home / skills / hexbee / hello-skills / claim-ledger

claim-ledger skill

/skills/claim-ledger

This skill transforms research claims into a verifiable Claim-Evidence-Boundary ledger, ensuring evidence, counterarguments, and publishability are established

npx playbooks add skill hexbee/hello-skills --skill claim-ledger

Review the files below or copy the command above to add this skill to your agents.

Files (5)
SKILL.md
3.5 KB
---
name: claim-ledger
description: 将研究与写作中的主张转成“Claim-Evidence-Boundary”可核验账本。用于在产出前强制补齐证据、反证、边界与可发布性判断,防止无依据断言并提升复用性。
---

# Claim Ledger

## 何时使用

当用户出现以下意图时使用本技能:

- “先做研究框架/证据链,再写内容”
- “判断某观点是否可发、是否站得住”
- “把资料堆整理成可发布结论”
- “让 AI 先出 Ledger 再出线程/长文”

## 核心定义

Claim Ledger 把每条观点拆成三部分:

- `Claim`:要让读者相信的主张
- `Evidence`:支持该主张的证据
- `Boundary`:成立条件与失效条件

目标:让每个结论都可追溯、可反驳、可复核。

## 最小字段(必填 9 项)

每条 Claim 一行,必须包含:

1. `Claim ID`:`C01/C02/...`
2. `Claim`:一句可检验陈述
3. `Type`:`事实/解释/预测/建议`
4. `Confidence`:`高/中/低`
5. `Evidence`:`E01,E02...`(至少 2 条)
6. `Source Grade`:`S/A/B/⚠️`
7. `Counter`:至少 1 条反证/反例;无则写“未找到,风险=高”
8. `Boundary`:成立条件 + 失效条件
9. `Publishability`:`可发/需改写/不可发`

## Evidence Card 规范

每条证据必须是可复核卡片,而非泛化描述。字段:

- `Source`:标题 + 作者/机构 + 日期
- `Link`:URL 或文档定位
- `Quote/Data`:关键原话或数字
- `Supports`:支持 Claim 的哪一部分
- `Reliability`:`S/A/B/⚠️` + 评级理由
- `Freshness`:`是/否`(是否过时)

经验规则:

- 事实型 Claim:至少 1 条 `S` 或 `A`
- 解释型 Claim:至少 2 条跨来源交叉
- 预测/建议型 Claim:必须有历史对比或失败条件,否则降为低置信度

## 六步流程(从资料堆到可发布)

1. 先写 5 条候选 Claim(先定义要证明什么)
2. 给每条 Claim 标注类型(事实/解释/预测/建议)
3. 每条补 2-4 条证据(优先一手源)
4. 强制补反证/反例(没有也要显式记录)
5. 写边界(成立条件与失效条件)
6. 打置信度和可发布性(必要时降调措辞)

## 质量闸门(必须执行)

以下任一不满足,则禁止进入内容生成:

- 任一 Claim 证据少于 2 条
- 任一 Claim 缺少 Counter
- 任一 Claim 缺少 Boundary
- `预测/建议` Claim 无失败条件

处理规则:

- 证据不足:降级为“猜测(低置信度)”
- 语气要求:不得使用确定性措辞

## 与内容生产的映射

### 线程映射(推荐)

- `3 个 Claim × 4 条推 = 12 条线程`
- 每组 4 条顺序:
  1. Claim
  2. Evidence
  3. Boundary
  4. Action

### 长文映射(推荐)

每个 Claim 作为一节:

- 结论 -> 证据 -> 反方风险 -> 边界 -> 小结/清单

## 默认输出格式

1. `Claim Ledger` 表(至少 3 条 Claim)
2. `Evidence Cards`(与 Ledger 的 E# 对齐)
3. `Gate Check`(逐条列出通过/失败项)
4. 若 Gate 全通过,才生成目标内容(线程/长文/脚本)

## 执行指令模板

```text
先输出 Claim Ledger(至少 3 条 Claim,每条 >=2 条证据 + >=1 条反证 + 边界)。
未满足质量闸门前,不允许生成正文。
若证据不足,必须降级为“猜测(低置信度)”,并避免确定性语气。
```

## 参考模板与示例

需要直接套用表格和卡片时,读取:

- `references/templates.md`

需要自动校验 Ledger 质量闸门时,运行:

- `scripts/validate_ledger.py --file <ledger.md>`

Overview

This skill structures research and writing into a verifiable "Claim–Evidence–Boundary" ledger to prevent unsupported assertions and improve reuse. It forces users or agents to supply claims, multiple evidence cards, counter-evidence, and explicit boundaries before producing publishable content. The skill automates gate checks and maps validated ledgers to threads or long-form sections.

How this skill works

The skill ingests candidate claims and enforces a six-step workflow: propose claims, tag types, attach 2–4 evidence cards per claim, record at least one counterexample, define boundaries, and assign confidence and publishability. It validates each claim against quality gates (e.g., minimum evidence, counter, boundary) and outputs a Claim Ledger table, aligned Evidence Cards, and a Gate Check report. Only when all gates pass does the skill allow generation of threads, long-form content, or action lists.

When to use it

  • Before writing: to force a research-first, evidence-first workflow
  • When judging whether a viewpoint is publishable or defensible
  • When organizing source material into traceable conclusions
  • When you want AI to produce a validated ledger first, then generate content
  • When preparing content that must be auditable or rebuttable

Best practices

  • Start with 5 candidate claims and then narrow to the strongest 3+ for the ledger
  • Use at least one first-party or high-grade (S/A) source for factual claims
  • Always add an explicit counterexample; if none found, mark risk high
  • Write boundaries as both conditions of validity and failure modes
  • Avoid definitive language for low-confidence or predictive claims

Example use cases

  • Create a 3–5 claim ledger for an investigative thread before drafting tweets
  • Validate whether a policy recommendation is publishable and under what conditions
  • Turn a literature review into Evidence Cards that map to specific claims
  • Generate long-form article sections where each section is a validated claim
  • Run an editorial gate check to decide if a draft should be published or rewritten

FAQ

What happens if a claim lacks two evidence items?

The skill downgrades the claim to low confidence (guess) and blocks content generation until evidence is added or the claim is rephrased.

How are evidence reliability grades used?

Grades (S/A/B/⚠️) guide minimum requirements: factual claims need at least one S/A; grades must include rationale and freshness to assess trustworthiness.

Can I bypass the gates for speed?

No. The quality gates are enforced to prevent unsupported publication. You can mark a claim as low confidence to continue, but outputs must avoid definitive tone.