Fact Checker

fact-checker 从 markdown 草稿中提取数字、模型引用、日期、分数与百分比等可核验声明，按优先级比对 FINDINGS.md、本地 /status API、scores JSON、memory 日志、git 历史与 CHANGELOG，输出 CONFIRMED/UNVERIFIABLE/CONTRADICTED 三类结果及汇总，并对矛盾项建议修正。

nissan

热度 90.78

最近更新 2026-05-11 17:27:32

分类：数据依赖：代码执行 / 读写文件应用场景：校对润色 / 报告撰写 / 研究

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 fact-checker

Metadata

name: fact-checker
version: 1.0.3
description: Verify claims, numbers, and facts in markdown drafts against source data. Use when: reviewing blog posts, reports, or documentation for accuracy before publication. Cross-references against FINDINGS.md, live APIs, scored run files, memory logs, and git history.
author: nissan
tags: ["fact-checking","content","verification"]
requires: {"bins":["python3"]}
metadata: {"openclaw":{"emoji":"🔍","network":{"outbound":false,"reason":"All verification against local files and APIs only"},"subprocess":{"note":"Runs python3 scripts for verification — legitimate use"},"security_notes":"All operations are performed locally. No data leaves the user's machine. Verification cross-references local files, local git history, and a local API (localhost). The subprocess note is for running the bundled Python fact-check script — not arbitrary code execution."}}
allowed-tools: ["Read","exec"]

Last used: 2026-03-24 Memory references: 1 Status: Active

Fact-Checker: Verify Markdown Claims Against Source Data

Given a markdown draft file, this skill extracts every verifiable claim (numbers, dates, model names, scores, causal statements) and cross-references them against available source data to produce a verification report.

Usage

python3 skills/fact-checker/scripts/fact_check.py <draft.md>
python3 skills/fact-checker/scripts/fact_check.py <draft.md> --output report.md

What It Checks

Claim types extracted

Numeric claims — integers and floats with surrounding context
Model references — model/task (phi4/classify) and model:tag (phi4:latest)
Dates — YYYY-MM-DD format
Score values — decimal scores like 0.923, 1.000
Percentages — 42%, 95.3%

Source data consulted (in priority order)

projects/hybrid-control-plane/FINDINGS.md — primary source of truth
Control Plane /status API at http://localhost:8765/status — live scored run data
projects/hybrid-control-plane/data/scores/*.json — raw scored run files on disk
memory/*.md — daily logs with timestamps and decisions
git log in projects/hybrid-control-plane/ — commit hashes, dates, authorship
projects/hybrid-control-plane/CHANGELOG.md — sprint history

Output Format

Each claim produces one line:

✅ CONFIRMED:    "phi4/classify scored 1.000" → /status API: phi4_latest_classify mean=1.000 n=23
⚠️ UNVERIFIABLE: "this took about a day" → no timestamp correlation found in logs
❌ CONTRADICTED: "909 runs" → /status API shows 958 total runs (stale number?)

Followed by a summary count of confirmed / unverifiable / contradicted claims.

When To Use This Skill

When asked to "fact-check" or "verify" a draft blog post, report, or documentation file — run this skill and present the report to the user. If any claims are ❌ CONTRADICTED, flag them prominently and suggest corrections.

Instructions for Agent

Run the script with the path to the draft file.
Parse the output report.
Summarise key findings — especially any ❌ CONTRADICTED claims.
Suggest specific corrections with the correct values from the evidence.
If the /status API is unavailable, note it and rely on FINDINGS.md + score files.