mcpField
F
fact-checker

fact-checker

github.com/kirkluokunDocuments6637 installsUpdated 32w ago
Repository

About fact-checker

fact-checker is a multi-agent skill for checking document claims. It first splits text and extracts claims, then verifies them with Google Search and reports issues.

Key features

  1. 01Two-phase splitting and verification
  2. 02Classifies factual and subjective text
  3. 03Filters documents by chapter
  4. 04Concurrency controls and dry runs
  5. 05JSON outputs for sentences, checks, and issues

Use cases

  1. 01Check factual claims in long documents
  2. 02Review sentence splitting and classification first
  3. 03Verify selected chapters in batches

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/kirkluokun/awesome-a-stock-openclawskills --skill fact-checker

After installing, type/fact-checkerin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

.env.example
ARCHITECTURE.md
DELIVERY.md
README.md
scripts/
fact_checker.py

Details

From the project README

Fact Checker — 多 Agent 文档事实核查

两阶段流水线架构：

  • Phase 1 (Splitter): gemini-2.0-flash → 拆句、分类(factual/subjective)、提取声明
  • Phase 2 (Verifier): gemini-2.5-flash + Google Search → 逐句逐声明核查

使用方法

# 完整核查
python {SKILL_DIR}/scripts/fact_checker.py <file.md>

# 指定章节
python {SKILL_DIR}/scripts/fact_checker.py <file.md> --chapters 2,3

# 仅拆句（检查 Phase1 质量，不花搜索费用）
python {SKILL_DIR}/scripts/fact_checker.py <file.md> --split-only

# 预览文档结构
python {SKILL_DIR}/scripts/fact_checker.py <file.md> --dry-run

参数

参数说明默认值
-o, --output输出目录文件同目录
--split-modelPhase1 模型gemini-2.0-flash
--verify-modelPhase2 模型gemini-2.5-flash
--split-concurrencyPhase1 并发5
--verify-concurrencyPhase2 并发3
--chapters章节过滤全部
--split-only仅拆句false
--dry-run仅解析false

输出文件

文件说明
{stem}_sentences.jsonPhase1: 全部句子 + 分类 + 声明
{stem}_factcheck.jsonPhase2: 完整核查结果
{stem}_issues.json仅 incorrect / not_found 条目

编号规则

C{章}.S{节}.SS{小节}.P{段落}.T{句子}.K{声明}

示例：C2.S1.P3.T2.K1 = 第1章(C2) > 第1节 > 第3段 > 第2句 > 第1个声明

工作流建议

  1. --dry-run → 确认文档结构和章节编号
  2. --split-only → 检查拆句和分类质量
  3. 完整核查 → 查看 _issues.json 定位问题
  4. 对大文档按章节分批核查（--chapters 2,3 + --chapters 4,5）

Comments

Markdown supported