fact-check-x-unified
Orchestrates the fact-checking of AI responses. It summarizes, validates, and evaluates different answers to create a complete, auditable final report.
What it does
Checks the accuracy of information by comparing answers from different sources and using trusted sources to confirm facts. It generates detailed reports on the veracity of the information.
When to use it
- ✓Comparing answers from multiple platforms on the same topic.
- ✓Distinguishing between unverified drafts and verified conclusions.
- ✓Maintaining an auditable and transferable record of fact-checks.
Example of what to ask your AI
“Fact-check the information about the benefits of solar energy.”
Want more? Open the sections below.
How to use
The easiest way: copy the message below and paste it into Claude, ChatGPT, Cursor or any AI. It reads the skill and follows the instructions.
Skill link
Ready-to-paste message
How it works
This agent orchestrates calls to other tools that gather answers, compare information locally, and verify facts with authoritative sources. It doesn't perform the checks itself but manages the overall process.
Installation
Easiest way: copy the skill link and paste it into Claude, ChatGPT or Cursor asking it to follow the instructions.
In the terminal (Claude Code, Codex, Cursor), in your project folder:
npx skills add sickn33/agentic-awesome-skills --skill fact-check-x-unifiedOr download just the file:
curl -L https://raw.githubusercontent.com/sickn33/agentic-awesome-skills/main/plugins/agentic-awesome-skills-claude/skills/fact-check-x-unified/SKILL.md -o SKILL.mdTechnical content (original SKILL.md)
Original text by the author, straight from GitHub.
When to Use
- 用户需要比较一个或多个平台的回答、区分未核验草案与已核验结论,并保留来源、逐知识点裁决和可迁移报告。
- 已有采集结果或准备好独立安装并审查的三个依赖技能;联网、账户配置和本机写入必须符合用户授权。
- 中文是原始工作流的目标语言;用户指定其他语言时用其语言交互,保留数据契约中的字段名。
Fact-Check-X 统一入口
本技能只做编排,不内嵌三个业务层。它按以下顺序调用兄弟技能:
llm-answer-reference-compare:多平台回答、引用与现场存证采集。fact-check-x-knowledge-compare:本地知识点结构化对比,可选。fact-check-x-authoritative-verify:逐知识点并发权威核验与平台表现报告。
所有语义拆解和证据裁决由当前承载技能的智能体完成。脚本不调用任何模型 API。
依赖定位
AAS 仅分发本编排技能的脚本与测试,未分发 llm-answer-reference-compare、fact-check-x-knowledge-compare、fact-check-x-authoritative-verify 的采集器、渲染器和夹具。因此单独安装本技能不能运行完整流水线,也不能假称已经打开登录页面。
依赖恢复:从 已核对的官方来源 取得四个完整技能及其依赖。先审查固定修订的来源、许可证、锁文件、浏览器与账户数据边界;不要由本技能静默下载或安装。单独安装的完整三个依赖目录可通过以下变量定位;自动配置组件使用相同的目录:
export FACTCHECK_SKILLS_DIR="<已审查的完整上游分发包的 skills 绝对目录>"
脚本路径相对于本技能安装目录;执行前进入该目录。先检查依赖(缺少可执行入口时明确失败,不以仅存在 SKILL.md 作为就绪证明):
python3 scripts/fact_check_x.py locate
完整流程
先采集 results.json,然后准备知识点对比:
python3 scripts/fact_check_x.py prepare-comparison \
--results <results.json> \
--run-dir <run>
命令会同步生成“各方答案汇总”,在运行目录顶层生成 01-capture-report.html。每个 deliverables 条目同时返回已验证存在的本机 path、标准 fileUri 和可直接展示的 markdownLink。调用方必须原样展示 markdownLink,不得自行拼接 Windows 路径或把 path 直接塞进 Markdown;产物不存在时命令会在当前阶段失败。
默认交互模式下,程序会把阶段状态写入 stage-checkpoints.json。调用方发送本阶段产物并收到用户“继续下一步”后,必须使用 checkpoint.acknowledgement.token 执行:
python3 scripts/fact_check_x.py acknowledge-stage \
--run-dir <run> --stage <checkpoint.stage> \
--token <checkpoint.acknowledgement.token> --decision continue
未确认时,下一阶段命令会直接失败。当前程序实现的是逐阶段确认模式;执行完整流程前说明这一约束。用户要求连续执行时不要假称已经支持自动模式,也不要伪造确认令牌。
当前智能体读取 <run>/comparison-task.json,写入 <run>/comparison-analysis.json,再执行:
comparison-task.json 是自包含的紧凑工作集。阶段确认后只允许“读取任务包一次、写入分析一次、执行验收一次”,以紧凑任务包为主要输入,避免无必要的重复读取或完整转储。发生错误、歧义或契约问题时应读取必要的源码或证据定位原因;三次调用是上游效率建议,不得用它阻止正确性或安全检查。capturedText 只是有界原文预览,完整正文由程序在验收时从 results.json 校验和重建。
拆解时,原子性同时约束知识点和各平台 claim。一个原句包含多个独立义务、条件、对象、数值或后果时必须拆点,每个 claim 只保留当前事实;平台独有的实质新增事实也要另起无锚点知识点,不能并入宽泛知识点后复用深知晓锚点免查。每个知识点同时标明 claimType=fact|recommendation:纯操作建议可标为 recommendation,不因缺少逐句脚标而判引用不忠实;建议中包含的制度事实、条件、数字或时效必须拆成独立 fact。
python3 scripts/fact_check_x.py complete-comparison \
--results <results.json> \
--run-dir <run>
命令会在运行目录顶层生成 02-comparison-report.html,并通过 deliverables[].markdownLink 返回可点击链接;调用方必须原样展示。报告必须包含明确标为“未核验”的综合草案。默认交互模式下,用户确认继续后才能进入权威核验。
