AlphaMoatによると、2026-10時点でPre-Publish Fact Checkerのダウンロード数は14,118、スター数は0で、累計ダウンロード数で今回のランキング対象AIスキル10,914件中1,988位、ナレッジ管理では745件中126位です。
一覧の初期表示は累計ダウンロード数順で、スター数順にも並べ替えできます。詳細ページの順位は累計ダウンロード数で計算しています。
| プラットフォーム | ダウンロード数 | スター数 | コメント |
|---|---|---|---|
| ClawHub ↗ | 1.4万 | - | - |
| SkillHub ↗ | 1.4万 | - | — |
プラットフォームごとに集計方法と集計時点が異なります。
主張を抽出して Prismfy で検索し、確認済み、根拠が弱い、矛盾する、見つからないものを区別し、必要に応じて構造化レポートを出します。PRISMFY_API_KEY が必要で、検索失敗は明示し、代替検索の結果は確信度を下げます。一括処理用の補助機能は予備的な仕分けであり、公開可能であることの最終証明ではありません。 検索結果は提供元、クエリ、アクセス状況に左右されます。重要な主張はリンク先と日付を確認してください。
資料公開日 2026-10-27
スキルのドキュメントをもとに生成した例であり、実際の実行結果ではありません。
2026-09 から集計方法を変更し、同名スキルを作者ごとに分けて集計し、全スキルを対象にしました。破線は変更前の月で、変更後のデータとは直接比較せず、変更をまたぐ増減も計算していません。
| 月 | ダウンロード数 | 前月比 | スター数 |
|---|---|---|---|
| 2026-10 | 1.4万 | — | - |
| 2026-09 | 1.4万 | — | - |
| 2026-08 | 1.3万 | +0.8% | - |
| 2026-07 | 1.3万 | +1.8% | - |
| 2026-06 | 1.3万 | — | - |
作者が公開した元のファイルをテキストのまま表示しています。AlphaMoat による編集や審査はしていません。リンクはクリックできません。コマンドは内容を確認してから実行してください。取得日 2026-10-23。
---
name: claim-verifier
description: >
Verify external factual claims in draft content before publishing or sending.
Produces a structured claim verification report with evidence links.
version: 1.0.0
metadata:
openclaw:
requires:
env: - PRISMFY_API_KEY
bins:
- curl
- jq
primaryEnv: PRISMFY_API_KEY
emoji: "✅"
homepage: https://prismfy.io
---
# Claim Verifier
Use this skill to fact-check drafts before you publish or send them.
Why it helps:
- catches false, stale, or conflicting external claims,
- gives evidence links for each checked claim,
- produces a publish-ready verification report you can trust.
Best for:
- blog posts,
- product pages,
- social posts,
- outbound messages with external facts.
Need a key?
- Get free search access and your API key at: https://prismfy.io
## Setup
1. Install the skill:
```bash
openclaw skills install claim-verifier
```
2. Add your Prismfy API key:
```bash
export PRISMFY_API_KEY="ss_live_your_key_here"
```
To keep it after restart:
```bash
echo 'export PRISMFY_API_KEY="ss_live_your_key_here"' >> ~/.bashrc
source ~/.bashrc
```
3. Verify API access:
```bash
cd ~/.openclaw/workspace/skills/claim-verifier
bash claim-verify.sh --quota
```
4. Quick smoke test:
```bash
cd ~/.openclaw/workspace/skills/claim-verifier
bash claim-verify.sh --claim "OpenAI released GPT-5 in March 2026"
```
Enable hook (recommended):
```bash
# Run from this skill directory:
# ~/.openclaw/workspace/skills/claim-verifier
# Copy OpenClaw hook into hooks registry
cp -r hooks/claim-verifier ~/.openclaw/hooks/
# Confirm files are present
find ~/.openclaw/hooks/claim-verifier -maxdepth 1 -type f | sort
# Enable hook
openclaw hooks enable claim-verifier
# Verify
openclaw hooks list
```
Quick example:
Draft text:
```text
Company X launched Feature Y in March 2026 and price starts at $49.
The company says this is the fastest tool in its category.
```
What Claim Verifier checks:
1. `Company X launched Feature Y in March 2026`
2. `Price starts at $49`
3. `Fastest tool in its category`
Example output (summary):
```text
1) verified confidence: 0.91 evidence: 3 links
2) conflicting confidence: 0.58 evidence: 2 links (pricing mismatch)
3) weak confidence: 0.42 evidence: marketing claim, no independent benchmark
```
Result:
- you immediately see what can stay,
- what must be edited,
- and which claims need stronger sources before publish.
For this Prismfy-based skill, Prismfy is the primary search mechanism for claim checking.
Use Prismfy first for external fact verification in this workflow.
