← 返回 Skills 市场

AI幻觉检测专家

Name: AI幻觉检测专家
Author: ai-gaoqian

作者 ai-gaoqian · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ✓ 安全检测通过

总下载

当前安装

版本数

在 OpenClaw 中安装

/install ai-hallucination-detector

功能描述

精准识别和分析LLM输出内容中的事实错误和不实信息，辅助内容真实性验证与质量评估。

使用说明 (SKILL.md)

AI骞昏妫€娴嬩笓瀹?(Hallucination Detector)

姒傝堪

AI骞昏妫€娴嬩笓瀹舵槸涓€涓笓涓氱殑LLM杈撳嚭璐ㄩ噺瀹℃牳鎶€鑳斤紝鑳藉瀵逛换浣旳I鐢熸垚鐨勬枃鏈繘琛岀郴缁熸€у够瑙夋娴嬶紝璇嗗埆鍏朵腑鐨勪簨瀹炴€ч敊璇€侀€昏緫鐭涚浘鍜屼俊鎭笉涓€鑷淬€?

鏍稿績妫€娴嬬淮搴?1. **瀹炰綋瀛樺湪鎬ч獙璇?* - 妫€娴嬫枃鏈腑鎻愬埌鐨勪汉鐗┿€佺粍缁囥€佸湴鐐规槸鍚︾湡瀹炲瓨鍦?2. **鏃堕棿涓€鑷存€ф鏌?* - 楠岃瘉鏃堕棿绾块€昏緫锛屾娴嬫棩鏈熺煕鐩?3. **鏁板€煎噯纭€ф牳楠?* - 缁熻鏁版嵁銆佹暟瀛楀紩鐢ㄧ殑浜ゅ弶楠岃瘉

*寮曠敤鐪熷疄鎬у鏌? - 璁烘枃/鏂囩尞/鎶ュ憡寮曠敤鐨勭湡瀹炴€ф鏌?5. 鍥犳灉閫昏緫鍒嗘瀽 - 妫€娴嬪洜鏋滃€掔疆銆佽櫄鍋囧洜鏋滃叧绯?6. *棰嗗煙鐭ヨ瘑涓€鑷存€? - 鍩轰簬涓撲笟鐭ヨ瘑搴撻獙璇侀鍩熸柇瑷€
*璺ㄦ钀借嚜娲芥€? - 妫€娴嬪叏鏂囧墠鍚庣煕鐩?8. *閲忓寲鏁版嵁鍚堢悊鎬? - 缁熻鏁版嵁鏄惁鍦ㄥ悎鐞嗚寖鍥村唴
*鏈浣跨敤姝ｇ‘鎬? - 涓撲笟鏈鏄惁浣跨敤鎭板綋
*鏉ユ簮鍙俊搴﹁瘎浼? - 瀵规湁鏉ユ簮澹版槑鐨勪俊鎭繘琛屽彲淇″害璇勫垎
*甯歌瘑涓€鑷存€ф鏌? - 鍩虹甯歌瘑鎬ч敊璇瘑鍒?12. 鏁板杩愮畻楠岃瘉 - 鏂囦腑璁＄畻鐨勫噯纭€?13. *澶氳瑷€缈昏瘧鍑嗙‘鎬? - 缈昏瘧鍐呭鐨勫繝瀹炲害妫€鏌?14. *璇箟婕傜Щ妫€娴? - 闀跨瘒鏂囨湰涓涔夊亸绂?15. *缂栭€犱俊鎭瘑鍒? - 璇嗗埆鍏稿瀷鐨凙I缂栭€犳ā寮?16. *鍋忚涓庡埢鏉垮嵃璞℃娴? - 绀句細鍋忚鍐呭绛涙煡

浣跨敤鏂瑰紡

clawhub install hallucination-detector

瀹夎鍚庯紝Agent鑷姩鑾峰緱骞昏妫€娴嬭兘鍔涖€傚浠讳綍AI鐢熸垚鍐呭璋冪敤妫€娴嬶細

璇峰浠ヤ笅鍐呭杩涜骞昏妫€娴嬶細[鍐呭]

