consensus. Irrelevant sources distract from patterns that matter. Fix: Assess corpus composition explicitly. Remove low-quality sources. Weight sources by independence. Note when "multiple sources" are actually one source repeated.
Pattern: Finding patterns that exist only in the selection of sources, not in the underlying reality. Why it fails: Confirmation bias shapes what sources you find. If you search for "X causes Y," you'll find sources discussing X and Y. That's not evidence of a pattern. Fix: Actively seek disconfirming sources. Note absence of pattern where expected. Distinguish "all my sources agree" from "I selected sources that agree."
Pattern: Synthesizing contradictory sources into a middle position—"the truth is somewhere between." Why it fails: Contradictions often indicate real disagreement, not measurement error. The middle position may be held by no one and supported by no evidence. Fix: Map contradictions explicitly. Understand why sources disagree. Present the landscape of positions rather than an artificial consensus.
Pattern: Citing a synthesis as if it were primary evidence, losing the chain back to original sources. Why it fails: Meta-analysis is only as good as its sources. When the chain collapses, you can't evaluate reliability or identify where disagreement actually lies. Fix: Maintain source-to-claim indices. Always know which original source supports which synthesis claim. Make verification pathways explicit.
Pattern: Focusing on what sources say without mapping what they don't say—the knowledge gaps and blind spots. Why it fails: What's missing is often more important than what's present. Systematic gaps reveal biases, under-researched areas, and opportunities. Fix: Explicitly map negative space. What questions do no sources address? What methodologies are absent? What perspectives are unrepresented?
| Skill | What it provides |
|---|---|
| research | Individual source discovery and query expansion |
| claim-investigation | Verified individual claims for synthesis |
| fact-check | Quality-checked individual analyses |
| Skill | What this provides |
|---|---|
| research | Identified gaps for further investigation |
| (content creation) | Synthesized knowledge for original work |
| (knowledge organization) | Structure for information architecture |
| Skill | Relationship |
|---|---|
| research | Research finds sources; meta-analysis synthesizes them. Use iteratively—synthesis reveals gaps that research fills |
| claim-investigation | Claim-investigation verifies individual claims; meta-analysis traces how claims connect across sources |
Weekly Installs
64
Repository
GitHub Stars
40
First Seen
Feb 16, 2026
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
codex59
gemini-cli58
github-copilot57
cursor57
opencode57
kimi-cli56
前端代码审计工具 - 自动化检测可访问性、性能、响应式设计、主题化与反模式
49,600 周安装