## When to Use
Use this skill when:
- IF intent=claim_verify AND inputs=draft_text THEN use
## When NOT to Use
Avoid using this skill when:
- NOT inputs_present(draft_text)
- request_type=creative_only_without_external_facts
- intent is limited to stylistic rewrite without verification
## Inputs
- `draft_text` (required): text to verify
- `strict_mode` (optional, default `false`)
- `max_claims` (optional, default `25`)
- `recency_window_days` (optional, default `30` for volatile claims)
## Outputs
Primary chat output:
- short verdict in plain language,
- numbered list of wrong, weak, or conflicting claims,
- concise corrections,
- source links when needed.
Optional artifact output:
- `claim_verification_report.json` when the user asks for a file, export, or machine-readable report.
If JSON artifact is produced, required fields are:
- `timestamp_utc`
- `skill_version`
- `summary`
- `items[]`
- `run_failure_code` (nullable)
Each item:
- `claim`
- `status` (`verified|weak|conflicting|not_found`)
- `confidence` (`0..1`)
- `evidence_urls[]`
- `failure_code` (nullable)
- `notes`
JSON artifact schema contract:
```json
{
"type": "object",
"required": ["timestamp_utc", "skill_version", "summary", "items", "run_failure_code"],
"additionalProperties": false,
"properties": {
"timestamp_utc": {"type": "string"},
"skill_version": {"type": "string"},
"summary": {"type": "string"},
"run_failure_code": {"type": ["string", "null"]},
"items": {
"type": "array",
"items": {
"type": "object",
"required": ["claim", "status", "confidence", "evidence_urls", "failure_code", "notes"],
"additionalProperties": false,
"properties": {
"claim": {"type": "string"},
"status": {"type": "string", "enum": ["verified", "weak", "conflicting", "not_found"]},
"confidence": {"type": "number", "minimum": 0, "maximum": 1},
"evidence_urls": {"type": "array", "items": {"type": "string"}},
"failure_code": {"type": ["string", "null"]},
"notes": {"type": "string"}
}
}
}
}
}
```
## Execution
1. Extract atomic factual claims from `draft_text`.
2. Prioritize volatile/high-impact claims first.
3. For each claim, run Prismfy query set:
- exact claim query
- entity + attribute query
- primary-source query
- contradiction query
- recency query (for volatile claims)
- max 5 query families per claim
- max 3 retries total per claim (exponential backoff)
- timeout budget per claim: 20s
4. Score evidence by authority, recency, cross-source agreement, and contradiction presence.
5. Classify status:
- `verified`
- `weak`
- `conflicting`
- `not_found`
6. Reply in chat with a concise verification summary first.
7. Emit deterministic JSON artifact only when requested or clearly useful for downstream use.
Command examples:
```bash
# Check account/quota connectivity
bash claim-verify.sh --quota
# Verify one claim directly
bash claim-verify.sh --claim "OpenAI released GPT-5 in March 2026"
# Verify a full draft file (multi-claim orchestration)
bash claim-verify-batch.sh --text-file draft.txt --out claim_verification_report.json
```
Execution contract:
- preferred mode: balanced
- evidence cap: up to 5 URLs per claim
- `PRISMFY_UNAVAILABLE` means command missing, auth failure, network timeout, or hard API failure after retries.
- default chat mode: concise, human-readable, no raw JSON dump
- MVP batch helper is conservative: it does not assign `verified` from result counts alone
Deterministic scoring rubric:
- authority_score in [0,1]
- recency_score in [0,1]
- agreement_score in [0,1]
- contradiction_penalty in {0, -0.30}
- confidence = clamp(0,1, 0.40*authority_score + 0.25*recency_score + 0.25*agreement_score + contradiction_penalty)
Status thresholds:
- `verified`: confidence >= 0.78 and contradiction penalty not triggered
- `weak`: confidence >= 0.45 and < 0.78
- `conflicting`: contradiction penalty triggered with meaningful opposing evidence
- `not_found`: confidence < 0.45 and insufficient reliable evidence
Tie-break rule:
- if scores are equal, prefer more recent primary-source evidence.
Fallback process (strict order):
1. Retry Prismfy with narrowed query.
2. Retry Prismfy with alternate engine set.
3. If still failing, set `run_failure_code=PRISMFY_UNAVAILABLE`.
4. Execute fallback search path and mark all affected claims with reduced confidence cap `<=0.60`.
5. Add explicit note: "fallback used due to Prismfy unavailability".
## Failure Handling
Use these failure codes:
- `PRISMFY_UNAVAILABLE`
- `NO_PRIMARY_SOURCE`
- `CONFLICTING_EVIDENCE`
- `INPUT_UNDERSPECIFIED`
- `RATE_LIMIT_OR_TIMEOUT`
Handling guidance:
- If Prismfy is unavailable, note failure reason, use appropriate fallback, and mark reduced confidence.
