evidence-chain-builder

把 AI 论断拆解为原子证据片段，按来源类型（official/paper/data/internal/assertion）标注可独立验证性与可信度，汇总整体证成度并列出未支撑论断。通过命令行运行 Python 脚本，输出结构化 JSON；声明不判定论断真伪，只评估证据质量。

zhaoxinghua09-cell

热度 61.67

最近更新 2026-10-17 07:26:50

分类：知识依赖：代码执行应用场景：研究 / 报告撰写

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add evidence-chain-builder

Metadata

name: evidence-chain-builder
slug: evidence-chain-builder
display_name: AI 论断有证证据链（LGD-II 有证）
display_name_en: AI Claim Evidence-Chain Builder (LGD-II Evidenced)
description: 当用户要『验证 AI 论断 / 防幻觉 / 给结论找证据 / 判断信源可信度』时用。把论断拆成可验证证据链：每条证据标来源类型(official/paper/data/internal/assertion)、是否可独立验证、可信度，输出『证成度』与未支撑论断清单。对齐 EIFP 不编造原教旨——本工具不判定论断真假，只评估证据质量；缺可验证证据的论断明确标『勿作结论』。这是 LGD-II 有证的落地执行器。触发词：证据链、论断验证、防幻觉、信源可信、有证、claim 证据、论断举证。
version: 1.0.1
agent_created: true
author: 诺声(Logos)@SynomosAI
license: MIT
license_scope: 代码（.py 文件）MIT；本 SKILL.md 与其中理论文本不在 MIT 覆盖范围内（见文末「License & attribution」）
allowed-tools: Read, Glob, Grep, Bash, Write, WebFetch
category: AI 治理
platforms: ["windows","macos","linux"]
read_when: ["用户要验证某个 AI 论断是否站得住脚","用户要防止幻觉、给结论配可核验证据","做事实核查、合规举证、技术写作的引用背书","要落地「凡所行必有证据」的有证原则"]
tags: ["LGD","有证","证据链","防幻觉","事实核查","信源","证成度","不编造"]
copyright: SynomosAI
description_zh: 当用户要『验证 AI 论断 / 防幻觉 / 给结论找证据 / 判断信源可信度』时用。把论断拆成可验证证据链：每条证据标来源类型(official/paper/data/internal/assertion)、是否可独立验证、可信度，输出『证成度』与未支撑论断清单。对齐 EIFP 不编造原教旨——本工具不判定论断真假，只评估证据质量；缺可验证证据的论断明确标『勿作结论』。这是 LGD-II 有证的落地执行器。触发词：证据链、论断验证、防幻觉、信源可信、有证、claim 证据、论断举证。
description_en: Decompose AI claims into verifiable evidence chains (LGD-II): each item tagged by source type (official, paper, data, internal, assertion) with independent verifiability status.
classification: {"skillhub":["ai-governance"],"iso_25010":["Security","Maintainability"]}
risk_tier: medium
risk_rationale: 门禁批量补丁默认 medium（按默认策略设计为中等风险）：调用 API/读写文件/本地执行，最小权限，密钥走 Secret；若实际涉对外代执行(push/发布/发送/上架)须人工复评，改 high 并走人工批准。
ai_protocol: true
spec: usk/1.0
interface: {"type":"cli","entry_point":"scripts/evidence_chain.py","runtime":"python3","call_pattern":"args"}
input_schema: {"args":{"type":"string","description":"命令行参数透传给主脚本，具体取值见各包 --help","required":false}}
output_schema: {"stdout":{"type":"string","description":"标准输出为 JSON 结果，含逐项检查与总体结论"},"exit_code":{"type":"integer","description":"0=PASS/完成；1=FAIL/有警告或待补强；2=用法或参数错误"}}
capabilities: ["cli_tool"]
permissions: {"network":false,"filesystem":false,"subprocess":false}
requirements: {"python_packages":[],"min_python":"3.8"}
changelog: 1.0.1: Bilingual rewrite (full EN+ZH, English first); 1.0.0: initial upload (USK v3)

凡自治之物 · LGD 治理学 · 创造术家族｜评测协议：见平台质量门禁规范

English

When to use

When the user presents an unverified AI claim, opinion summary or news item and needs to prevent hallucination while finding supporting evidence.
When verifying conclusions or assessing source credibility, needing to be explicit about the evidence quality of each source kind (official / paper / data, etc.).
When running the LGD-II "evidenced" executor, making sure every conclusion is traceable and EIFP-compliant.
When trigger words such as "evidence chain", "anti-hallucination" or "claim evidence" demand a structured decomposition of the verification process.

Usage

Call the Python script evidence_chain.py from the CLI or API, passing the claim text to be analyzed as the core input.
Optionally configure the source types (e.g. official / paper) and verification depth; by default the system runs the EIFP no-fabrication logic.
To get a detailed evidence-chain report, enable --verbose to surface the unsupported-claim list in full.
After reading the input, the script returns structured JSON: each evidence item's id, source and credibility score.

Workflow

Step 1: receive the target claim and extract the core assertion, scoping what must be verified.
Step 2: decompose complex claims into atomic evidence fragments, identifying each fragment's type and independence.
Step 3: match candidate evidence sources; assess how credible and independently verifiable the information is.
Step 4: per the EIFP no-fabrication principle, flag missing evidence or information gaps.
Step 5: aggregate all evidence quality into an overall justification score and generate a structured verification report.
Step 6: emit the result; claims that cannot be supported are explicitly marked "do not conclude" together with the gap list.

Evidence & safety statement

Evidence: TBD (no evaluation / badge link yet; update here once it is produced)
Version: 1.0.1
Security: no independent security scan has been performed (TBD); the conclusion is pending and no endorsement is given until it is completed.

中文

何时用

当用户提出未经证实的 AI 论断、观点摘要或新闻信息，需防止幻觉并寻找支撑证据时。
进行结论验证或信源可信度评估，需明确各类来源（官方/论文/数据等）的证据质量时。
执行 LGD-II 有证落地执行器任务，确保所有结论均有据可查且符合 EIFP 原则时。
当触发「证据链」「防幻觉」或「论断举证」等关键词，要求结构化拆解验证过程时。

用法

通过命令行或 API 调用 Python 脚本 evidence_chain.py，指定待分析论断文本作为核心输入参数。
可选配置源信源类型（如 official/paper）及验证深度，系统默认按 EIFP 标准执行不编造逻辑。
若需输出详细证据链报告，需在脚本调用中开启 --verbose 模式以获取未支撑清单细节。
脚本读取输入后，将返回结构化 JSON 数据，包含每条证据的标号、来源及可信度评分。

流程

步骤 1：接收用户输入的目标论断，提取核心主张并明确待验证的范围。
步骤 2：将复杂论断拆解为原子级证据片段，识别每个片段的类型与独立性。
步骤 3：匹配潜在证据库或来源，评估信息的真实可信度与是否可独立验证。
步骤 4：依据 EIFP 不编造原教旨原则，标记是否存在缺位证据或缺口信息。
步骤 5：综合所有证据质量计算整体证成度数值，生成结构化验证报告。
步骤 6：输出结果，对无法得到支持的论断明确标注『勿作结论』及缺失清单。

证据与安全声明

证据：TBD（暂无评测/徽章链接；产生后在此更新实际链接）
版本：1.0.1
安全：未做独立安全扫描（TBD），结论待补；补全前不作推荐背书。