杈撳嚭鏍煎紡

{
  "overall_score": 0.85,
  "risk_level": "low",
  "total_issues": 3,
  "dimensions": [
    {
      "name": "瀹炰綋瀛樺湪鎬ч獙璇?,
      "score": 0.95,
      "issues": [],
      "status": "pass"
    }
  ],
  "hallucination_instances": [
    {
      "type": "factual_error",
      "dimension": "鏃堕棿涓€鑷存€?,
      "location": "绗?娈电2鍙?,
      "original": "...",
      "correction": "...",
      "confidence": 0.92
    }
  ],
  "recommendations": ["寤鸿鏍稿疄..."]
}

瀹夊叏涓庨殣绉?- 绾湰鍦版帹鐞嗭紝妫€娴嬪唴瀹逛笉涓婁紶浠讳綍澶栭儴鏈嶅姟鍣?- 浣跨敤鏈湴宓屽叆妯″瀷杩涜璇箟鐩镐技搴﹁绠?- 鏀寔绂荤嚎杩愯锛堥櫎棣栨妯″瀷涓嬭浇锛?- 涓嶈褰曟垨缂撳瓨鐢ㄦ埛妫€娴嬪唴瀹?

閫傜敤鍦烘櫙

鍐呭鍒涗綔鑰咃細鍙戝竷鍓嶅鏍窤I杈呭姪鐢熸垚鐨勬枃绔?- 瀛︽湳鐮旂┒鑰咃細鏍￠獙鏂囩尞缁艰堪涓殑寮曠敤鍜屾柇瑷€
娉曞緥浠庝笟鑰咃細瀹℃煡AI鐢熸垚鐨勬硶寰嬫枃涔?- 鍖荤枟鍋ュ悍锛氶獙璇丄I鍋ュ悍寤鸿鐨勪簨瀹炲噯纭€?- 鏂伴椈濯掍綋锛欰I鏂伴椈绋跨殑浜嬪疄鏍告煡
浼佷笟鍚堣锛氬唴閮ㄦ姤鍛婂拰瀵瑰澹版槑鐨勫噯纭€у鏌?

安全使用建议

Installers should expect this skill to analyze text they deliberately provide for fact-checking. Avoid sending sensitive documents unless you are comfortable with local model processing and any first-run model download behavior in your agent environment.

能力评估

✓ Purpose & Capability

The stated purpose is to evaluate AI-generated text for factual, logical, citation, numerical, and consistency issues; that capability matches the documented use cases and output format.

ℹ Instruction Scope

The skill describes checking any AI-generated content, but the artifact shows an explicit user-invoked prompt pattern rather than automatic interception of unrelated conversations.

✓ Install Mechanism

The artifact contains only SKILL.md, with no executable scripts or install-time code; declared Python packages are consistent with local text analysis.

✓ Credentials

The declared Python, sentence-transformers, spaCy, NumPy, CPU, and memory needs are proportionate for local semantic analysis; no API keys are requested.

✓ Persistence & Privilege

No persistence, credential access, background execution, destructive action, or external upload behavior is present in the artifact; it claims local inference and no content logging.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install ai-hallucination-detector
安装完成后，直接呼叫该 Skill 的名称或使用 /ai-hallucination-detector 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

AI Hallucination Detector v1.0.0 - Initial release of a professional tool for comprehensive LLM output hallucination and factuality detection. - Supports 16 evaluation dimensions, including fact consistency, temporal alignment, statistical accuracy, and use of proper terms. - Structured JSON output provides detailed scoring, identified hallucination instances, and targeted recommendations. - Designed for diverse scenarios such as content moderation, academic review, legal text verification, medical review, news validation, and enterprise report quality assurance. - Runs locally, ensures privacy by not uploading content externally, and works offline except for initial model download.

元数据

Slug ai-hallucination-detector

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题