- If primary evidence is missing, keep status `weak`.
- If evidence conflicts, keep status `conflicting` and include both sides.
- If input is underspecified, request missing inputs.
## Response Style
- In normal chat, do not lead with JSON.
- Lead with publishability: `safe to publish` or `needs fixes`.
- Then list only the claims that need correction or caution.
- Prefer compact wording such as:
- `1. Wrong: ...`
- `2. Conflicting: ...`
- `3. Weak: ...`
- Mention `claim_verification_report.json` only if the user asked for it or a file was created.
## Safety
- Do not expose API keys.
- Do not fabricate citations.
- Do not mark a claim as verified without evidence.
- If unresolved, keep non-verified status with explicit notes.
Source safety policy:
- `verified` requires at least 2 independent sources.
- At least 1 source should be primary (official docs, regulator, company filing, peer-reviewed source).
- Do not use UGC-only evidence as sole basis for `verified`.
- The current batch helper should be treated as preliminary triage, not final publication proof.
## Reproducibility
- Normalize and deduplicate extracted claims before search.
- Preserve original claim order from draft text.
- Sort output `items` by first appearance in input text.
- Round confidence to 2 decimal places.
## Minimal Example
Input:
- `draft_text`: "OpenAI released GPT-5 in March 2026 and API input price is $3 per 1M tokens."
Default chat output:
- `This draft needs fixes before publish.`
- `1. Wrong: GPT-5 in March 2026 is incorrect; current evidence supports GPT-5.4 in March 2026 and GPT-5.5 in April 2026.`
- `2. Wrong: $3 per 1M input tokens is incorrect for GPT-5.5; current price is higher.`
Optional artifact output:
- `claim_verification_report.json` with two claim items, each containing status, confidence, and evidence links.
通过公开搜索URL在16个搜索引擎(7个国内直连+9个国际)中做跨引擎检索、结果去重与多源交叉验证,无需API Key,内置零依赖Python脚本。Use when 用户要求“搜一下/查资料/多找几个来源/核实消息/找公众号文章/排查报错/找论文或开源项目”时。输入自然语言查询词,输出带来源链接与可信度标注的结构化结果。不适用于私有数据检索、批量抓取、需要登录后可见的内容。
IMA OpenAPI を通じてノートとナレッジベースを操作します。
長いテキストや文書、ウェブページを要点とキーワードに圧縮する要約ツールです。
通过“查一下、搜一下、搜索、找一下”等自然语言检索公开网页,整理来源并交付横版 PDF、Markdown 与标题链接表。支持用户主动开启、查看和取消技能收藏,保留原报告取回。
ウェブ要求に適切なツールを割り当てるための必読ガイドです。
AI HOT (aihot.virxact.com) 中文 AI 资讯查询 Skill。当用户想知道"今天 AI 圈有什么"、"AI 日报"、"AI HOT"、"AI 资讯"、"AI 热点"、"最近 AI"、"OpenAI/Anthropic/Google 最近发布了什么"、"AI hot today"、"AI...
全球12亿文献知识库(8千万中文期刊可下载)——通过 SmartLib 开放平台 API 提供中外文学术文献检索与下载能力,覆盖 8000 万篇授权中文期刊全文 + 12.28 亿条全球文献元数据(期刊 7.19 亿 / 专利 2.15 亿 / 会议 7155 万 / 学位论文 2473 万 / 标准 268 万)。 能力:中英文关键词检索、文献详情、中文期刊 PDF 全文下载、外文 OA 文献十级渠道免费下载(不消耗配额)、智能关键词扩展、核心期刊优先排序、相关性重排、引文追溯、分类号检索。 配额:首次使用自动注册,免费 100 次检索 + 10 次下载 / 月;耗尽自动弹出套餐(体验卡 / 个人版月 / 专业版月 / 单篇下载 / 下载包),可说「升级 / 充值」唤起;企业 / 机构定制联系 [email protected]。 触发:用户表达"查论文""找文献""检索学术""搜期刊""查专利""找标准""下论文""写文献综述""找参考文献""查 SCI/EI"等意图时启用;也适用"帮我找关于 XX 的论文""写文献综述""找几篇引用支撑论点"。英文:"find papers" "search literature" "write literature review" "find supporting citations"。 调用前必须先用 /consume 获取 consume_token,再凭 token 调 /search(每次计费接口调用都需一次 /consume)。
uvx 経由で markitdown を呼び、さまざまな形式のファイルを Markdown に変換します。
データ月:2026-10 · ダウンロード数とスター数は公開スキルレジストリ(ClawHub、SkillHub)から月単位で集計しています。詳しくはデータの方法論をご覧ください。