Agent skill
claim-ledger
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。
Install this agent skill to your Project
npx add-skill https://github.com/hexbee/hello-skills/tree/main/skills/claim-ledger
SKILL.md
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:
Claim ID:C01/C02/...Claim:一句可检验陈述Type:事实/解释/预测/建议Confidence:高/中/低Evidence:E01,E02...(至少 2 条)Source Grade:S/A/B/⚠️Counter:至少 1 条反证/反例;无则写“未找到,风险=高”Boundary:成立条件 + 失效条件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)
- 先写 5 条候选 Claim(先定义要证明什么)
- 给每条 Claim 标注类型(事实/解释/预测/建议)
- 每条补 2-4 条证据(优先一手源)
- 强制补反证/反例(没有也要显式记录)
- 写边界(成立条件与失效条件)
- 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:
- Claim
- Evidence
- Boundary
- 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
Claim Ledger表(至少 3 条 Claim)Evidence Cards(与 Ledger 的 E# 对齐)Gate Check(逐条列出通过/失败项)- 若 Gate 全通过,才生成目标内容(线程/长文/脚本)
执行指令模板
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>
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
analogy-commentary
Turn source content into a familiar analogy framework, then produce concise commentary angles and a short take. Use when the user asks whether a post, article, speech, or idea is “like” a known story, theory, faction, historical pattern, or strategic lens.
deep-productivity
Master deep work productivity through the three types of work framework (Building, Maintenance, Recovery). Use when user needs to: (1) Build a sustainable deep work routine with just 1 hour/day, (2) Create vision/anti-vision for life direction, (3) Structure goals using the 10-year → 1-year → 1-month → 1-week hierarchy, (4) Apply project-based learning to bridge skill gaps, (5) Identify lever-moving tasks that actually progress goals, (6) Balance focus work with necessary recovery for creativity.
multi-agent-systems
Design and implement multi-agent LLM architectures using the orchestrator-subagent pattern. Use when: (1) Deciding whether to use multi-agent vs single-agent systems, (2) Implementing context isolation for high-volume operations, (3) Parallelizing independent research tasks, (4) Creating specialized agents with focused tool sets, (5) Building verification subagents for quality assurance, or (6) Analyzing context-centric decomposition boundaries.
agent-first-product-strategy
Reframe AI product and SaaS strategy from human-user assumptions to agent-first execution. Use when redefining product positioning, success metrics, API/docs priorities, go-to-market, or roadmap decisions for an AI-native market where agents are primary software users.
debog-yourself
Help users identify and escape psychological traps that are holding them back. Use when user feels stuck, unable to progress, facing a deadlock, or experiencing decision paralysis. Provide diagnostic frameworks and specific strategies to help users understand which trap pattern they're in and find a path forward.
china-model-selection-guide
China model selection and task-routing guide for Doubao-Seed-2.0-Code, GLM-5, MiniMax-M2.5, and Kimi-K2.5. Use when users need to choose the best-fit model by input type, task complexity, engineering constraints, and delivery goals, including staged multi-model workflows.
Didn't find tool you were looking for?