生成每个知识点的独立云端请求并并发取证:
python3 scripts/fact_check_x.py prepare-authority --run-dir <run>
python3 scripts/fact_check_x.py search-authority --run-dir <run> --max-workers 12
深知晓采集同时保留回答内逐句脚标和“知识专库”中的回答级来源;相同 URL 合并为 inline_and_global。回答级来源只有经当前知识点逐条语义匹配且原文可定位时才可支持该事实,不得整库继承。知识点已有深知晓/深知晓(深度溯源)本次回答所附的官方材料,或其他平台本次回答所附的 gov.cn 材料,且原文确实支持当前事实主张时,直接复用为官方证据,不调用可信搜索。纯操作建议同样不调用可信搜索;除此之外,已有材料不足以裁决的事实知识点才是非免查知识点。
若存在非免查知识点但本机尚无可信搜索配置,prepare-authority 和 search-authority 会返回 status=configuration_required、userPrompt 与 configuration.command 并以非零状态退出。调用方先展示登录提示和外部提供方、账户/Key 创建、可能费用、发送数据范围及本机保存位置,获得授权后再前台执行该命令。不得把“配置缺失”当成自动创建账户凭据或联网的授权。用户只需在自动打开的深知 MaaS 页面完成登录;组件会自动复用已有完整 Key,没有时创建 Fact-Check-X 专用 Key,验证后保存到 ~/.fact-check-x/credentials/trusted-search-key。Codex、Claude Code、WorkBuddy 等载体共享该配置,检测到已有 Key 时直接跳过登录。配置成功后调用方自动重跑 prepare-authority,不得要求用户复制 Key、编辑 shell 配置、回复“已配置”,也不得改用深知晓来源或普通搜索绕过。
只有可信搜索返回 401/403 时才把现有 Key 视为失效并进入自动配置。超时、断网或服务异常保留现有配置并自动重试;重试后仍失败则停留在搜索阶段返回技术错误,不要求用户重新登录,也不转成人工事实复核。
prepare-authority 会写入 authority-gate.json。只有可信搜索批次正常完成、门禁状态变为 searched 后,finalize-authority 才允许运行;手工写 evidence 或 result 不能绕过。
当前智能体逐个读取 authority/requests 与 authority/evidence,把裁决写入 authority/assessments/<知识点ID>.json,然后:
免查模式下,trustedAnchor.officialAnswer 是当前事实知识点的权威结论;锚点证据列表用于来源追溯。无论来自深知晓还是 gov.cn,都必须确认已捕获正文实际支持该结论,不得仅凭官方属性自动判定。trustedAnchor.claimEvidenceMap 记录各平台已经逐主张绑定的官方原文;命中该映射的平台直接复用本次回答自带证据,不得因另一轮可信搜索未召回同一材料而降为证据不足。其他平台主张与 officialAnswer 语义一致或可由其直接推出时,裁决为 supported 并引用有效锚点证据 ID。仅当平台主张增加了权威结论不能支持的实质事实,或确实无法判定时,才使用 insufficient。知识点对比阶段的引用忠实性与本阶段的事实正确性分别保留,不能因平台自身引用不足而拒绝判断已经被权威结论证实的主张。recommendation 作为“操作建议”单独展示,不进入事实准确率或幻觉率;若权威证据证明建议会误导,仍可判为 contradicted。
{
"authoritativeFinding": "由权威证据支持的知识点结论",
"verdicts": {
"<平台ID>": {
"verdict": "supported",
"reason": "主张与 E1 一致",
"evidenceIds": ["E1"]
}
}
}
…
FAQ
What are 'sibling skills'?
These are other tools or AI agents that this main agent calls upon to perform specific tasks, such as collecting answers or fact-checking.
Do I need to install other tools to use this one?
Yes, this tool acts as an orchestrator and requires other specific tools to function fully. Installation and setup instructions are provided.
The original language is Chinese, but can I use it in Portuguese?
Yes, the system can interact in Portuguese and process the information, while maintaining the data structure.
Details
Original description by the author
Fact-Check-X 流程编排能力,依次组织各方答案汇总、各方答案聚合(未核验)、权威核验后的最终答案和各方答案测评,生成可打开、可审计、可迁移的阶段产物与完整报告包。
Popularity
- GitHub stars
- 48k
- Ranking position
- #2970
- License
- MIT
- Updated
- 10/15/2026
Comments
- …
More skills from this repository
00-andruia-consultant
Helps your agent start or improve AI projects. It analyzes code, suggests tools, and creates action plans in Spanish.
sickn33/agentic-awesome-skills
007
Lets your agent perform security audits and threat modeling. It's useful for protecting projects against attacks and vulnerabilities.
sickn33/agentic-awesome-skills
10-andruia-skill-smith
Helps your agent create or improve AI tools. It defines how tools work, writes documentation, and validates the structure.
sickn33/agentic-awesome-skills
20-andruia-niche-intelligence
Lets your agent create a summary about a specific industry. It's useful for understanding the context, rules, usage patterns, and risks of a project, using reliable sources.
sickn33/agentic-awesome-skills
Similar skills
agent-architecture-audit
Makes your agent check if everything is working correctly. It finds and fixes problems in AI systems, ensuring they perform as expected.
affaan-m/ECC
agent-eval
Helps your agent compare different AI coding tools. It measures which one is faster, cheaper, and more reliable for specific tasks.
affaan-m/ECC
agent-harness-construction
Makes your agent better understand how to perform tasks. It optimizes the tools and instructions the agent uses to increase its chances of success.
affaan-m/ECC