Deep research skill for AI agents: live web research, source reputation checks, safer URL fetches, and structured evidence feedback.
-
Updated
Aug 25, 2026 - Python
Deep research skill for AI agents: live web research, source reputation checks, safer URL fetches, and structured evidence feedback.
Privacy-safe Hermes researcher for web, GitHub, YouTube, Reddit, documents, source-lineage checks, counterexamples, and decision-ready briefs
Open-source Claude Code skills — 6 cognitive firewalls block AI hallucination, bias & sloppy reasoning. npx skills add
保洁:它为研究工作做两类检查:研究过程中,把公开资料整理成可复核的证据;交付前,核对报告、图表和附件,并复查修改结果。
Stop hallucinations by verifying citations.
Verifiable Codex + Claude + Gemini collaboration framework with source checks, mutual cross-audit, merge gates, HTTP and MCP adapters.
MedFailBench exposes critical safety boundary failures in medical AI through clinician built, reproducible open source evaluation.
AI 조사 결과의 출처를 믿기 전에 검증하는 Claude Code 스킬 — 주장–근거 원장으로 발췌를 원문과 기계 대조, 4-Tier·재인용 체인·URL/DOI 실존 확인, 한국 학술DB(KCI·RISS·NKIS) 대응 | Source-verification gate with a claim–evidence ledger that checks excerpts against saved originals
StormGrill is an open-source dual-protocol orchestrator that pairs verified multi-agent research (Storm) with real-time adversarial interrogation (Grill) to help decision-makers stress-test strategic decisions against facts, not intuition. Make bad strategic decisions expensive to make.
Repeatable company-research pipeline for Claude Code: seven cited, confidence-tagged dossiers per company.
给 Agent 一个链接、仓库、文章或截图,格物会完整查看、追溯源头、拆解实现与价值,并判断是否值得使用、复刻或二开。
Claim it. Prove it. Settle it on GenLayer with AI-powered 1v1 prediction challenges.
Web research MCP server with claim verification — 45 tools for Claude Desktop and Cursor
World Bank data assistant for investigative journalists — ask questions, get cited tables and methodological notes from real World Bank APIs, with built-in anti-hallucination guardrails
Research with Perplexity, ChatGPT or Gemini and know which claims hold up. An agent skill for source-demanding research prompts, claim-by-claim answer checks and verified DOCX/PDF reports.
Local, fail-closed source quorum verification
Read-only fact-honesty gate: every statistic in your copy must trace to a cited source manifest. No network, no credentials, no auto-run. Exit 1 on an unsourced claim.
AI-powered fake news detection system using advanced NLP, fact verification, and source reliability analysis. Built with Next.js 14, featuring real-time credibility assessment, comprehensive RESTful API, and professional dark/light mode interface for combating misinformation.
Open-source Agent Skill for evidence-based assessment of information source trustworthiness across regions, domains, and source types.
卫生法庭 HEALTH COURT* — 对文化内容进行来源核查的漫画法庭(机制严谨 × 呈现漫画)。纯静态 BYOK,多智能体庭审管线。
To associate your repository with the source-verification topic, visit your repo's landing page and select "manage topics."