Skip to content
Back to skills

Hypothesis Synthesis

ASecurity

SOP: 综合所有中间产物,产出最终结构化假设集

  • 417 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added May 27, 2026
researchgo

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned May 27, 2026

npx -y skills add yogsoth-ai/de-anthropocentric-research-engine --skill hypothesis-synthesis --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Hypothesis Synthesis?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Hypothesis Synthesis
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/yogsoth-ai-hypothesis-synthesis/badge)](https://www.skillsdirectory.com/skills/yogsoth-ai-hypothesis-synthesis)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: hypothesis-synthesis
description: "SOP: 综合所有中间产物,产出最终结构化假设集"
version: 1.0.0
category: hypothesis-formation
type: sop
campaign: hypothesis-formulation
input: "所有中间产物(理论、机制、变量、关系、边界条件、可证伪性、竞争假设、对比矩阵)"
output: "最终假设文档(完整 6 组件结构 + 排序 + 关系说明)"
dependencies:
  skills:
    - subagent-spawning
---

# Hypothesis Synthesis
收集并整合所有中间产物,产出完整、去重、排序的最终假设集。

## HARD-GATE
<HARD-GATE>
前置条件(全部满足才能开始):
1. 已有 ≥2 个假设候选(来自上游任意步骤)
2. 每个假设至少有 statement + mechanism + falsification_scenario

不满足 → 停止,返回错误:假设候选不足或缺少必要组件。
</HARD-GATE>

## Pipeline
1. 前置检查:验证所有中间产物完整性
2. 收集全部假设候选:汇总来自 deductive/inductive/abductive 路径的所有假设
3. 去重合并:识别并合并机制相同的假设(保留最完整版本)
4. 完整性检查:确保每个假设包含全部 6 个组件(statement/variables/mechanism/boundary_conditions/falsification/measurability)
5. 排序:按证据支持度 + 可测试性 + 理论重要性综合排序
6. 关系说明:标注假设间关系(互补/竞争/独立/层级)
7. 格式化输出最终假设文档

## Output Format
```json
{
  "summary": {
    "total_hypotheses": 3,
    "primary_hypothesis": "H1",
    "reasoning_path": "deductive | inductive | abductive | mixed"
  },
  "hypotheses": [
    {
      "hypothesis_id": "H1",
      "priority": 1,
      "statement": "If X then Y under conditions Z",
      "variables": {
        "independent": "X",
        "dependent": "Y",
        "mediators": [],
        "moderators": [],
        "controls": []
      },
      "mechanism": "Causal chain explanation",
      "boundary_conditions": {
        "temporal": "...",
        "spatial": "...",
        "population": "...",
        "conditional": [],
        "exclusions": []
      },
      "falsification": {
        "scenario": "...",
        "testability": "high | medium | low"
      },
      "measurability": {
        "iv_measure": "...",
        "dv_measure": "..."
      },
      "evidence_support": "strong | moderate | weak | none",
      "theoretical_basis": "Theory name(s)"
    }
  ],
  "hypothesis_relationships": [
    {
      "pair": "H1 and H2",
      "relationship": "complementary | competing | independent | hierarchical",
      "notes": "..."
    }
  ],
  "recommended_next_step": "Which hypothesis to develop into a research question first, and why"
}
```

Files in this skill

  • SKILL.md2.7 KB
  • prompt.md3.7 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.